diff --git a/.braid/snapshot.jsonl b/.braid/snapshot.jsonl
index 0924d56f9f..6370dffbb2 100644
--- a/.braid/snapshot.jsonl
+++ b/.braid/snapshot.jsonl
@@ -17,7 +17,7 @@
 {"id":"bd-0gsj","title":"Fix CRLF handling in tree-sitter-qmd grammar.js for pipe tables","description":"Tree-sitter grammar at crates/tree-sitter-qmd/tree-sitter-markdown/grammar.js parses pipe tables incorrectly when input uses CRLF line endings. With LF input, the pipe table closes at the blank line and following content becomes a separate block. With CRLF input, the table absorbs the following paragraph as a malformed body row. Bug propagates through pampa end-to-end: rendered Pandoc AST is structurally wrong for any Windows user with core.autocrlf=true (default) authoring a pipe table followed by a paragraph. Affects all output formats (HTML, PDF, Word) since Pandoc AST is upstream of every writer. Surfaced 2026-04-24 during PR 109 Windows verification as 5 failing pipe-table corpus tests; investigation in bd-ntnx classified it as a real grammar bug, not a corpus formatting issue. Full reproducer and observed parse trees in the design field. Cross-reference q2-windows-verify-tests memory note for related Windows CRLF failures.","design":"Reproducer. Two layers, both diagnostic. Layer 1 tests grammar.js directly. Layer 2 tests the full pampa pipeline that real users hit. Fixtures live in the tree-sitter-markdown directory so tree-sitter parse picks up the configured grammar without an init-config step.\n\nLayer 1 (grammar.js):\n\nprintf \"before\\r\\n\\r\\n| a | b |\\r\\n|---|---|\\r\\n| 1 | 2 |\\r\\n\\r\\nafter\\r\\n\" > crates/tree-sitter-qmd/tree-sitter-markdown/test-crlf.qmd\nprintf \"before\\n\\n| a | b |\\n|---|---|\\n| 1 | 2 |\\n\\nafter\\n\" > crates/tree-sitter-qmd/tree-sitter-markdown/test-lf.qmd\ncd crates/tree-sitter-qmd/tree-sitter-markdown\ntree-sitter parse test-crlf.qmd > test-crlf.parse\ntree-sitter parse test-lf.qmd > test-lf.parse\ndiff test-crlf.parse test-lf.parse\n\nLayer 2 (pampa end-to-end):\n\ncargo run --quiet -p pampa --bin pampa -- crates/tree-sitter-qmd/tree-sitter-markdown/test-crlf.qmd > pampa-crlf.out\ncargo run --quiet -p pampa --bin pampa -- crates/tree-sitter-qmd/tree-sitter-markdown/test-lf.qmd > pampa-lf.out\ndiff pampa-crlf.out pampa-lf.out\n\nObserved parse trees from Layer 1 on this branch (fix/windows-tree-sitter-crlf at 80f07171). Both trees agree on the prefix, so the divergence is the trailing block:\n\nLF (correct, table closes at the blank line):\n  pipe_table [2, 0] - [5, 0]\n    pipe_table_header [2, 0] - [2, 9] ...\n    pipe_table_delimiter_row [3, 0] - [3, 9] ...\n    pipe_table_row [4, 0] - [4, 9] ...\n  pandoc_paragraph [6, 0] - [7, 0]\n    pandoc_str [6, 0] - [6, 5]\n\nCRLF (buggy, table absorbs following paragraph):\n  pipe_table [2, 0] - [7, 0]\n    pipe_table_header [2, 0] - [2, 9] ...\n    pipe_table_delimiter_row [3, 0] - [3, 9] ...\n    pipe_table_row [4, 0] - [4, 9] ...\n    pipe_table_row [6, 0] - [6, 5]   <- SPURIOUS, should be a separate paragraph\n      pipe_table_cell [6, 0] - [6, 5]\n        pandoc_str [6, 0] - [6, 5]\n\nObserved Pandoc AST from Layer 2 (pampa default -t native), confirming the bug propagates past pampa input handling:\n\nLF: [Para [Str \"before\"], Table ... [body row [1,2]], Para [Str \"after\"]]\nCRLF: [Para [Str \"before\"], Table ... [body rows [1,2] AND [Plain \"after\"]]]   <- no trailing Para\n\nConfirms pampa does NOT normalize line endings before feeding the grammar. End-user impact: any qmd document with CRLF endings, a pipe table, and following content produces structurally wrong output in every writer that consumes Pandoc AST.\n\nOpen design questions for the fix.\n\n1. Locate the bug in grammar.js. Likely candidates: newline character classes that match only \\n, or the external scanner (scanner.cc) handling of block boundary detection. Read both grammar.js and any external scanner before patching. Tree-sitter-markdown forks vary on this — confirm what our fork does versus what works.\n\n2. Upstream vs local fix. This grammar is forked from MDeiml/tree-sitter-markdown (or whichever upstream we track — confirm in tree-sitter.json or README). If upstream already fixes this, cherry-pick. If not, patch locally and consider an upstream PR.\n\n3. Test-corpus line-ending policy after the fix. Two viable strategies:\n   a. Keep test/corpus/*.txt files materialized as CRLF on Windows checkout and rely on the fixed grammar to parse them identically. Tests then exercise CRLF on Windows and LF on Linux, which is useful coverage but creates a small risk that future grammar changes regress CRLF handling silently when only Linux CI runs.\n   b. Pin test/corpus/*.txt to LF via .gitattributes (eol=lf for that subdirectory only) and add a separate Windows-only regression test that re-creates the fixtures with CRLF and asserts the parse trees match. More explicit, less coverage drift.\n\n   The choice depends on whether we expect frequent grammar changes and whether ubuntu-only CI is sustainable. Recommend deciding before landing the grammar fix.\n\n4. Audit other CRLF patterns. The 5 failing pipe-table tests are the surface; there may be other constructs whose grammar paths assume \\n only. Grep grammar.js (and scanner.cc if present) for \\n usages and audit each one.\n\nReference. Investigation history and original framing in bd-ntnx (closed). Memory note q2-windows-verify-tests catalogues related Windows CRLF failures.","acceptance_criteria":"tree-sitter parse on matched CRLF and LF qmd inputs produces identical S-expression trees for pipe-table fixtures (Layer 1 reproducer diff is empty). pampa default native output is identical for the same matched inputs (Layer 2 reproducer diff is empty). All 5 originally-failing pipe-table tests under cd crates/tree-sitter-qmd/tree-sitter-markdown && tree-sitter test pass on Windows: Example 201 (GFM), Pipe table can precede content (two variants), Pipe table can follow content, Pipe table with blank-line caption. cargo xtask verify step 4 (tree-sitter test) passes on Windows. cargo nextest run --workspace and cargo xtask verify continue to pass on Linux CI (no regression). Decision recorded in commit message or PR description for upstream-vs-local fix and for test-corpus line-ending policy. Audit findings recorded for any other CRLF patterns discovered in grammar.js or scanner.cc. q2-windows-verify-tests memory note updated with PR link and resolution.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-04-27T14:26:53.718601600Z","created_by":"cderv","updated_at":"2026-05-04T13:45:00.482193500Z","closed_at":"2026-05-04T13:45:00.481818100Z","close_reason":"PR #139 merged 2026-04-30: fixed CRLF pipe-table line ending in scanner.c (typo fix at line 2259), added Rust regression test and xtask CRLF parity check","external_ref":"https://github.com/quarto-dev/q2/issues/138","labels":["tree-sitter","windows"],"dependencies":{"bd-ntnx:discovered-from":{"depends_on_id":"bd-ntnx","type":"discovered-from","created_at":"2026-04-27T14:26:53.718601600Z","created_by":"cderv"}},"comments":{"c-2vcbuybu":{"id":"c-2vcbuybu","author":"cderv","created_at":"2026-04-28T16:05:51Z","text":"Tomorrow's verification (next session pickup): (1) check PR #139 CI status — gh pr checks 139 --repo quarto-dev/q2; (2) if green, merge; (3) on merge, close bd-0gsj with PR link as the close reason; (4) flush and commit accumulated .beads/issues.jsonl from main repo (this session created bd-vxl8 and bd-picv, plus comments on bd-238o, bd-3pe8, bd-0gsj); (5) update q2-windows-verify-tests memory note status to fully resolved."},"c-swxf3f4l":{"id":"c-swxf3f4l","author":"cderv","created_at":"2026-04-29T16:25:17Z","text":"CRLF parity check landed in commit 8f61f587 on branch fix/windows-tree-sitter-crlf (PR #139). cargo xtask verify step 4 now re-runs tree-sitter test against a CRLF-converted corpus copy; 451/451 pass, regression-tested by reverting scanner.c (446/451, build fails). Awaiting PR merge to close."},"c-xjxgh7ir":{"id":"c-xjxgh7ir","author":"cderv","created_at":"2026-04-28T15:37:36Z","text":"PR #139 opened (https://github.com/quarto-dev/q2/pull/139). Layer 1 + Layer 2 reproducer diffs empty post-fix. tree-sitter test on Windows: 451/451 (was 446/451). Decision: local fix, not upstream — typo introduced on this fork in 79489c4e. Decision: no .gitattributes change for test corpus — fixed grammar parses LF and CRLF identically, so existing corpus tests both modes (LF on Linux, CRLF on Windows checkout). Audit: line 2259 was the only outlier; all other newline checks in scanner.c use paired '\\n' || '\\r'. Added Rust regression test pipe_table_crlf_matches_lf for cross-platform coverage."}}}
 {"id":"bd-0gwekaem","title":"nested_cell_mask doesn't cover inline expressions, so a displayed `{r} x` still executes","description":"The nested-cell mask (claude-notes/plans/2026-09-02-nested-cell-masking.md, \"Make q2 self-documenting again: don't execute a cell that is being displayed\") suppresses execution of an executable *fence opener* inside a markdown-displaying CodeBlock. It does not cover *inline* expressions, so an inline brace expression inside a display block still runs.\n\nWHY IT DIDN'T APPLY BEFORE\nThe mask's Scope section lists what is deliberately out (RawBlock, doubled-brace, {=html} openers, cells nested in cells). Inline expressions are not on that list because they were not reachable: q2 never evaluated the `{r} expr` spelling at all. bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe made that spelling live, which is what opens this hole. The fix for that strand is correct and matches Quarto 1; this is the follow-on the mask needs to stay whole.\n\nMEASURED, q2 (with the inline fix) vs Quarto 1 99.9.9. Fixture: knitr document, `v <- \\\"EVALUATED\\\"`, then each form. EVALUATED means the escape failed.\n\n  form                              q2              Q1\n  plain prose                       executes        executes   (the feature)\n  `` `{r} v` `` extra backticks     executes        executes\n  `{{r}} v` doubled brace           shows {{r}} v   shows {r} v\n  inside a ````markdown block       executes        executes\n  inside a ````r block              executes        executes\n\nTwo consequences:\n1. The escape q2 documents for cells -- wrap it in a longer fence -- does not work for inline expressions, so guides/authoring/computations.qmd has to document a different rule for the inline case than for the cell case.\n2. The only working escape is the doubled brace, and q2 displays it verbatim ({{r}}) where Q1 collapses it to {r}. So a q2 author currently has no way to show a reader the text `{r} x`. Note q2's literal rendering is defensible on its own terms -- collapsing doubled braces is what Q-2-50 deliberately dropped so Jinja/Mustache templates render correctly -- which is why the fence-wrapper route, not brace-collapsing, is the right fix.\n\nSHAPE OF THE FIX\nSmall and contained. `mask_code_block_text` (crates/quarto-core/src/engine/nested_cell_mask.rs) is the single rewrite site; add an inline pattern alongside MASK_OPENER_RE that rewrites an inline `{lang}` marker inside an in-scope CodeBlock the same way the opener is rewritten. `unmask` needs no change: UNMASK_OPENER_RE is deliberately not line-anchored (H4, for blockquote prefixes), so it already restores a mid-line occurrence.\n\nMind the plan's invariants: byte-exact restore is load-bearing for reconcile (spec section 'Reconcile -- why byte-exactness is load-bearing'), and the test-seam spec is frozen, so this wants its own tier rather than being bolted onto an existing one.\n\nAlso worth deciding: whether the empty-classes / `markdown`-only display-class predicate is right for inline too, or whether the inline case should be widened (an author writing ````r to show inline syntax gets nothing today -- the E row above).\n\nDocs: guides/authoring/computations.qmd's 'Writing about an inline expression' section documents today's behaviour and should be simplified to match the cell rule once this lands.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-09-04T16:33:12.125548Z","created_by":"Gordon Woodhull","updated_at":"2026-09-04T18:12:17.724568Z","labels":["engine"],"dependencies":{"bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe:discovered-from":{"depends_on_id":"bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe","type":"discovered-from","created_at":"2026-09-04T16:33:12.125548Z","created_by":"Gordon Woodhull"}},"comments":{"c-5lgjuro9":{"id":"c-5lgjuro9","author":"Gordon Woodhull","created_at":"2026-09-04T18:12:17.724568Z","text":"PR #657 — https://github.com/quarto-dev/q2/pull/657 (remote branch bugfix/bd-0gwekaem-nested-cell-mask-inline-expressions, one commit 8e13ecf8b). Pushed after a full cargo xtask verify --skip-rust-tests --skip-hub-tests (WASM/hub-client leg included, all 14 steps green); workspace nextest 13692 passed / 199 skipped and workspace clippy -D warnings were already clean on the same commit."},"c-ndusewmr":{"id":"c-ndusewmr","author":"Gordon Woodhull","created_at":"2026-09-04T16:40:36.539600Z","text":"Dispatched to agent inline-executable-wrapped on workspace-2 (branch braid/bd-0gwekaem-nested-cell-mask-inline-expressions).\n\nConfirmed reproducible on main today with the classic spelling -- an inline `r v` inside a ````markdown display block renders EVALUATED on q2 0.29.0. The brace-spelling half stays latent until bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe merges; the agent was told to cover both spellings and to branch from main rather than from that unmerged branch, which is pending a history squash."},"c-wvs2dbvu":{"id":"c-wvs2dbvu","author":"Gordon Woodhull","created_at":"2026-09-04T17:32:06.840651Z","text":"Fixed on braid/bd-0gwekaem-nested-cell-mask-inline-expressions @ 8e13ecf8b (off main @ 914f20697). Not pushed.\n\nMASKED SPELLING. One rewrite for both inline spellings: the marker is inserted directly after the opening backtick, so ````r v```` becomes ````q2-nested-executable r v```` and ````{r} v```` becomes ````q2-nested-executable {r} v````. unmask deletes the same fixed string. This departs from the strand's suggestion that UNMASK_OPENER_RE could restore the brace form for free: it can, but the classic spelling has no brace to work with and would need its own rule regardless, and two masked spellings buy nothing over one. Insert-only/delete-only also gives the strongest byte-exactness argument available, which the plan says is load-bearing for reconcile.\n\nMASK_INLINE_RE = (^|[^`])`((?:\\{[A-Za-z0-9_]+\\}|r)[ \\t#][^`]+)`\n\n- Brace form: any [A-Za-z0-9_]+, engine-agnostic, matching MASK_OPENER_RE's language class and Q1's per-language executeInlineCodeHandler. Doubled braces excluded by the same first-char rule as openers.\n- Classic form: literal r only. It is knitr's own rmarkdown syntax and is hardcoded to r in all_patterns$md$inline.code; generalizing it to any word would rewrite every ordinary code span in displayed prose.\n- Separator [ \\t#] is the union of knitr's [ #] and q2's [ \\t] -- the mask has to be a superset of every scanner that could claim the expression.\n- The (^|[^`]) guard is the bd-knitr-inline-r-eats-fence-2ofk91x1 shape: without it a nested fence's third backtick anchors a match. It excludes only a backtick, not a backslash -- knitr does not exclude one either, and the YAML-fallback case that motivates INLINE_R_PATTERN's backslash guard is text this mask never sees. Accepted limitation, shared with INLINE_R_PATTERN and Q1: two expressions with no character between them leave the second unmasked.\n\nDISPLAY-CLASS PREDICATE: left narrow (empty or [\"markdown\"]), and filed as bd-tiidc899. Three reasons. It is not inline-specific -- the same predicate gates both rewrites, so widening changes the cell escape too, which 2026-09-02-nested-cell-masking.md scoped out explicitly. Widening for inline only would give the incoherent rule '````r protects your inline expressions but not your nested fences'. And T18 is frozen and revert-bound to the narrow predicate. Narrow is also parity: the strand's own matrix has Q1 executing in a ````r block too.\n\nTESTS: new tier, own hunk ids, T1-T24 untouched. H14 mask rewrites an inline marker, H15 unmask restores it, H16 the guard. T25/T26 render-tier (real knitr); T27 x9 round-trip byte-exactness (classic, brace, tab, hash, two-on-a-line, bare fence, blockquoted, multiline body, opener+inline together); T28 a live cell's backtick-quoted R name untouched; T29 the guard; T30 engine-agnostic brace + doubled brace out of scope; T31 binds to the production scanner -- resolve_inline_r_expressions leaves the masked text alone while the control unmasked text is rewritten, which is the inline counterpart of T14 and starts discriminating the brace half automatically when the {r} alternation lands.\n\nVACUITY FOUND AND FIXED: the first draft of T25 used INLINE-RAN as its runtime-only marker and passed with H14 reverted. q2's classic spelling escapes markdown specials, so the executed value reaches the HTML as INLINE\\-RAN and the contains() check was blind to it. Marker changed to INLINERAN; both render rows confirmed RED under a reverted H14 before the fix was restored. A runtime-only marker used with an inline expression must survive .QuartoInlineRender's escaping unchanged -- recorded in the test file.\n\nVERIFICATION: workspace 13692 passed / 199 skipped, +16 over the 13676 baseline, no removals. cargo clippy --workspace --all-targets -D warnings clean. End-to-end with a freshly built q2 on the strand's repro: the ````markdown row now renders 'Inline: `r v`' where main renders 'Inline: EVALUATED'; the ````r row still renders EVALUATED (bd-tiidc899).\n\nDOCS: no change here. The 'Writing about an inline expression' section the strand names does not exist on main -- it is added by bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe's second commit. Simplifying it belongs with that merge; docs/guides/authoring/computations.qmd on main only documents the cell rule, which stays accurate."}}}
 {"id":"bd-0h8crv6u","title":"Duplicate-seq self-heal for file documents (backoff + circuit breaker)","description":"Generalize the index-doc duplicate-seq self-heal (bd-6f21d4c6, ts-packages/quarto-sync-client/src/client.ts installDuplicateSeqRecovery) to file documents: a legacy stable-actor client colliding on a FILE doc currently wedges that file's sync permanently for the receiving browser (only remedy: manual IndexedDB wipe). Add per-document recovery with backoff/attempt cap and the state.unavailableFiles retry poll as circuit breaker. Plan: claude-notes/plans/2026-10-22-file-doc-duplicate-seq-recovery.md","status":"closed","priority":1,"issue_type":"feature","created_at":"2026-10-22T10:23:21.422509Z","created_by":"shikokuchuo","updated_at":"2026-10-22T12:40:50.024324Z","closed_at":"2026-10-22T12:40:50.024324Z","close_reason":"Implemented: recoverFileDocument with per-document backoff (3 attempts / 30s cooldown / 300s quiet window) + circuit breaker into the unavailable-files retry poll; routing generalized by documentId in installDuplicateSeqRecovery; change-handler tracking keyed by doc id; recoveryBackoff injectable via ConnectOptions. Tests: new full-stack-file-collision.test.ts (red-then-green gated) + file-collision-recovery.test.ts (backoff, circuit breaker, index + untracked-doc regressions). 191 sync-client + 462 hub-client tests green; build:all green. Plan: claude-notes/plans/2026-10-22-file-doc-duplicate-seq-recovery.md","dependencies":{"bd-6f21d4c6:related":{"depends_on_id":"bd-6f21d4c6","type":"related","created_at":"2026-10-22T10:23:21.422509Z","created_by":"shikokuchuo"}}}
-{"id":"bd-0i6fii0k","title":"product-doc-theme (posit-docs) on q2: exploratory assessment + fix plan","description":"Exercise posit-dev/product-doc-theme (the posit-docs custom project type) on a q2 website, compare against Quarto 1 (prima facie correct), and record what works, what breaks, and what fundamentally cannot work as-is. Covers q2 add (not implemented), contributes.project path handling, theme/include rebasing, env shortcodes, and the example site's markdown. Deliverable: claude-notes plan with assessment + fix plan split between q2 and product-doc-theme.","status":"in_progress","priority":1,"issue_type":"task","created_at":"2026-10-21T19:39:12.691821Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-26T16:27:40.583704Z","labels":["extensions"],"dependencies":{"bd-ad7i1pc6:related":{"depends_on_id":"bd-ad7i1pc6","type":"related","created_at":"2026-10-21T19:39:12.691821Z","created_by":"Carlos Scheidegger"}},"comments":{"c-38tbbbtl":{"id":"c-38tbbbtl","author":"Claude Fable 5.1","created_at":"2026-10-22T15:05:23.080050Z","text":"Filed per-finding strands (2026-10-22): F1 bd-74ktvzv7 (logo doubling, P1 bug), F3 bd-t3n1nghb (link-external-*), F4 bd-rfz4e3bw (q2 add), F6 bd-7u2xu4gp ({mermaid} fence warning), F7 bd-hfghrqvh (table caption classes), F8 bd-3z5bubrf (quarto-required), F9 bd-f5l0n6oa (Q-2-9 in config, P4), F10 bd-2pmjmjmj (diagram fig-align). F5 goes to the extension (raw {=html} block). F2 experiment: replacing the footer <img> tags with markdown images (![Posit](/_extensions/posit-docs/assets/images/x.svg){.light-content width=\"65px\\}) works on q2 — all four footer images are copied to _site/_extensions/posit-docs/assets/images/ and load (0 broken footer images in headless Chromium; the only broken image left is the F1 logo), and Q-2-9 drops from 24 to 2 (the <small> in website.title). On Quarto 1 the left item (text+href) renders the markdown and relocates the files to site_libs/quarto-contrib/; but the right item (a markdown link wrapping two images with {#open_preferences_center title=...} attributes and no href) is emitted verbatim, unrendered, so the cookie-preferences link needs a different Q1-compatible spelling or stays raw HTML."},"c-et30yzmn":{"id":"c-et30yzmn","author":"Carlos Scheidegger","created_at":"2026-10-26T16:27:40.583704Z","text":"Status refresh 2026-10-26: rebased branch onto main @ 4d966ea05. Re-ran the example on nightly 0.36.0-nightly.20261024: F1/F4/F7/F10 verified fixed; F3 merged today (#906), not in the nightly yet. Filed F2(2) as bd-5r97wehd (raw <img> in footer not copied, no Q-5-6). Closed bd-t3n1nghb, bd-hfghrqvh, bd-2pmjmjmj (PRs merged). Example still 5/6 pages (F5 grid.qmd, extension-side). Plan updated with a Status 2026-10-26 section."},"c-oper168q":{"id":"c-oper168q","author":"Carlos Scheidegger","created_at":"2026-10-21T19:44:22.496560Z","text":"Assessment + plan written: claude-notes/plans/2026-10-21-product-doc-theme-q2.md (commit 9f3c57e44 on braid/bd-0i6fii0k-product-doc-theme-q2). Key findings: custom project types already work (bd-ad7i1pc6) — type: posit-docs resolves, themes/includes/env shortcodes all land; 5/6 example pages render. Gaps: F1 navbar logo src doubled (resolve_metadata_path re-resolves the rebased extension path against _extension.yml's dir; P1 q2 bug), F2 /_extensions/ asset URLs in footer raw <img> not copied/relocated (Q1 relocates to site_libs/quarto-contrib/quarto-project), F3 link-external-icon/newwindow unimplemented, F4 q2 add is a stub, F5 inline <style> block in grid.qmd is a hard parse error (example fix: {=html} fence), F6 {mermaid} fences fall through silently (bd-vg8p5yoj), F7 table {.striped .hover} caption attrs dropped, F8 quarto-required ignored, F9 Q-2-9 noise ×24 for footer HTML. Nothing fundamentally blocked. Awaiting review of the 5 open questions before filing per-finding strands."}}}
+{"id":"bd-0i6fii0k","title":"product-doc-theme (posit-docs) on q2: exploratory assessment + fix plan","description":"Exercise posit-dev/product-doc-theme (the posit-docs custom project type) on a q2 website, compare against Quarto 1 (prima facie correct), and record what works, what breaks, and what fundamentally cannot work as-is. Covers q2 add (not implemented), contributes.project path handling, theme/include rebasing, env shortcodes, and the example site's markdown. Deliverable: claude-notes plan with assessment + fix plan split between q2 and product-doc-theme.","status":"in_progress","priority":1,"issue_type":"task","created_at":"2026-10-21T19:39:12.691821Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T17:28:56.537182Z","labels":["extensions"],"dependencies":{"bd-ad7i1pc6:related":{"depends_on_id":"bd-ad7i1pc6","type":"related","created_at":"2026-10-21T19:39:12.691821Z","created_by":"Carlos Scheidegger"}},"comments":{"c-38tbbbtl":{"id":"c-38tbbbtl","author":"Claude Fable 5.1","created_at":"2026-10-22T15:05:23.080050Z","text":"Filed per-finding strands (2026-10-22): F1 bd-74ktvzv7 (logo doubling, P1 bug), F3 bd-t3n1nghb (link-external-*), F4 bd-rfz4e3bw (q2 add), F6 bd-7u2xu4gp ({mermaid} fence warning), F7 bd-hfghrqvh (table caption classes), F8 bd-3z5bubrf (quarto-required), F9 bd-f5l0n6oa (Q-2-9 in config, P4), F10 bd-2pmjmjmj (diagram fig-align). F5 goes to the extension (raw {=html} block). F2 experiment: replacing the footer <img> tags with markdown images (![Posit](/_extensions/posit-docs/assets/images/x.svg){.light-content width=\"65px\\}) works on q2 — all four footer images are copied to _site/_extensions/posit-docs/assets/images/ and load (0 broken footer images in headless Chromium; the only broken image left is the F1 logo), and Q-2-9 drops from 24 to 2 (the <small> in website.title). On Quarto 1 the left item (text+href) renders the markdown and relocates the files to site_libs/quarto-contrib/; but the right item (a markdown link wrapping two images with {#open_preferences_center title=...} attributes and no href) is emitted verbatim, unrendered, so the cookie-preferences link needs a different Q1-compatible spelling or stays raw HTML."},"c-et30yzmn":{"id":"c-et30yzmn","author":"Carlos Scheidegger","created_at":"2026-10-26T16:27:40.583704Z","text":"Status refresh 2026-10-26: rebased branch onto main @ 4d966ea05. Re-ran the example on nightly 0.36.0-nightly.20261024: F1/F4/F7/F10 verified fixed; F3 merged today (#906), not in the nightly yet. Filed F2(2) as bd-5r97wehd (raw <img> in footer not copied, no Q-5-6). Closed bd-t3n1nghb, bd-hfghrqvh, bd-2pmjmjmj (PRs merged). Example still 5/6 pages (F5 grid.qmd, extension-side). Plan updated with a Status 2026-10-26 section."},"c-j577kwo9":{"id":"c-j577kwo9","author":"Carlos Scheidegger","created_at":"2026-10-27T17:28:56.537182Z","text":"2026-10-27: F2(2) / bd-5r97wehd closed as working-as-designed (q2 does not look inside raw HTML, per fc5pvkcv). Resolved on the content side: posit-dev/product-doc-theme now has a q2 branch (q2 add posit-dev/product-doc-theme@q2, commit 6e4d99c) with markdown footer images, the grid.qmd <style> block in a {=html} fence (F5), and the website.title <small> marked as raw HTML (F9 noise gone for the example). Example renders 6/6 with 0 errors/warnings on q2 main. Details: claude-notes/plans/2026-10-26-footer-raw-html-img.md on main."},"c-oper168q":{"id":"c-oper168q","author":"Carlos Scheidegger","created_at":"2026-10-21T19:44:22.496560Z","text":"Assessment + plan written: claude-notes/plans/2026-10-21-product-doc-theme-q2.md (commit 9f3c57e44 on braid/bd-0i6fii0k-product-doc-theme-q2). Key findings: custom project types already work (bd-ad7i1pc6) — type: posit-docs resolves, themes/includes/env shortcodes all land; 5/6 example pages render. Gaps: F1 navbar logo src doubled (resolve_metadata_path re-resolves the rebased extension path against _extension.yml's dir; P1 q2 bug), F2 /_extensions/ asset URLs in footer raw <img> not copied/relocated (Q1 relocates to site_libs/quarto-contrib/quarto-project), F3 link-external-icon/newwindow unimplemented, F4 q2 add is a stub, F5 inline <style> block in grid.qmd is a hard parse error (example fix: {=html} fence), F6 {mermaid} fences fall through silently (bd-vg8p5yoj), F7 table {.striped .hover} caption attrs dropped, F8 quarto-required ignored, F9 Q-2-9 noise ×24 for footer HTML. Nothing fundamentally blocked. Awaiting review of the 5 open questions before filing per-finding strands."}}}
 {"id":"bd-0idqzj33","title":"pampa: give the remaining uncoded parse-error sources Q-codes (backticks in code spans, '$ ' at line start, braces in prose, indented lines)","description":"Split out of bd-bare-at-literal-w3ytmu8e to keep that strand scoped to @. Its original scope also asked to 'check the other uncoded parse-error sources in the same corpus and give each a Q-code or fix it'. Per the 2026-09-24 comment on that strand, grouping the claude-notes uncoded errors by the character at the error position found: code spans containing backticks (127), '$ ' at line start in prose (32), braces in prose (~40), indented lines (~18). Many end-of-line uncoded errors are cascades after an earlier coded error (Q-2-7 / Q-2-17). For each source, decide whether to fix the parse or add a Q-code, error-corpus cases and a docs/errors page. Measure with the claude-notes scan method in claude-notes/plans/2026-09-25-star-as-str.md.","status":"open","priority":2,"issue_type":"task","created_at":"2026-09-25T21:41:38.102787Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-25T21:41:38.102787Z","labels":["parser"],"dependencies":{"bd-bare-at-literal-w3ytmu8e:discovered-from":{"depends_on_id":"bd-bare-at-literal-w3ytmu8e","type":"discovered-from","created_at":"2026-09-25T21:41:38.102787Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-0jo6fnb7","title":"Upgrade copy_navbar_logo / copy_favicon missing-file warnings to spanned Q-5-6","description":"Phase 4 of bd-page-footer-image-items-stmpikgo gave copy_footer_images the uniform missing-resource shape: a ResourceCopyIntent + missing_resource_diagnostic (Q-5-6) anchored at the reference in _quarto.yml. Its siblings in website_post_render.rs — copy_navbar_logo and copy_favicon — still emit uncoded, span-less DiagnosticMessage::warning strings ('website.navbar.logo refers to missing file ...'). Give them the same treatment: build an intent with origin = the config scalar's SourceInfo (navbar.logo / website.favicon values carry one) and push missing_resource_diagnostic. Existing integration tests pipeline_navbar_logo_missing_diagnoses_continues and the favicon one assert on d.title containing the filename; update them to assert code Q-5-6 + location like assert_footer_q_5_6. Also consider the sidebar logo if/when it grows a copy hook.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-18T21:36:41.846236Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-18T21:36:41.846236Z","dependencies":{"bd-page-footer-image-items-stmpikgo:discovered-from":{"depends_on_id":"bd-page-footer-image-items-stmpikgo","type":"discovered-from","created_at":"2026-08-18T21:36:41.846236Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-0jtai5dh","title":"Five near-copies of pandoc stringify with subtly different semantics","description":"Rendering Inlines to plain text is implemented five times in the tree, and no two agree on the details. Any consumer picking one by name rather than by semantics gets a silent behaviour surprise.\n\n- pampa::lua::utils::stringify_inline/stringify_inlines - private + behind the lua-filter feature. SoftBreak/LineBreak -> \"\\\\n\". Backs pandoc.utils.stringify.\n- pampa::writers::plaintext::inlines_to_string - public. Keeps a Code span's backticks (locked by a test at plaintext.rs:483) and emits Math as raw TeX.\n- quarto_core::template::inlines_to_text - private. Breaks -> \" \", Code -> text without backticks.\n- quarto_core::transforms::listing_render::inlines_to_text - private, another copy.\n- quarto_core::transforms::reference_link_diagnostics::push_rendered - added by bd-q249-message-drops-inline-content-pacg3qeu, because none of the four above had the semantics that diagnostic needs (Code without backticks AND breaks as spaces, since the result goes inside a one-line warning).\n\nThe differences are real, not accidental: backtick handling and break handling each have a defensible answer per consumer. So this is not a straight de-duplication - it wants one canonical walk with explicit options (break rendering, code rendering) and the call sites migrated deliberately, each with a test pinning the choice it depends on.\n\nNoticed while fixing Q-2-49's message. Out of scope there: that fix needed one correct string, not a refactor of the other four call sites.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-26T19:40:33.282472Z","created_by":"Gordon Woodhull","updated_at":"2026-08-26T19:40:33.282472Z","dependencies":{"bd-q249-message-drops-inline-content-pacg3qeu:discovered-from":{"depends_on_id":"bd-q249-message-drops-inline-content-pacg3qeu","type":"discovered-from","created_at":"2026-08-26T19:40:33.282472Z","created_by":"Gordon Woodhull"}}}
@@ -43,7 +43,7 @@
 {"id":"bd-0vfgz2cl","title":"Per-word source spans for classes promoted from class=\"...\" kv attributes","description":"Follow-up from bd-heading-id-attr-duplicated-xbpcmejr (design question 4; user accepted the simple version for now). When kv class=\"c1 c2\" is promoted into the classes slot, each resulting class gets an AttrSourceInfo classes entry pointing at the WHOLE quoted value span, not at its own word. Precise per-word sub-spans need offset math inside the quoted value (quote/escape handling) — likely wants a helper in or alongside the quarto-source-map crate for slicing a SourceInfo by byte range. Low priority: only affects diagnostic precision for multi-class kv attributes.","status":"open","priority":4,"issue_type":"task","created_at":"2026-08-19T14:24:26.413893Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-19T14:24:26.413893Z","labels":["quarto-source-map"],"dependencies":{"bd-heading-id-attr-duplicated-xbpcmejr:discovered-from":{"depends_on_id":"bd-heading-id-attr-duplicated-xbpcmejr","type":"discovered-from","created_at":"2026-08-19T14:24:26.413893Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-0vje43lp","title":"qmd parser: ':Caption' with no space after the colon is not a table caption after a blank line (Pandoc accepts it)","description":"Pandoc treats ':NoSpace {.striped}' as a table caption both directly under the table and after a blank line. q2 (tree-sitter-qmd) only recognises the directly-under form (via the caption-as-last-row fixup in crates/pampa/src/pandoc/treesitter_utils/postprocess.rs; attribute lifting added in bd-hfghrqvh). After a blank line the line becomes a Para ':NoSpace' and the attribute block is dropped. ': Spaced' works in both positions. Likely a grammar change in the caption rule (requires whitespace after ':'). Repro: printf '| a |\\n|---|\\n| 1 |\\n\\n:NoSpace {.striped}\\n' | pampa -t native  vs  pandoc -f markdown -t native.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-10-22T20:46:28.895825Z","created_by":"t","updated_at":"2026-10-22T20:46:28.895825Z","labels":["parity"],"dependencies":{"bd-hfghrqvh:discovered-from":{"depends_on_id":"bd-hfghrqvh","type":"discovered-from","created_at":"2026-10-22T20:46:28.895825Z","created_by":"t"}}}
 {"id":"bd-0wyo","title":"Server-precomputed other_metadata_html for default listing","description":"Q1's item-default.ejs.md iterates over fields *not* in the curated set ('title','image','image-alt','date','author','subtitle','description','reading-time','categories') and emits a <div class=\"metadata-value listing-{field}\">…</div> per such field. This is dynamic in EJS — there is no equivalent in doctemplate without server-side precomputation.\n\nL3 v1 ships the default listing with curated-fields-only output. This issue picks up the gap by adding an other_metadata_html string to the per-item TemplateValue::Map binding, computed server-side in the listing render transform. The built-in item-default.template gets one more interpolation site: $item.other-metadata-html$.\n\nSource-code TODO marker lands in crates/quarto-core/src/project/listing/templates/item-default.template at the otherFields gap location (added during L3).\n\nDiscovered while writing L3 (bd-ml8z) sub-plan; see D15 in claude-notes/plans/2026-05-06-listings-L3-resolve-transform.md.","status":"open","priority":3,"issue_type":"task","created_at":"2026-05-06T18:26:31.323613Z","created_by":"cscheid","updated_at":"2026-05-06T18:26:31.323613Z","dependencies":{"bd-ml8z:discovered-from":{"depends_on_id":"bd-ml8z","type":"discovered-from","created_at":"2026-05-06T18:26:31.323613Z","created_by":"cscheid"}}}
-{"id":"bd-0x1znhgb","title":"Defer metadata path resolution: carry the authoring base dir instead of re-deriving it from the source span","description":"resolve_metadata_path (crates/quarto-core/src/transforms/navigation_href.rs) derives the base directory for a relative metadata path from the source span's file. That conflates 'where the bytes were authored' with 'what the path is relative to'. bd-74ktvzv7 patched the immediate case by returning extension-manifest-sourced values unchanged (project-root-relative), but the general problem remains: a rebased value keeps a span in _extension.yml; an extension could ship a _metadata.yml in a subdirectory with its own relative paths; directory _metadata.yml files generally need their own base. Reconsider the approach: record the resolution base (project root / document dir / declaring-file dir / extension dir) when the value is merged or rebased, and resolve once at the consumer against that base, instead of re-deriving it from SourceInfo at render time. Also covers the ConfigValueKind::Path doc-relativization that double-applied for rebased website.* values (bd-74ktvzv7 phase 2).","status":"open","priority":2,"issue_type":"task","created_at":"2026-10-22T15:52:38.273706Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-22T15:52:38.273706Z","dependencies":{"bd-74ktvzv7:discovered-from":{"depends_on_id":"bd-74ktvzv7","type":"discovered-from","created_at":"2026-10-22T15:52:38.273706Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-0x1znhgb","title":"Defer metadata path resolution: carry the authoring base dir instead of re-deriving it from the source span","description":"resolve_metadata_path (crates/quarto-core/src/transforms/navigation_href.rs) derives the base directory for a relative metadata path from the source span's file. That conflates 'where the bytes were authored' with 'what the path is relative to'. bd-74ktvzv7 patched the immediate case by returning extension-manifest-sourced values unchanged (project-root-relative), but the general problem remains: a rebased value keeps a span in _extension.yml; an extension could ship a _metadata.yml in a subdirectory with its own relative paths; directory _metadata.yml files generally need their own base. Reconsider the approach: record the resolution base (project root / document dir / declaring-file dir / extension dir) when the value is merged or rebased, and resolve once at the consumer against that base, instead of re-deriving it from SourceInfo at render time. Also covers the ConfigValueKind::Path doc-relativization that double-applied for rebased website.* values (bd-74ktvzv7 phase 2).","status":"closed","priority":2,"issue_type":"task","created_at":"2026-10-22T15:52:38.273706Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T22:51:30.736415Z","closed_at":"2026-10-28T22:51:30.736415Z","close_reason":"Merged in PR #959 (adcc7efb6): extension contributes.project fragments rekeyed to a virtual <manifest>#project file so config_paths::file_base_dir yields the project root; rebased website.* assets use site-rooted /_extensions/... spelling; _extension.yml filename check removed. Per-value base deferred (recorded in plan Phase 3 handoff).","dependencies":{"bd-74ktvzv7:discovered-from":{"depends_on_id":"bd-74ktvzv7","type":"discovered-from","created_at":"2026-10-22T15:52:38.273706Z","created_by":"Carlos Scheidegger"}},"comments":{"c-lzbup1j1":{"id":"c-lzbup1j1","author":"Carlos Scheidegger","created_at":"2026-10-28T20:41:54.334437Z","text":"PR #959 (stacked on #958). Site-rooted /_extensions spelling for rebased website.* strings; the project fragment is a virtual file (config_paths::project_fragment_file_id + file_base_dir) instead of the _extension.yml file-name rule. Per-value Generated wrapper rejected: published error-reporting renders start_offset()=0 for Generated (snippets lost). See plan §Phase 3 handoff."}}}
 {"id":"bd-0xghpvij","title":"Lua parity C1: OrderedList discards its ListAttributes argument","description":"constructors.rs binds _list_attr and hardcodes (1, Default, Default). Differential case: orderedlist-listattributes. Catalog: claude-notes/research/2026-07-13-lua-api-mismatch-catalog.md. Epic plan: claude-notes/plans/2026-07-13-lua-api-pandoc-parity.md.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-13T17:09:26.361234Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-13T17:13:03.014521Z","closed_at":"2026-07-13T17:13:03.014521Z","close_reason":"Fixed: OrderedList now parses its ListAttributes argument via parse_list_attributes. Differential case orderedlist-listattributes matches oracle; xfail shrunk 6->5. Commit on braid/bd-grkrb9nj-lua-api-pandoc-parity.","dependencies":{"bd-grkrb9nj:parent-child":{"depends_on_id":"bd-grkrb9nj","type":"parent-child","created_at":"2026-07-13T17:09:26.361234Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-0xmt","title":"Phase A.0: Lift WASM JS bridge to @quarto/wasm-js-bridge workspace package","description":"Move hub-client/src/wasm-js-bridge/ to a new @quarto/wasm-js-bridge workspace package. Wire hub-client + q2-preview-spa + preview-renderer test configs to alias /src/wasm-js-bridge -> the package's src. Removes the cross-platform symlink/duplication risk for Phase A.3. See claude-notes/plans/2026-05-13-q2-preview-phase-a.md §A.0.","status":"closed","priority":1,"issue_type":"task","created_at":"2026-05-13T16:53:11.353454Z","created_by":"cscheid","updated_at":"2026-05-13T17:07:44.781502Z","closed_at":"2026-05-13T17:07:44.781355Z","close_reason":"Bridge files moved to @quarto/wasm-js-bridge workspace package; consumers alias /src/wasm-js-bridge. Verified end-to-end (sass compile through bridge proven by hub-client test:wasm; cargo xtask verify 11/11). Commits ea1bb889 + changelog 043f65e1.","dependencies":{"bd-kw93:parent-child":{"depends_on_id":"bd-kw93","type":"parent-child","created_at":"2026-05-13T16:53:11.353454Z","created_by":"cscheid"}}}
 {"id":"bd-0yp90370","title":"Case-sensitivity of index-file stem matching is inconsistent across sidebar code","description":"During bd-sidebar-dir-index-md-5khf3lds the user decided that section_for_dir's directory-index stem match must be case-SENSITIVE (stem == \"index\" exactly), because case-insensitive matching misbehaves across case-sensitive filesystems vs case-preserving/insensitive ones (macOS). But is_top_level_index in the same file (crates/quarto-core/src/transforms/sidebar_auto.rs, ~line 300) still matches the stem with eq_ignore_ascii_case, so 'Index.qmd' at the project root is excluded from auto: true sibling lists while 'guides/Index.md' is treated as an ordinary child. Neither behavior is obviously wrong in isolation, but the codebase has no stated policy for path case handling, and macOS's case-preserving-insensitive default means a file can be openable under a name that differs in case from what discovery recorded. Low priority per the user ('case-sensitive errors are not particularly high-priority'), but worth a pass to (a) state a policy, (b) align is_top_level_index and any similar stem/extension matches with it, and (c) make sure nothing egregious (silent drop, wrong link) happens when only case differs.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-19T16:41:46.293109Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-19T16:41:46.293109Z","labels":["navigation"],"dependencies":{"bd-sidebar-dir-index-md-5khf3lds:discovered-from":{"depends_on_id":"bd-sidebar-dir-index-md-5khf3lds","type":"discovered-from","created_at":"2026-08-19T16:41:46.293109Z","created_by":"Carlos Scheidegger"}}}
@@ -255,7 +255,7 @@
 {"id":"bd-3zst4hwy","title":"Clean up clippy debt and gate clippy in CI","description":"CI never runs clippy; ~69 violations have accrued across 11 crates (quarto-hub 42, xtask 11, quarto-system-runtime 7, others) once the existing [workspace.lints.clippy] allow-policy is respected. Dominant lints: collapsible_if (32), map_unwrap_or (13), needless_borrows_for_generic_args (5) — mostly clippy --fix auto-fixable. A few hand-fixes / #[allow]: should_implement_trait, naive_bytecount, get_unwrap. Plan: fix all 69, then add a CI step + xtask leg running 'cargo clippy --workspace --all-targets -- -D warnings' so it stays clean (gate-and-grow, but small enough to finish in ~1-2 sessions). Architectural lints (result_large_err, large_enum_variant, too_many_arguments, type_complexity, ptr_arg) are already allowed in the workspace table and stay allowed. Discovered while adding q2 mcp --print-config (bd-9a8yu2gw).","status":"closed","priority":2,"issue_type":"chore","created_at":"2026-06-13T15:45:10.467531Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-13T16:38:17.617590Z","closed_at":"2026-06-13T16:38:17.617590Z","close_reason":"Workspace clippy clean (cargo clippy --workspace --all-targets -- -D warnings exits 0); gate added to CI (test-suite.yml) + cargo xtask verify Step 1. ~525 violations fixed (auto + hand). 10036 tests pass. Branch beads/clippy-cleanup-gate, commits c9445494/25c1e187/d1cb1dec(+plan). Not pushed; hub-client test leg fails on pre-existing missing-WASM env, orthogonal.","dependencies":{"bd-9a8yu2gw:discovered-from":{"depends_on_id":"bd-9a8yu2gw","type":"discovered-from","created_at":"2026-06-13T15:45:10.467531Z","created_by":"Carlos Scheidegger"}},"comments":{"c-g0rfqzgg":{"id":"c-g0rfqzgg","author":"Carlos Scheidegger","created_at":"2026-06-13T16:09:44.958893Z","text":"SCOPE CORRECTION + progress checkpoint. The initial '69 violations' was a measurement artifact: clippy -D warnings aborts a crate on its first lint, masking the rest, so quarto-core (~240) was hidden until upstream crates were fixed. True scope is 250+ across the workspace, quarto-core-dominated. Progress: auto-fixed the bulk via cargo clippy --fix (now ~150-file diff, +/-); hand-fixed/allowed the judgment lints. Down to 28 remaining (mostly MaybeIncorrect single_match/filter_map_next + test-data approx_constant + deliberate from_str should_implement_trait). NOTE: had to use --broken-code to apply quarto-core fixes (normal cargo fix reverts a whole crate if any fix breaks compile); this exposed a clippy bug that expanded a matches!() macro into its raw template (compile_theme_css.rs) + cascaded an import removal — caught and hand-fixed; workspace cargo check --all-targets is green. NOT YET DONE: full nextest run to validate the auto-fixes, finish last 28, add the CI gate. Remaining list saved to claude-notes/2026-06-13-clippy-remaining.txt. Uncommitted on branch beads/clippy-cleanup-gate."}}}
 {"id":"bd-3zuggmsr","title":"q2 preview: active page outside _quarto.yml render: list shows \"Project render produced no output\"","description":"In project preview, navigating to (or launching on) a page that is NOT in the project's `render:` list (e.g. a scratch .qmd at the project root) shows the Render Error banner \"Project render produced no output for the active page\", even though the page is valid.\n\nROOT CAUSE (confirmed): RenderMode::ActivePage filters project.files, which discovery builds from the render: globs. A page outside those globs is absent from project.files, so pass_two produces zero outputs; the WASM entry point (wasm-quarto-hub-client/src/lib.rs:1633) then hits the catch-all because the active page is not among pass1_failures either. It is NOT sibling-capture-failure poisoning and NOT a poisoned engine host (fail_fast defaults false; proven with a clean no-broken-siblings native repro at the orchestrator level).\n\nQ2 has no single-file escape hatch for an in-project file: crates/quarto/src/commands/preview.rs resolve_project_and_initial_page forces project mode for any file under a _quarto.yml ancestor (single_file: None, pinned by test at preview.rs:575); --no-project is mutually exclusive with a positional path. So without a fix, a render:-excluded in-project page is unreachable in preview.\n\nQ1 (TypeScript quarto-cli) comparison, source-read only: it is mode-dependent. Direct 'quarto preview foo.qmd' renders a render:-excluded file (cmd.ts single-file render() fallback bypasses project.files.input). But navigating to a non-member inside a running project preview REFUSES via previewUnableToRenderResponse() (serve.ts:877) / 404 -- the direct analog of Q2's message. So Q2's current behavior matches Q1's navigation mode; the real defect is (a) the confusing message and (b) no way to preview the page at all.\n\nDESIGN DECISION (open -- no fix landed):\n- Option 1: orchestrator injects the ActivePage target into project.files so it renders on-demand in project mode. Prototyped and verified end-to-end (native RED->GREEN test + q2 preview + headless Chromium: sum_sines2.qmd rendered with 5 marimo islands). BUT the render:-excluded page renders with DEGRADED nav chrome: sidebar container present but EMPTY (no links, missing 'nav-sidebar docked' body classes) because the page is not a member of any website.sidebar.contents entry. Looks half-wired.\n- Option 2: route the render:-excluded active page through the single-doc renderer (standalone, no project chrome) -- closer to Q1's direct-invocation single-file semantics; sidesteps the empty-sidebar artifact; but does not give project theme/nav and is a larger CLI+WASM change.\n\nSTATUS: Option 1 prototype was REVERTED at Gordon's request pending the design decision above. Nothing committed. See conversation 2026-07-24.","status":"open","priority":1,"issue_type":"bug","created_at":"2026-07-24T16:57:57.818860Z","created_by":"Gordon Woodhull","updated_at":"2026-07-24T22:04:58.757367Z","labels":["bug","preview"],"comments":{"c-0b5awo7m":{"id":"c-0b5awo7m","author":"Gordon Woodhull","created_at":"2026-07-24T22:04:58.757367Z","text":"2026-07-24: Prototyped Option 1 (orchestrator injects ActivePage target into project.files). Verified end-to-end but found it renders render:-excluded pages with an empty sidebar rail (page is not a member of any configured sidebar). Reverted the prototype (both orchestrator.rs and the regression test) pending decision between Option 1 (project render, degraded nav) and Option 2 (single-file render for excluded pages, Q1-parity). No code committed."}}}
 {"id":"bd-4062e8tn","title":"preview hub aborts: automerge fork_at MissingOps panic in periodic sync (upstream #1327)","description":"q2 preview --allow-edit --share aborts after minutes: sync_document's fork_at(checkpoint) hits automerge 0.10.0's fork_at dependency-traversal bug (upstream automerge#1327): a dep hash can be scheduled twice in some DAG shapes, duplicate BuildChangeMetadata makes the change collector return MissingOps and the unwrap at collector.rs:761 panics; the unwind escapes with_document so samod's WithDocGuard::drop panics ('dropped without comitting') and the process aborts. Root cause VERIFIED: issue's deterministic repro panics identically on our 0.10.0; PR #1366's fix (mark deps seen at schedule time) applied to a local 0.10.0 copy makes the repro pass with all 553 quarto-hub+quarto-preview tests green. Fix tracks: (A) git-pin automerge to a fork carrying #1366 on 0.10.0 (precedent: runtimelib pin); (B) containment in sync_document (catch_unwind + save/load heal) so collector-class panics can't abort the process; recommended: both. Diagnosis: claude-notes/plans/2026-08-07-automerge-fork-at-missingops.md","status":"closed","priority":0,"issue_type":"bug","created_at":"2026-08-07T15:45:23.994955Z","created_by":"shikokuchuo","updated_at":"2026-08-20T09:27:33.652557Z","closed_at":"2026-08-20T09:27:33.652557Z","close_reason":"Fixed: automerge fork_at MissingOps panic resolved (upstream automerge#1327 / fix PR #1366)."}
-{"id":"bd-40luf359","title":"xtask lint: config-path-base — ban ad-hoc base-dir joins on config-originated path strings","description":"New lint rule in crates/xtask/src/lint/ (precedents: metadata-as-str, add-file-with-id). Flag joining a config-originated string onto a base directory (doc_dir.join / Path::new(...) / CWD reads on values from as_plain_text()/as_str()) outside a blessed-module list (the blessed resolvers + metadata_merge; see 'Blessed mechanisms' in claude-notes/designs/path-resolution-model.md). Suppress with lint:allow(config-path-base) + reason. Sequencing: most effective AFTER the convergence work (unified path-shaped-key registry generalizing the css layer_base marking, bd-oejuizi9/bd-hjv5o) shrinks the blessed list; a first cut restricted to stage/ and transforms/ consumers is viable earlier. Optional stronger form once consumers migrate: make raw-string access to ConfigValueKind::Path values a distinctively named accessor and lint that name instead. Design: claude-notes/research/2026-08-19-path-resolution-class-assessment.md §6.3.","status":"open","priority":2,"issue_type":"task","created_at":"2026-08-19T15:03:28.493703Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-19T15:03:28.493703Z","dependencies":{"bd-oejuizi9:related":{"depends_on_id":"bd-oejuizi9","type":"related","created_at":"2026-08-19T15:03:28.493703Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-40luf359","title":"xtask lint: config-path-base — ban ad-hoc base-dir joins on config-originated path strings","description":"New lint rule in crates/xtask/src/lint/ (precedents: metadata-as-str, add-file-with-id). Flag joining a config-originated string onto a base directory (doc_dir.join / Path::new(...) / CWD reads on values from as_plain_text()/as_str()) outside a blessed-module list (the blessed resolvers + metadata_merge; see 'Blessed mechanisms' in claude-notes/designs/path-resolution-model.md). Suppress with lint:allow(config-path-base) + reason. Sequencing: most effective AFTER the convergence work (unified path-shaped-key registry generalizing the css layer_base marking, bd-oejuizi9/bd-hjv5o) shrinks the blessed list; a first cut restricted to stage/ and transforms/ consumers is viable earlier. Optional stronger form once consumers migrate: make raw-string access to ConfigValueKind::Path values a distinctively named accessor and lint that name instead. Design: claude-notes/research/2026-08-19-path-resolution-class-assessment.md §6.3.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-08-19T15:03:28.493703Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T23:15:49.086697Z","closed_at":"2026-10-28T23:15:49.086697Z","close_reason":"Merged in PR #960: xtask lint config-path-base bans consumer-chosen base dirs for metadata path strings in quarto-core stage/ and transforms/; fixed book_cover_image.rs; four annotated allows; AGENTS.md and path-resolution-model.md updated.","dependencies":{"bd-oejuizi9:related":{"depends_on_id":"bd-oejuizi9","type":"related","created_at":"2026-08-19T15:03:28.493703Z","created_by":"Carlos Scheidegger"}},"comments":{"c-mekkr7ub":{"id":"c-mekkr7ub","author":"Carlos Scheidegger","created_at":"2026-10-28T20:59:11.134468Z","text":"Implemented on branch braid/bd-40luf359-xtask-lint-config-path (remote feature/bd-40luf359-config-path-base-lint), stacked on PR #959. Rule config-path-base in crates/xtask/src/lint/config_path_base.rs; scope stage/ + transforms/; taint = metadata reads, string params beside a Path param, closure params over tainted receivers; resolve* sanitizes. Found and fixed book_cover_image.rs; 4 allows with reasons. Plan §Phase 2 handoff."}}}
 {"id":"bd-41dros0u","title":"node_tests.rs unused PathBuf import fails clippy on Windows (-D warnings)","description":"std::path::PathBuf is imported unconditionally at crates/quarto-mcp-launcher/tests/integration/node_tests.rs:11, but its only consumer is fake_node at line 22, which is #[cfg(unix)]-gated. On Windows this makes the import unused, failing cargo clippy --all-targets -D warnings. CI (Linux/Mac) is unaffected. Fix: move the use inside the #[cfg(unix)] block or gate it with #[cfg(unix)]. Discovered during bd-i9i5ad2t Phase 5 verify while fixing the sibling bd-nj9nnkn1 dead-code issue.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-02T11:54:46.698369Z","created_by":"cderv","updated_at":"2026-07-02T11:56:14.007894Z","closed_at":"2026-07-02T11:56:14.007894Z","close_reason":"Moved the PathBuf import into the #[cfg(unix)] fake_node fn (its only consumer); verified with cargo clippy -p quarto-mcp-launcher --all-targets -- -D warnings","labels":["windows"],"dependencies":{"bd-i9i5ad2t:discovered-from":{"depends_on_id":"bd-i9i5ad2t","type":"discovered-from","created_at":"2026-07-02T11:54:46.698369Z","created_by":"cderv"}}}
 {"id":"bd-42202ipf","title":"Preview capture-splice mis-pairs cells when a hidden cell is immediately followed by another cell","description":"Found while implementing bd-nn2fou8h (execute-visibility). PRE-EXISTING, not introduced by that work — but bd-nn2fou8h makes it reachable in more documents.\n\nThe A1<->B1 walk in crates/quarto-core/src/engine/capture_splice.rs (derive_cell_outputs) pairs each engine cell in A1 with 'the next cell wrapper Div in B1'. Its doc comment states the invariant 'the engine always emits exactly one wrapper Div per executed cell'. That invariant is FALSE for a fully hidden cell, which emits nothing at all:\n- knitr has always done this: a chunk with #| include: false produces no wrapper. Verified on main before any bd-nn2fou8h change: a two-cell doc (first #| include: false) renders with exactly one class=\"cell\" div.\n- bd-nn2fou8h makes jupyter do the same, and extends it to document-scope execute: include:/echo:+output: combinations.\n\nWHEN IT IS FINE: prose between the cells re-anchors the walk. Traced: A1=[prose, cell(hidden), prose, cell(visible), prose] against B1=[prose, prose, Div.cell, prose] pairs correctly, because the lookahead hits a non-wrapper, takes the conservative break branch, and leaves j in place for the following prose match.\n\nWHEN IT BREAKS: two ADJACENT cells with no prose between, the first fully hidden. A1=[cell(hidden), cell(visible)], B1=[Div.cell(visible)]. The hidden cell's key maps to the VISIBLE cell's wrapper, then the visible cell finds nothing and falls through to raw source. Net effect in q2 preview: the hidden cell displays the next cell's output, and the next cell shows its raw source. Render (non-preview) is unaffected — the splice is preview-only.\n\nFIX DIRECTIONS (not chosen):\n1. Make the pairing verify the wrapper belongs to the cell before consuming it (e.g. compare the wrapper's .cell-code content to the A1 cell's code). Has false negatives: an echo:false wrapper carries no code to compare.\n2. Carry explicit cell identity through the capture rather than inferring it positionally — the robust fix, but it changes the capture wire shape ({input_qmd, result.markdown}).\n3. Give the engines a marker block for hidden cells so B1 keeps one block per cell. Rejected while implementing bd-nn2fou8h: an empty ::: {.cell} paints a real <div class=\"cell\"> box in the output, and knitr's suppression happens inside R where q2 cannot inject a marker anyway.\n\nRepro sketch: a two-cell jupyter doc, first cell '#| include: false', no prose between, opened in q2 preview.\n\nAnalysis: claude-notes/plans/2026-08-14-issue-523-execute-visibility.md","status":"open","priority":2,"issue_type":"bug","created_at":"2026-08-14T15:50:04.352796Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-14T16:24:32.511344Z","labels":["deferred-website-push","q2-preview"],"dependencies":{"bd-7hqea3qi:related":{"depends_on_id":"bd-7hqea3qi","type":"related","created_at":"2026-08-14T16:24:30.879110Z","created_by":"Carlos Scheidegger"},"bd-nn2fou8h:discovered-from":{"depends_on_id":"bd-nn2fou8h","type":"discovered-from","created_at":"2026-08-14T15:50:04.352796Z","created_by":"Carlos Scheidegger"},"bd-nn2fou8h:related":{"depends_on_id":"bd-nn2fou8h","type":"related","created_at":"2026-08-14T16:24:32.511335Z","created_by":"Carlos Scheidegger"}},"comments":{"c-2wn1ccmb":{"id":"c-2wn1ccmb","author":"Carlos Scheidegger","created_at":"2026-08-14T16:01:58.454689Z","text":"EMPIRICALLY CONFIRMED in a live q2 preview session (2026-08-14), replacing the code-trace reasoning in the description. Both halves of the claim reproduce exactly as predicted.\n\nREPRO (jupyter). Single-file 'q2 preview' on:\n\n  Prose before.\n\n  ```{python}\n  #| include: false\n  hidden = 1\n  ```\n  ```{python}\n  print(\"SECOND_CELL_OUTPUT\")\n  ```\n\n  Prose after.\n\nRendered iframe DOM (read via CDP):\n  div.cell count: 1\n  pre classes: ['sourceCode python :: print(\"SECOND_CELL_OUTPUT\")',\n                'code-with-copy :: SECOND_CELL_OUTPUT',\n                '{python} code-with-copy :: print(\"SECOND_CELL_OUTPUT\")']\n  visible text: Prose before. / print(\"SECOND_CELL_OUTPUT\") / SECOND_CELL_OUTPUT / print(\"SECOND_CELL_OUTPUT\") / Prose after.\n\nSo: the HIDDEN cell's position renders the SECOND cell's spliced output, and the second cell's own position falls through to raw source (note the literal '{python}' class — the un-executed pre-engine form). The second cell's content therefore appears TWICE, and the cell the author asked to hide is the one showing output.\n\nPRE-EXISTENCE PROVEN, not merely argued. The identical fixture with engine: knitr and an R chunk (`#| include: false`, then `cat(\"SECOND_CELL_OUTPUT\\n\")`) produces the same DOM: 1 div.cell, ['sourceCode r :: cat(...)', 'code-with-copy :: SECOND_CELL_OUTPUT', '{r} code-with-copy :: cat(...)']. knitr's include:false path is untouched by bd-nn2fou8h — it has always emitted no wrapper — so this is a splice defect that predates the visibility work. bd-nn2fou8h only widens the set of documents that can reach it (jupyter cells, and document-scope execute: options).\n\nMETHOD NOTE for whoever picks this up: editing the .qmd while 'q2 preview' is running did NOT trigger a re-capture in single-file mode — the page kept rendering both cells as raw source (the documented no-capture fall-through), which masks the bug. Restarting 'q2 preview' against the final content is what makes the initial capture match the document and exposes the mis-pairing. Budget for that or you will misread a stale-capture render as a fix.\n\nThe prose-separated case renders correctly, as traced in the description — the intervening prose re-anchors the two-pointer walk."}}}
 {"id":"bd-42owi5yb","title":"Selection-sync: pure helpers + parent applySelection (TDD scaffold)","description":"Pure, unit-testable core: sourceOffsetToLineCol(byteLineMap, offset)->{line,col}; parent-side applySelectionMessage(msg)->monaco.Selection with echo-guard (isSyncingRef window + editorHasFocusRef). No wiring yet. jsdom/vitest tests are the deliverable. Phase 0 of bd-6dib1198.","status":"open","priority":2,"issue_type":"task","created_at":"2026-06-25T21:27:50.121566Z","created_by":"shikokuchuo","updated_at":"2026-06-25T21:27:50.121566Z","dependencies":{"bd-6dib1198:parent-child":{"depends_on_id":"bd-6dib1198","type":"parent-child","created_at":"2026-06-25T21:27:50.121566Z","created_by":"shikokuchuo"}}}
@@ -293,6 +293,7 @@
 {"id":"bd-4ltdl","title":"Author error-docs pages for template subsystem (7 codes)","description":"Author stub-quality pages for all 7 template subsystem error codes under docs/errors/template/. Follow template in docs/errors/yaml/Q-1-10.qmd. See claude-notes/plans/2026-05-22-error-docs-content.md.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-05-24T21:07:57.022926Z","created_by":"cscheid","updated_at":"2026-05-24T21:15:22.653391Z","closed_at":"2026-05-24T21:15:22.653051Z"}
 {"id":"bd-4ly7ne01","title":"Desugar bare caption-id pipe-table into canonical float Div (tbl- crossref)","description":"Bare pipe-table with caption-attribute id (': cap {#tbl-data}') never reaches the float path: it produces a bare Block::Table that FloatRefTargetSugarTransform doesn't classify (only Div/Figure/Div>Table). Fails identically in HTML and revealjs (verified 2026-06-18). Fix (matching 'all syntaxes desugar into divs') = a desugar step wrapping a caption-id Table into Div(#tbl-..) > Table, then the existing uniform classifier handles it. Q1 ref: parsefiguredivs.lua. Parent audit: bd-w0c6d38k.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-06-18T16:59:12.994039Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-18T17:53:33.221449Z","closed_at":"2026-06-18T17:53:33.221449Z","close_reason":"Implemented: maybe_wrap_bare_table_into_div in float_ref_target.rs wraps bare caption-id tables into the canonical Div(#tbl-)>Table so the uniform classifier handles them. fixture_bare_table_caption_target (RED->GREEN); verified e2e in HTML+revealjs (Table 1/Table 2 resolve). 2386 quarto-core tests pass. Commit db3fc6b5.","labels":["crossref"],"dependencies":{"bd-w0c6d38k:discovered-from":{"depends_on_id":"bd-w0c6d38k","type":"discovered-from","created_at":"2026-06-18T16:59:12.994039Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-4m2n6qf1","title":"Table floats render their caption twice (inner <table><caption> + synthesized <figcaption>)","description":"Found by end-to-end render while preparing the bd-hcp8m3ve PR; the 10511-test workspace suite does not catch it.\n\nA crossref table float emits the caption text TWICE:\n\n<div id=\"tbl-one\" class=\"quarto-float quarto-figure quarto-figure-center\">\n  <figure class=\"quarto-float quarto-float-tbl\">\n    <div aria-describedby=\"tbl-one-caption\">\n      <table class=\"caption-top table\">\n        <caption>A table caption</caption>   <-- still here\n        ...\n      </table>\n    </div>\n    <figcaption id=\"tbl-one-caption\" ...>A table caption</figcaption>  <-- and here\n  </figure>\n</div>\n\nConfirmed that <table><caption> is the normal non-float table rendering (a bare captioned table with no #tbl- id renders the same inner caption), so the float transform synthesizes the figcaption from the table's caption without clearing the table's own. Q1's shape 1 carries the caption only in the figcaption — design doc says table floats 'replace the current Div+caption-paragraph shape with shape 1 (figure.quarto-float-tbl + <figcaption>)'.\n\nImpact: duplicated caption text visually and for screen readers (aria-describedby points at the figcaption while the table also self-captions).\n\nSecondary observation for the same fix: the synthesized figcaption carries quarto-float-caption-bottom while the inner table carries caption-top — worth confirming which is right for tbl floats against Q1.\n\nRepro: render a doc containing ::: {#tbl-one} <pipe table> : A table caption ::: through cargo run --bin q2 -- render.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-27T20:04:57.037293Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-27T21:56:34.675179Z","closed_at":"2026-07-27T21:56:34.675179Z","close_reason":"MERGED: fixed in 32f06010, shipped with PR #423 (4310c173 on main). Table floats no longer emit their caption twice — the Table's own caption is elided at figcaption-synthesis time in render_float_ref_target, matching Q1's parse-time elision, so both the pampa writer and the React preview renderer get it. TDD (table_float_clears_the_tables_own_caption watched red first); verified e2e on both repro forms.","dependencies":{"bd-hcp8m3ve:discovered-from":{"depends_on_id":"bd-hcp8m3ve","type":"discovered-from","created_at":"2026-07-27T20:04:57.037293Z","created_by":"Carlos Scheidegger"}},"comments":{"c-8egiwxw2":{"id":"c-8egiwxw2","author":"Carlos Scheidegger","created_at":"2026-07-27T20:25:36.840892Z","text":"FIX IMPLEMENTED (TDD) on the bd-hcp8m3ve branch, commit b1d72b25, folded into PR #423 per the where-to-fix decision.\n\nRED first: new unit test table_float_clears_the_tables_own_caption in crates/quarto-core/src/transforms/crossref_render.rs, watched failing with 'the table's own caption must be cleared once hoisted into the figcaption, else it renders twice: Some([Paragraph(...Cap...)])'.\n\nRoot cause of the test-coverage gap, worth recording: the existing table_target_renders_q1_float_shape supplies the caption as a SIBLING PARAGRAPH (content: vec![table, para(\"Table caption\")]), never as the Table's own caption.long — so the duplicating path had no coverage at all. That is why 10511 passing tests missed it and only an e2e render caught it.\n\nFIX: in render_float_ref_target, inside the html_float_dom && is_float_kind branch, clear caption.long/short on top-level Block::Table entries in the float content. Guarded by !is_uncaptioned — when the float is uncaptioned nothing was hoisted, so the Table's caption is the only copy of that text and must be preserved.\n\nChose the transform rather than the writers deliberately: the caption is gone from the AST before either writer runs, so the pampa HTML writer AND the React preview renderer (Figure.tsx) both get the fix, with no parity drift between render and preview.\n\nVERIFIED end-to-end through the real binary on BOTH repro forms (': Cap {#tbl-one}' and '::: {#tbl-one}'): 'Cap' now appears exactly once each, inside <figcaption id=\"tbl-one-caption\">; the inner <table><caption> is gone. Workspace 10512/10512 (one more than before = the new test), ZERO snapshot churn, only crossref_render.rs touched. Full cargo xtask verify running before push (quarto-core change -> WASM leg).\n\nStill open and NOT addressed here: the figcaption carries quarto-float-caption-bottom while the inner table carries caption-top. Left alone deliberately — it is a caption-location question, not a duplication one."},"c-jqv57yno":{"id":"c-jqv57yno","author":"Carlos Scheidegger","created_at":"2026-07-27T20:15:03.178651Z","text":"CHECKED AGAINST MAIN (2026-07-27) — the duplication is PRE-EXISTING, not introduced by PR #423.\n\nBuilt origin/main's q2 (78d55deb, detached worktree, shared CARGO_TARGET_DIR) and rendered the identical minimal fixture:\n\nMAIN (78d55deb):\n<div id=\"tbl-one\">\n<table class=\"caption-top table\"><caption>Cap</caption>...</table>\nCap                                    <-- bare trailing text node\n</div>\n\nPR #423 branch:\n<div id=\"tbl-one\" class=\"quarto-float quarto-figure quarto-figure-center\">\n<figure class=\"quarto-float quarto-float-tbl\">\n<div aria-describedby=\"tbl-one-caption\"><table class=\"caption-top table\"><caption>Cap</caption>...</table></div>\n<figcaption id=\"tbl-one-caption\" ...>Cap</figcaption>\n</figure>\n</div>\n\nBoth emit the caption text twice. #423 changes the SHAPE of the second copy (stray trailing text node -> proper figcaption) but does not fix the double emission. So #423 is not a regression on this axis; it is arguably an improvement, since the duplicate is now at least semantic markup.\n\nQ1'S SOLUTION CONFIRMED (Carlos's recollection is right — elide the table's own caption when it becomes a float). Two sites in quarto-cli, both in quarto-pre/parsefiguredivs.lua, at float-PARSE time rather than render time:\n- line 280 (div-wrapped case): walks the div content for a Table with a non-empty caption, hoists it into the float caption, then 'table caption should be removed from the table as we'll handle it' -> table.caption = pandoc.Caption{}\n- line 544 (caption-attr case, startsWith(attr.identifier, 'tbl-')): 'we've parsed the caption, so we can remove it from the table' -> el.caption.long = pandoc.Blocks({})\n\nRECOMMENDATION ON WHERE TO FIX: on top of PR #423, not against main. The duplication predates #423, but #423 rewrites render_float_ref_target — the exact code a fix must touch — so a fix landed on main would conflict. Cleanest: fold it into #423 or ship it as the immediate follow-up PR after #423 merges.\n\nNote Q1 elides at parse time; Q2's float DOM is built in the Finalization-phase transform, so the natural Q2 site is wherever the figcaption is synthesized (clear the Table's own caption at the same moment)."},"c-jri7zqih":{"id":"c-jri7zqih","author":"Carlos Scheidegger","created_at":"2026-07-27T20:31:25.861098Z","text":"PUSHED: commit 32f06010 on feature/bd-hcp8m3ve-float-taxonomy (PR #423). Full cargo xtask verify GREEN — all 14 steps, true exit 0, 10512/10512 tests, zero snapshot churn. Both repro forms verified end-to-end: caption appears exactly once. Close when #423 merges."},"c-ophsw8e3":{"id":"c-ophsw8e3","author":"Carlos Scheidegger","created_at":"2026-07-27T20:08:19.986463Z","text":"MINIMAL REPRO (verified 2026-07-27). Five lines, no front matter, no cross-reference, no div wrapper:\n\n| a |\n|---|\n| 1 |\n\n: Cap {#tbl-one}\n\nRender with: cargo run --bin q2 -- render tbl.qmd\n\nOutput has 'Cap' twice — once in <table class=\"caption-top table\"><caption> inside the aria wrapper, once in <figcaption id=\"tbl-one-caption\" class=\"quarto-float-caption-bottom quarto-float-caption quarto-float-tbl\">.\n\nControls:\n- Remove the {#tbl-one} id -> single <caption>, no float wrapper, no figcaption. So it is the float path, not table rendering in general.\n- No @tbl-one reference is needed; being a float target is sufficient.\n- The ::: {#tbl-one} div-wrapped form behaves identically, so it is not specific to caption-attr syntax.\n\nNOT yet established: whether main (pre-bd-hcp8m3ve) also duplicates via its older Div+caption-paragraph shape, i.e. whether PR #423 introduces a user-visible regression or merely carries an existing one into the new DOM. Needs a build of main to settle."}}}
+{"id":"bd-4nypw3wq","title":"Docs: how to run {r} cells on a Jupyter kernel (e.g. Ark) — engine: {jupyter: {claims: {r: N}}} overrides knitr's Primary(1)","description":"By design (claude-notes/designs/engine-resolution.md §4.3) jupyter claims every language as Fallback(0) and knitr claims {r} as Primary(1), so {r} cells go to knitr even with 'engine: jupyter' in frontmatter. This is not discoverable: the render silently uses knitr (log shows 'processing file: example.rmarkdown') and the user's Jupyter R kernel (Ark, IRkernel) is never started.\n\nVerified workaround (q2 0.36.0-nightly.20261027): a per-entry claims table on the NAMED engine raises jupyter above knitr for r:\n\nengine:\n  jupyter:\n    claims:\n      r: 1\n\nNotes from testing:\n- The 'engines:' form ('engines: [{jupyter: {claims: {r: {kind: primary, priority: 1}}}}]') in document frontmatter did NOT route r to jupyter; 'engine:' (which names the engine) did. Per the design doc, engines: only configures engines and does not name them. Worth confirming the intended scope before documenting engines: as a doc-level option.\n- Masking knitr instead ('engines: [{knitr: {claims: []}}, jupyter]') also works but disables ALL knitr claims (r, plus reticulate python/sql/bash interop) for the doc; prefer raising jupyter's r claim.\n- Claims grammar: integer = Primary with that priority; or {kind: primary|interop|fallback, priority, whenClass}. Whole-table replacement per engine (does not merge with built-in claims).\n- Kernel selection is separate: today q2 maps r -> 'ir' and ignores jupyter.kernel (bd-9pvolzwl, bd-db68qtuf), so Ark must currently be registered under the kernelspec name 'ir'. Doc should say so until those land, or be updated when they do.\n\nDeliverable: a short docs page/section (under docs/guides/, engines or computations) with the frontmatter above, the knitr-vs-jupyter claim explanation, and the kernel-name caveat. Working example: ~/docs/ark/example.qmd.","status":"open","priority":3,"issue_type":"docs","created_at":"2026-10-28T17:38:59.960354Z","created_by":"Gordon Woodhull","updated_at":"2026-10-28T17:42:36.384836Z","labels":["jupyter"],"dependencies":{"bd-9pvolzwl:related":{"depends_on_id":"bd-9pvolzwl","type":"related","created_at":"2026-10-28T17:38:59.960354Z","created_by":"Gordon Woodhull"}},"comments":{"c-0r71fgdg":{"id":"c-0r71fgdg","author":"Gordon Woodhull","created_at":"2026-10-28T17:42:36.384836Z","text":"r: 1 vs r: 2 (supersedes the 'r: 2 is clearest' remark above; description and ~/docs/ark/example.qmd now use 1).\n\nr: 1 — pros: reads as 'same weight as knitr, but I named it', matching the intent; no magic number above knitr's built-in Primary(1). Verified working (q2 0.36.0-nightly.20261027). Cons: wins only via the tie-break (explicitly named engine precedes knitr in candidate order, engine-resolution.md 'candidate_engines'), so it depends on an ordering rule the doc must state and that a future resolver change could flip silently; also hard-codes knitr's current priority, so if knitr's r claim is ever raised to 2+, r: 1 silently reverts to knitr with no error.\n\nr: 2 — pros: wins on priority alone, independent of tie-break order; stays correct against any knitr Primary(1). Cons: also silently breaks if knitr's r priority is raised to 2+ (any fixed number does); the '+1 over knitr' value looks arbitrary and needs explaining.\n\nNet: both work today and share the knitr-priority-change fragility; 2 additionally removes the tie-break dependency, 1 reads more naturally. Whichever is documented, the page should say the value is relative to knitr's built-in Primary(1), and ideally a test pins the chosen behavior."},"c-10oie26x":{"id":"c-10oie26x","author":"Gordon Woodhull","created_at":"2026-10-28T17:39:52.707723Z","text":"Priority sweep (q2 0.36.0-nightly.20261027, ~/docs/ark/example.qmd, engine: {jupyter: {claims: {r: N}}}): r: 2 routes {r} to jupyter/Ark (knitr is Primary(1), so the minimum winning priority is 2). r: 1 ALSO routes to jupyter, because a tie is broken by candidate order and an explicitly named engine is listed before knitr. So r: 2 is the clearest value to document (wins outright; does not rely on the tie-break); 10 was just a safe high number. Not tested: r: true (priority None -> default) or r: 0."}}}
 {"id":"bd-4qflzhwh","title":"Wire TS-engine checkInstallation into q2's engine protocol + 'q2 check' (Q1 parity)","description":"From plan-4c 4cE (2026-07-03): checkInstallation is INERT in q2 — the ToEngine wire enum (crates/quarto-core/src/engine/ts_protocol.rs:33-101) has 11 variants and no checkInstallation; the only trace is the optional declaration at ts-packages/quarto-types/src/execution-engine.ts:209. Q1 invokes engines' checkInstallation via its install/check tooling. q2 ALREADY HAS 'q2 check' ('Verify correct functioning of Quarto installation') — the natural home: add a CheckInstallation wire variant + host dispatch, and have q2 check enumerate installed TS engines and invoke it (marimo's is a no-op delay+spinner stub today, so assert only runs-without-error; julia's may do real checks). Same family as the populateCommand gap (bd-m1jeqhhz): Q1 engine-contributed CLI surfaces not yet in q2's protocol. Documented: migration guide §4 (claude-notes/research/2026-07-03-marimo-migration-guide.md), compat doc §16.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-07-03T17:11:53.843746Z","created_by":"Gordon Woodhull","updated_at":"2026-09-28T09:53:24.297254Z","dependencies":{"bd-m1jeqhhz:related":{"depends_on_id":"bd-m1jeqhhz","type":"related","created_at":"2026-07-03T17:11:54.976729Z","created_by":"Gordon Woodhull"}},"comments":{"c-4d1x4pgd":{"id":"c-4d1x4pgd","author":"Gordon Woodhull","created_at":"2026-07-05T05:01:32.657777Z","text":"Plan 10 authored: claude-notes/plans/2026-07-04-plan10-check-installation.md (branch braid/bd-4qflzhwh-wire-ts-engine-checkinstallation, rebased on feature/ts-engine-extensions incl. Plan 9 merge). Frozen Test Seam Spec W1-W4/T1-T2/H1-H4/R1/E1/C1-C8/K1-K3/P1-P2/J1-J3 + accepted-untested log. Ready for execution."},"c-5k6u8jwq":{"id":"c-5k6u8jwq","author":"Gordon Woodhull","created_at":"2026-07-03T17:44:35.468727Z","text":"Research + design complete: claude-notes/research/2026-07-03-plan10-check-installation-research.md (on braid/bd-4qflzhwh-wire-ts-engine-checkinstallation). Recommendation presented to Gordon; awaiting approval before authoring Plan 10."},"c-x5azj1vv":{"id":"c-x5azj1vv","author":"cderv","created_at":"2026-09-28T09:53:24.297254Z","text":"2026-09-28: q2 check on main is still a bare NotImplemented stub (crates/quarto/src/commands/check.rs), and no braid/bd-4qflzhwh or feature/ts-engine-extensions branch exists on the remote -- the 'ready for execution' note in bd-elhgm6ff appears stale. bd-elhgm6ff (q2 check jupyter) is sequenced after this strand and needs to know: should it wait, execute Plan 10 itself, or build the minimal dispatch per your frozen spec? See claude-notes/plans/2026-09-28-jupyter-windows-detection-and-env.md Phase 3."}}}
 {"id":"bd-4qi4j3sa","title":"quarto-hub-mcp 'live:' vitest suites are network-flaky and unconditionally gate cargo xtask verify","description":"ts-packages/quarto-hub-mcp/src/hub-mcp.test.ts has two 'live:' describe suites (~12 tests, 15s timeouts each) that hit the third-party public sync server wss://sync.automerge.org with no gate: plain 'vitest run' — the ts-packages step of cargo xtask verify — always runs them. Observed 2026-07-14: two of three full verify runs failed with 'Test timed out in 15000ms' on a DIFFERENT subset of live tests each time (9 then 2), while direct vitest reruns minutes later and GitHub Actions CI were green; the changed code (pampa lua) was unrelated. Two problems: (1) an external-network hiccup fails the whole pre-push verify gate; (2) the 'live: create project and mutate files' suite creates an orphan project on the public server on EVERY full test run, from every dev machine and CI job. Proposed: gate live suites behind an opt-in env var (describe.runIf(process.env.HUB_MCP_LIVE_TESTS)), keep them enabled in a dedicated CI job, and consider pointing the mutate suite at a locally spawned hub binary (hub --allow-insecure-auth) instead of sync.automerge.org.","status":"closed","priority":2,"issue_type":"chore","created_at":"2026-07-14T20:38:25.031343Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-15T10:59:46.171334Z","closed_at":"2026-10-15T10:59:46.171334Z","close_reason":"Fixed by PR #839 (merged 2026-10-14): sync tests run against an in-process test hub; the sync.automerge.org suite is gated behind HUB_MCP_LIVE=1 (describe.skipIf at hub-mcp.test.ts:622), so plain vitest run / cargo xtask verify no longer hits the network.","dependencies":{"bd-ixnp4uqj:discovered-from":{"depends_on_id":"bd-ixnp4uqj","type":"discovered-from","created_at":"2026-07-14T20:38:25.031343Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-4qjl87ax","title":"Engine-output headings can collide with document heading ids (dedup runs pre-engine)","description":"Follow-up scoped out of bd-duplicate-heading-ids-mou5z7ux by design decision (2026-08-18): the document-level heading-id dedup pass runs at the tail of IncludeExpansionStage, but EngineExecutionStage runs later, so a heading emitted by a knitr/jupyter code cell can still duplicate an existing heading id. Q1 does not have this problem because engines run before pandoc parses (ids assigned last). Fixing it in q2 means renumbering after engine execution — in tension with DocumentProfileStage (outline) running pre-engine and wanting final ids. Needs its own design pass: either a second post-engine dedup for engine-inserted headers only, or moving the authoritative pass later and teaching the profile to tolerate it. No known real-world hit yet; filed for bookkeeping.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-08-18T17:24:48.182022Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-18T17:34:06.297280Z","labels":["markdown"],"dependencies":{"bd-duplicate-heading-ids-mou5z7ux:discovered-from":{"depends_on_id":"bd-duplicate-heading-ids-mou5z7ux","type":"discovered-from","created_at":"2026-08-18T17:24:48.182022Z","created_by":"Carlos Scheidegger"}},"comments":{"c-hrdrt8vt":{"id":"c-hrdrt8vt","author":"Carlos Scheidegger","created_at":"2026-08-18T17:34:06.297280Z","text":"Design note from parent strand (2026-08-18): the include-stage fix landed on a scoped uniqueIdent mechanism (renameable set = injected headers, seen-set = whole document, ids monotonic once assigned). This strand's intended fix is re-running that same routine post-engine with scope = engine-inserted headers — no renaming of ids DocumentProfileStage already observed, so the profile tension noted in the description dissolves."}}}
@@ -312,6 +313,7 @@
 {"id":"bd-56b0","title":"Cross-doc dependency channel audit for q2 preview","description":"Enumerate every Quarto feature that creates a cross-document dependency the preview pipeline needs to react to: include shortcodes, listing content globs, bibliography/csl paths, theme SCSS imports, project-scoped resources, _extensions/ Lua filters, and anything else. For each: which DocumentProfile channel encodes it today (if any), which need to be added, which stay manual-refresh-only. Drives Phase B (channels already on DocumentProfile) and informs Phase D (new channels). Until this lands, the always-visible manual force-refresh button (Phase A.6) is the user escape hatch. See claude-notes/plans/2026-05-11-q2-preview-epic.md §Recommended next steps item 3 + Risk 2.","status":"open","priority":2,"issue_type":"task","created_at":"2026-05-11T15:40:54.896280Z","created_by":"cscheid","updated_at":"2026-05-11T15:40:54.896280Z","dependencies":{"bd-kw93:related":{"depends_on_id":"bd-kw93","type":"related","created_at":"2026-05-11T15:40:54.896280Z","created_by":"cscheid"}}}
 {"id":"bd-56hbab5q","title":"Cut q2 v0.2.0 release","description":"Bump workspace version 0.1.1 -> 0.2.0, tag v0.2.0, ship multi-platform signed binaries via Release workflow. Follows claude-notes/instructions/release-runbook.md","status":"closed","priority":1,"issue_type":"task","created_at":"2026-06-16T13:18:30.266125Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-16T14:14:09.411561Z","closed_at":"2026-06-16T14:14:09.411561Z","close_reason":"v0.2.0 released and end-to-end verified","comments":{"c-f6u34rjk":{"id":"c-f6u34rjk","author":"Carlos Scheidegger","created_at":"2026-06-16T13:49:38.218204Z","text":"PR #292 merged to main (HEAD 7de45f9c). Tagged v0.2.0 and pushed; Release workflow run 27622326574 triggered (queued). Watching."},"c-pcpmstkn":{"id":"c-pcpmstkn","author":"Carlos Scheidegger","created_at":"2026-06-16T14:14:08.561366Z","text":"Release published: https://github.com/quarto-dev/q2/releases/tag/v0.2.0 — all 5 build legs green, GitHub Release job published all assets (5 platforms; Unix tar.gz + .sha256 + .minisig, Windows zip + .sha256, combined checksums.sha256). Step-6 verification all green: install.sh one-liner resolved v0.2.0 with checksum+signature verified; q2 --version='q2 (quarto 2) 0.2.0'; all 3 hub defaults report 'bundled'; resolves_latest_version_from_github network test passed."},"c-sqj020bf":{"id":"c-sqj020bf","author":"Carlos Scheidegger","created_at":"2026-06-16T13:25:08.741618Z","text":"Version bumped 0.1.1->0.2.0 on branch release/v0.2.0 (commit eaa26e6c). Full cargo xtask verify green after fixing a stale node_modules (npm install added @vitejs/plugin-basic-ssl). q2 --version reports 0.2.0. Awaiting user approval to push branch, merge PR, and push v0.2.0 tag."}}}
 {"id":"bd-5706gcrq","title":"Rich Markdown in document titles is stringified in the title-block h1","description":"Document titles with inline Markdown (code spans, emphasis, etc.) render as plain text in <h1 class=\"title\">. Root cause: PandocInlines title metadata is flattened to plain text before reaching the template (Site 1: template.rs config_value_to_template_value:660-663; Site 2: title_block.rs extract_title/create_title_header). pagetitle (head <title>) is correctly plain via derive_pagetitle and must stay so. Fix: render allowlisted body title-block fields to HTML via pampa::writers::html::write_inlines_to. Affects both q2 render and q2 preview (shared quarto-core path). Plan: claude-notes/plans/2026-06-15-rich-markdown-titles.md","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-06-15T21:52:09.222689Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-27T19:22:34.307502Z","closed_at":"2026-07-27T19:22:34.307502Z","close_reason":"Landed: PR #291 merged 2026-06-16 (rich Markdown preserved in title-block h1). Audit bd-a0eyjshu.","labels":["bug"],"comments":{"c-046f8azz":{"id":"c-046f8azz","author":"Carlos Scheidegger","created_at":"2026-06-15T22:00:06.286454Z","text":"Decisions: (D1) defer author/date richness to a follow-up; this strand does title/subtitle/abstract. (D2) fix both Site 1 (full HTML) and Site 2 (minimal HTML); non-HTML formats don't render in Q2 yet (only render_html stage exists), out of scope. (D3) ship a CLEAN h1 (no data- annotations) via write_inlines_to — matches the unannotated body in the CLI render path (include_source_locations:false) and fixes the bug in render+preview. Source-annotated h1 is a preview-only follow-up: h1 is built in ApplyTemplateStage from an already-serialized body + meta, with no access to the AST/pointer-keyed source map; title inlines do carry source_info substring spans (pampa/src/pandoc/meta.rs:358), so the follow-up would annotate from those directly."},"c-3y5u51em":{"id":"c-3y5u51em","author":"Carlos Scheidegger","created_at":"2026-06-15T22:29:18.349602Z","text":"Verification complete. cargo xtask verify: all 14 steps pass (incl WASM build + hub-client build/tests). Workspace: 10049 tests pass. Zero .snap churn (only template.rs + title_block.rs changed). Gotcha: first verify run failed on a clippy unused-import (-D warnings) for a test-only Code import at module scope — moved it into the tests module; re-verified clean. Outstanding: live-browser q2 preview confirmation not yet done (the WASM/SPA leg rebuilt fine and preview shares the same quarto-core template path the render e2e exercised, so the fix is in the image — flagging per the end-to-end policy). Implementation done across template.rs (Site 1) and title_block.rs (Site 2); ready for review. Follow-up bd-z37euevy tracks preview-path source annotations."},"c-c4eagmkh":{"id":"c-c4eagmkh","author":"Carlos Scheidegger","created_at":"2026-06-15T22:16:52.509927Z","text":"Implemented both sites (TDD). Site 1 (template.rs): title/subtitle/abstract now rendered to HTML via pampa write_inlines_to/write_blocks_to (metadata_entry_to_template_value + titleblock_field_to_html); pagetitle stays plain. Site 2 (title_block.rs): extract_title_inlines + create_title_header(Vec<Inline>) preserve inline markup in minimal-mode injected h1, recursively stamping title-block Generated provenance (atomic boundary + no-preimage contract preserved). 6 new unit tests pass; full workspace 10049 tests pass with no snapshot churn. E2E verified: q2 render brand.qmd -> <h1 class=title>...<code>_brand.yml</code></h1>; theme:none fixture -> <h1>...<code>_brand.yml</code></h1>. cargo xtask verify (full, incl WASM/hub) running. Plan: claude-notes/plans/2026-06-15-rich-markdown-titles.md"},"c-ez6ld1qw":{"id":"c-ez6ld1qw","author":"Carlos Scheidegger","created_at":"2026-06-15T23:17:01.305355Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/291 (branch feature/bd-5706gcrq-rich-markdown-titles off main). Commit aae92460. Awaiting CI."}}}
+{"id":"bd-57kue1fs","title":"website.navbar.background/foreground should feed $navbar-bg/$navbar-fg into the theme SCSS","description":"## Symptom\n\nSetting `website.navbar.background: light` (as quarto.org does) only changes the rendered markup in Q2: `navbar_to_html` in `crates/quarto-navigation/src/render_html.rs` adds the Bootstrap `bg-light` class and `data-bs-theme=\"light\"` to the `<nav>`. The compiled theme CSS still contains the `.navbar { background-color: $primary; color: $navbar-fg }` rule computed from `_bootstrap-variables.scss`, with `$navbar-fg` chosen for contrast against `$primary`. The result is a near-white navbar (from `bg-light !important`) with white nav links, i.e. unreadable.\n\nFound while porting the quarto.org theme to `docs/` (commit after 3f7d4b0e8). Stopgap in `docs/theme/quartoorg.scss`: a hand-written `$navbar-bg: #f8f9fa;` default, which is what Q1 would have synthesized. Remove that line (and its comment) once this lands.\n\n## What Q1 does\n\n`quartoBootstrapDefaults` in `external-sources/quarto-cli/src/format/html/format-html-scss.ts` (around line 530) forwards three navbar keys into the SCSS defaults layer:\n\n- `background` → `$navbar-bg: <value>`; a Bootstrap color name (`primary`, `secondary`, `success`, `info`, `warning`, `danger`, `light`, `dark`) becomes `$<name>` and the matching `$<name>: <hex> !default;` fallback is prepended (`bootstrapColorVars`, line ~791); any other string is emitted verbatim.\n- `foreground` → `$navbar-fg`, same treatment.\n- `toggle-position` → `$navbar-toggle-position`.\n\nThe same block also forwards `website.sidebar.background/foreground/border` → `$sidebar-bg/$sidebar-fg/$sidebar-border`.\n\n`_bootstrap-variables.scss` then derives `$navbar-fg`, `$navbar-hl`, `$navbar-brand`, etc. with `theme-contrast()` from the supplied `$navbar-bg`, which is why the links come out readable in Q1.\n\n## Where it goes in Q2\n\n`CompileThemeCssStage` in `crates/quarto-core/src/stage/stages/compile_theme_css.rs` already builds a doc-vars layer (currently only `$sidebar-border` from `website.sidebar.style`, line ~86) and hands it to `compile_with_doc_vars` in `crates/quarto-sass/src/compile.rs`, whose doc comment lists `$navbar-bg` as a planned case. Add `$navbar-bg` / `$navbar-fg` (and ideally `$sidebar-bg` / `$sidebar-fg`, `$navbar-toggle-position`) there, with the Bootstrap color-name mapping. The cache key (line ~440) already hashes the doc-vars string, so distinct values get distinct cache entries.\n\nKeep the existing class/attribute emission in `render_html.rs`; Q1 emits both.\n\n## Acceptance\n\n- A site with `website.navbar.background: light` and the default cosmo theme renders a light navbar with dark, readable nav links in light mode; `background: dark` / `primary` / a hex value likewise get contrast-correct links.\n- A dark theme that sets its own `$navbar-bg` (as `docs/theme/quartoorg-dark.scss` does) still wins, since user theme layers come after the doc-vars layer... verify the layer order gives the user's theme file precedence, matching Q1.\n- `docs/theme/quartoorg.scss` can drop its `$navbar-bg` stopgap and the docs navbar is unchanged.\n- Unit test in `compile_theme_css.rs` alongside the `$sidebar-border` one.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-27T21:26:27.397809Z","created_by":"cscheid","updated_at":"2026-10-27T21:26:27.397809Z","labels":["theme","website"]}
 {"id":"bd-57y4","title":"Vendor and integrate quarto-listing.scss with theme-CSS pipeline","description":"Per L3 D5 (decided 2026-05-06), the listing JS pair (list.min.js + quarto-listing.js) shipped via the artifact store as Project-scoped js: artifacts in the L3 phase 7 commit. The third asset, quarto-listing.scss, was deferred because it requires Bootstrap variable wiring + media-breakpoint mixins from the existing theme-CSS pipeline (CompileThemeCssStage / quarto_sass::SassLayer). That's a larger design task than the static JS bundling.\n\nThis issue: integrate the listing SCSS as a SassLayer that gets composed with the website's theme bundle. The Q1 source file lives at external-sources/quarto-cli/src/resources/projects/website/listing/quarto-listing.scss; copy to resources/listing/quarto-listing.scss per the External Sources Policy and add it to the theme bundle when at least one listing is rendered.\n\nWhile the SCSS isn't compiled, listings render with default browser styling — the markup and JS are unchanged. The list / grid / table layouts work but lack the polished card / table styling Q1 ships.\n\nDiscovered while shipping L3 phase 7 (bd-ml8z); see D5 in claude-notes/plans/2026-05-06-listings-L3-resolve-transform.md.\n\nL5 (bd-5vsr) lands the markup that consumes this SCSS; merging bd-57y4 restores Q1 visual parity for category sidebars and per-item category chips.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-05-06T20:52:15.894830Z","created_by":"cscheid","updated_at":"2026-07-29T19:19:10.196570Z","closed_at":"2026-07-29T19:19:10.196570Z","close_reason":"Vendored quarto-listing.scss into resources/scss/html/templates/ and composed it as a built-in SassLayer at all HTML assembly sites (native+wasm, themed+default). Deviation from the strand sketch, recorded in the plan: unconditional like Q1 rather than listing-gated (stage ordering + single shared theme CSS per site). $theme-name overrides already flow from the vendored bootstrap layer. Plan: claude-notes/plans/2026-07-29-listing-scss.md. e2e verified on the bd-r1by4u2a blog scaffold.","dependencies":{"bd-ml8z:discovered-from":{"depends_on_id":"bd-ml8z","type":"discovered-from","created_at":"2026-05-06T20:52:15.894830Z","created_by":"cscheid"}}}
 {"id":"bd-5adgk9hr","title":"Typst: css-property-processing: none is silently ignored (filter param never forwarded)","description":"Setting `css-property-processing: none` has no effect — q2 always\nperforms CSS-to-Typst border/color/vertical-align translation for raw\nHTML (tables especially), the same as the (correct) default `translate`\nbehavior. Found via `css-property-processing/none.qmd` in Q1's\nsmoke-all corpus (`external-sources/quarto-cli/tests/docs/smoke-all/typst/css-property-processing/`).\n\n## Not a missing-feature bug — a missing-param-forward bug\n\nq2 does not reimplement Q1's Typst CSS-property-translation logic in\nRust. It shells out to the real, unmodified `pandoc -L\n<QUARTO_SHARE_PATH>/filters/main.lua` chain\n(crates/quarto-core/src/pandoc_filters/harness.rs:35,102) — i.e. Q1's\nactual Lua filters run as-is. `main.lua` imports\n`quarto-post/typst-css-property-processing.lua` (main.lua:77), whose\ntop-level guard is:\n\n    if not _quarto.format.isTypstOutput() or\n      param(_quarto.modules.constants.kCssPropertyProcessing, 'translate') ~= 'translate' then\n      return {}\n    end\n\n`param(key, 'translate')` falls back to `'translate'` when the named\nparam is absent from the filter-params blob Pandoc receives — which is\nexactly what always happens today, because q2 never puts\n`css-property-processing` into that blob in the first place.\n\n`crates/quarto-core/src/pandoc_filters/typst_params.rs`'s\n`TypstFilterParamsContributor` forwards `brand`, `logo`, `brand-mode`,\n`typst-available-fonts`, `citation-location`, `reference-location`,\n`typst-root-dir`, `cite-method` into the params blob (`contribute()`,\nlines 63-103) — no `css-property-processing` field exists. Compare to\nQ1's own generic (non-typst-specific) equivalent:\n`external-sources/quarto-cli/src/command/render/filters.ts:951`,\n`[kCssPropertyProcessing]: format.metadata[kCssPropertyProcessing]`.\n\nSame shape as bug #1 (font-paths, already fixed via #750/#752): a real,\nworking Lua filter with a real config surface, made unreachable\nbecause the Rust glue that assembles filter params drops the\ndocument's own metadata key on the floor.\n\n## Fix\n\nAdd a `css_property_processing: Option<String>` field to\n`TypstFilterParamsContributor` (typst_params.rs), populated from the\ndocument's merged `css-property-processing` metadata (same\nmerge/flattening path `brand_mode` already uses per this file's own\ndoc comment — `MetadataMergeStage`), and forward it in `contribute()`\nas the `\"css-property-processing\"` key when `Some`. Since the Lua\nfilter's own `isTypstOutput()` guard already scopes this to Typst,\nfollowing the existing typst-only contributor pattern here (rather\nthan Q1's generic-for-every-format approach) is fine and consistent\nwith how `brand`/`brand-mode` are handled in this same file today.\n\n## Non-goals\n- Do not touch `juice()`/`astpipeline.lua` — it is NOT gated by\n  `css-property-processing` in Q1 either (it's an earlier, unconditional\n  CSS-inlining normalization pass feeding the property-processing\n  filter's input; it already runs correctly today since q2 runs the\n  real, unmodified `main.lua` chain). No juice-related work is needed\n  for this strand.\n- Do not attempt a Rust reimplementation of the CSS-to-Typst\n  translation logic — it already exists and works (real Lua filter);\n  this is purely a param-forwarding gap.\n\n## Verification\n- `default.qmd` / `translate.qmd` in the same fixture directory already\n  pass today (they want the default `translate` behavior, which\n  happens \"by accident\" via the missing-param fallback) — make sure\n  the fix doesn't regress them once the param is explicitly forwarded.\n- Port `css-property-processing/none.qmd` into\n  `crates/quarto/tests/smoke-all` alongside the fix.\n\nOriginally surfaced during Typst smoke-all triage:\nclaude-notes/plans/2026-09-29-typst-smoke-all-followup.md (workspace-4),\nfinding #13.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-09-30T19:35:42.639402Z","created_by":"Gordon Woodhull","updated_at":"2026-10-02T17:21:17.138995Z","closed_at":"2026-10-02T17:21:17.138995Z","close_reason":"Fixed in PR #764 (merged)","labels":["typst"],"comments":{"c-jk3qi60y":{"id":"c-jk3qi60y","author":"Gordon Woodhull","created_at":"2026-09-30T19:42:41.730875Z","text":"Dispatched to agent css-property-processing on workspace-7 (branch braid/bd-5adgk9hr-css-property-processing)."}}}
 {"id":"bd-5aklrxgi","title":"Q2 has no number-sections / @sec- header-numbering implementation for any format","description":"Confirmed by direct code investigation (2026-09-17), while auditing the pandoc-writer-hybrid\nepic's crossref-numbering:external mode for docx/pptx.\n\nQ2 appears to have NO implementation of two related, standard Quarto/Pandoc features, for ANY\noutput format (not just Pandoc targets):\n\n1. `number-sections: true` — visible \"1.2 Section Title\" numbering injected into headers.\n2. `@sec-<id>` crossref resolution — resolving a `@sec-intro` reference to a linked, numbered\n   section reference (e.g. \"see Section 1.2\").\n\nEvidence:\n- `grep -rn number-sections crates/quarto-core/src/` finds exactly 2 hits, both metadata\n  passthrough/tests (extension/read.rs, stage/stages/metadata_merge.rs) — no header-numbering\n  logic anywhere.\n- `grep -rln number-sections resources/scss/` finds nothing — not even a CSS-counter-based\n  implementation (a common alternative for HTML).\n- `crates/quarto-core/src/transforms/crossref_index.rs` DOES track a section-counter stack\n  (`advance_sections`, driven by `Header` blocks) — but its own module doc says this exists only\n  to compute compound numbers for OTHER crossref targets (\"Figure 1.2.3\"), not to number or\n  register sections themselves. `visit_header` never calls `index_custom_target` for the header\n  itself, and never injects a number into the header's own inline content.\n- `crates/quarto-core/src/crossref/registry.rs:83` registers `sec` → \"Section\" as a builtin\n  ref-type display name, but `crates/quarto-core/src/transforms/crossref_resolve.rs` has no\n  handling at all for a `sec`-typed ref or for `CrossrefIndex::headings` (the `HeadingRecord` list\n  `crossref_index.rs` does populate, apparently for a different, not-yet-built future use —\n  \"needed for cross-file book fixup in future phases\").\n\nRelevance to the pandoc-writer-hybrid epic (see `claude-notes/designs/pandoc-hybrid-architecture.md`\n§11): under `crossref-numbering: external` (the mode the epic sets for docx/pptx targets), Q1's\nown `sections.lua` — which normally provides BOTH capabilities natively via Pandoc's own header\nnumbering — is suppressed wholesale as part of the same filter group P3 audited\n(`quarto_crossref_filters`, gated at `main.lua:718`). If Q2 has no replacement, `--to docx` with\n`number-sections: true` (or any `@sec-` reference) will silently produce unnumbered headers /\nunresolved section refs, with no error. This is a real regression risk for docx/pptx v1, not a\ntheoretical one.\n\nScope note: this strand is about whether/how to implement the feature at all (which predates and\nis broader than the pandoc-hybrid epic — it would also be missing for a plain HTML render today,\nif anyone tested it, which nobody appears to have). The pandoc-hybrid epic itself should NOT try\nto build this from scratch as part of its own scope; the epic's design doc records this as an\nopen, not-yet-assigned-to-a-plan question and should reference this strand instead of\nre-investigating.","status":"open","priority":1,"issue_type":"feature","created_at":"2026-09-18T02:18:13.666545Z","created_by":"Gordon Woodhull","updated_at":"2026-09-18T02:18:13.666545Z"}
@@ -335,10 +337,10 @@
 {"id":"bd-5oyk1xce","title":"q2 preview drops engine include-in-header (marimo islands script) — widgets never hydrate","description":"ROOT CAUSE (code + browser confirmed). In q2 preview, marimo cells render their static server-side island markup but never hydrate / execute dynamically. The engine's include-in-header content (the @marimo-team/islands cdn.jsdelivr.net <script type=module>, the __MARIMO_EXPORT_CONTEXT__ trust-marker, the <marimo-code hidden> notebook source, and islands style.css) never reaches the preview pane's <head>.\n\nMECHANISM: On the render path, EngineExecutionStage (crates/quarto-core/src/stage/stages/engine_execution.rs:416-424) copies result.includes.header_includes into ctx.includes, and apply_template.rs:149 late-folds ctx.includes into the rendered <head> slots. On the PREVIEW path, EngineExecutionStage is bypassed and CaptureSpliceStage (crates/quarto-core/src/stage/stages/capture_splice.rs:138-146) extracts ONLY capture.result.markdown, explicitly discarding capture.result.includes. So ctx.includes.header_includes stays empty -> the islands <script> is never emitted -> no client JS -> inert widgets. The full ExecuteResult (incl. includes.header_includes) IS present in the capture (serialized at engine_execution.rs:385-391); it is dropped only by the splice stage.\n\nBROWSER EVIDENCE (headed Chromium via q2 preview on the marimo mo.ui.slider fixture): iframe __MARIMO_EXPORT_CONTEXT__=undefined; <marimo-code> absent; <marimo-island>/<marimo-ui-element> present (body markup delivers fine); ZERO network requests to jsdelivr (script tag simply absent, not blocked); console clean (matches 'no errors'); iframe sandbox='allow-scripts allow-same-origin'; CSP meta = [] (empty). q2 render of the same doc emits all three markers inside <head>. => Hypothesis #1 (dropped include-in-header) CONFIRMED; hypothesis #2 (CSP/sandbox blocking) REFUTED.\n\nPROPOSED FIX: In CaptureSpliceStage, after (or alongside) the block splice, read capture.result.includes.{header_includes,include_before,include_after} and extend ctx.includes.* with them — mirroring engine_execution.rs:416-424. The existing apply_template.rs:149 late fold then delivers them to the head via the same channel q2 render uses. Small, surgical, no new architectural surface. Multi-engine: fold each capture's includes in the same order the blocks are folded. Needs a native seam test (ctx.includes populated after splice) + an e2e assertion that the islands <script src=jsdelivr> is present in the preview iframe <head> (extend engine-capture-splice-marimo.spec.ts SC23, whose header currently scopes hydration as out-of-scope).\n\nRelated: bd-5jxcio5d (splice, CLOSED — block delivery is correct and separate). Plan: claude-notes/plans/2026-07-07-plan4c2-marimo-preview-e2e.md (Open decision / FINDING #5). Do NOT conflate with bd-5m1ni9if (RawBlock-misconsume edge).","status":"open","priority":1,"issue_type":"bug","created_at":"2026-07-08T15:08:36.148724Z","created_by":"Gordon Woodhull","updated_at":"2026-07-27T19:23:43.293669Z","dependencies":{"bd-5jxcio5d:discovered-from":{"depends_on_id":"bd-5jxcio5d","type":"discovered-from","created_at":"2026-07-08T15:08:36.148724Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-5px05kui","title":"hub-client changelog.md breaks WASM render test: '~37MB' parses as unclosed subscript (Q-2-17)","description":"The 2026-07-07 changelog entry (commit 6cd4dd5a) contains '~37MB'; qmd's parser treats the unmatched '~' as an opening subscript delimiter and errors with Q-2-17 'Unclosed Subscript', failing src/services/changelogRender.wasm.test.ts (changelog.md renders successfully) and therefore 'cargo xtask verify' / hub-client test:ci on main. Immediate fix: escape as '\\\\~37MB'. Open UX question (separate): Pandoc treats an unmatched '~' as literal text rather than an error — Q-2-17's hard error means any prose tilde (file sizes, approximations) breaks a render; consider whether unclosed subscript should degrade to literal text or stay a hard error.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-07-07T19:39:40.066006Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-07T19:40:22.075992Z","closed_at":"2026-07-07T19:40:22.075992Z","close_reason":"Duplicate: an open PR already fixes the changelog tilde / Q-2-17 render failure","dependencies":{"bd-abo9m23f:discovered-from":{"depends_on_id":"bd-abo9m23f","type":"discovered-from","created_at":"2026-07-07T19:39:40.066006Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-5qa1zvmr","title":"JSON reader errors lack Q-* catalog codes","description":"The entire JSON reader path (readers/json.rs JsonReadError, surfaced via 'Error reading JSON: {e}' in main.rs) reports plain error enums with no Q-* codes, unlike the writer path which uses DiagnosticMessage + catalog codes. Noticed while adding the raw-json reader rejections (bd-en2hvrwn) — those followed the existing convention rather than bolt on a one-off catalog integration. If/when reader errors get catalog treatment, do the whole JsonReadError enum at once.","status":"open","priority":3,"issue_type":"chore","created_at":"2026-07-17T16:50:09.673503Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-17T16:50:09.673503Z","dependencies":{"bd-en2hvrwn:discovered-from":{"depends_on_id":"bd-en2hvrwn","type":"discovered-from","created_at":"2026-07-17T16:50:09.673503Z","created_by":"Carlos Scheidegger"}}}
-{"id":"bd-5qepzwst","title":"Lua quarto.version is hardcoded to {0,1,0}; should derive from the effective CLI version","description":"crates/pampa/src/lua/quarto_api.rs register_quarto_version sets quarto.version = {0,1,0} unconditionally. Extensions doing minimum-version checks via quarto.version see 0.1.0 regardless of the real version (0.32.0 today; nightlies report e.g. 0.33.0-nightly.YYYYMMDD via QUARTO_VERSION_OVERRIDE, see bd-p4ljdp2e). Should parse quarto_util::cli_version() into {major, minor, patch} (prerelease suffix dropped or exposed separately). Found while inventorying CARGO_PKG_VERSION consumers for the nightly override; out of scope there.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-09-19T16:51:18.246258Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-19T16:51:18.246258Z","dependencies":{"bd-p4ljdp2e:discovered-from":{"depends_on_id":"bd-p4ljdp2e","type":"discovered-from","created_at":"2026-09-19T16:51:18.246258Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-5qepzwst","title":"Lua API: quarto.version is the plain table {0,1,0}; make it a real, comparable pandoc.types.Version (GH #942)","description":"register_quarto_version (crates/pampa/src/lua/quarto_api.rs) sets quarto.version = {0,1,0} unconditionally. In Quarto 1 it is a pandoc Version object, so filters do quarto.version >= '1.4' or pandoc.types.Version(...). In q2 the value is 0.1.0 regardless of the real version (0.36.0), comparisons fail with 'attempt to compare string with table', and pandoc.types does not exist. PANDOC_VERSION / PANDOC_API_VERSION have the same problem (plain tables), and q2's own vendored resources/pandoc-filters/filters/layout/table.lua:155 does PANDOC_VERSION < pandoc.types.Version('2.11.3.2'). Reproduced 2026-10-27 on main at 3f7d4b0e8 (GH #942, reported by atheriel porting the Connect docs).\n\nFix: new crates/pampa/src/lua/version.rs with a LuaVersion userdata: pandoc.types.Version(spec) accepting string/number/table/Version; __eq/__lt/__le registered as meta *functions* over (Value, Value) so either operand may be the Version (Lua calls __le('1.4', v) for v >= '1.4'); lexicographic comparison, shorter prefix is less; __tostring -> '0.36.0'; __index for integer components and must_be_at_least; __len (with __index + __len, Lua 5.4 table.concat accepts the userdata, so table.concat(quarto.version, '.') keeps working -- pin by test); pandoc.utils.type -> 'Version'. quarto.version = Version(leading dotted-numeric prefix of quarto_util::cli_version()), so nightlies '0.37.0-nightly.YYYYMMDD' give 0.37.0 (Q1's leadingDottedNumber). PANDOC_VERSION and PANDOC_API_VERSION become Version objects with today's components. Check the lua-conformance xfail lists for Version cases that start passing.\n\nPlan: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-09-19T16:51:18.246258Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T21:54:35.502055Z","external_ref":"https://github.com/quarto-dev/q2/issues/942","labels":["lua","pampa","parity"],"dependencies":{"bd-p4ljdp2e:discovered-from":{"depends_on_id":"bd-p4ljdp2e","type":"discovered-from","created_at":"2026-09-19T16:51:18.246258Z","created_by":"Carlos Scheidegger"},"bd-pqj37chw:parent-child":{"depends_on_id":"bd-pqj37chw","type":"parent-child","created_at":"2026-10-27T20:22:40.559453Z","created_by":"claude"}},"comments":{"c-5lqrcp7h":{"id":"c-5lqrcp7h","author":"claude","created_at":"2026-10-27T21:17:53.501959Z","text":"Implemented on braid/bd-5qepzwst-lua-api-quartoversion-plain, merged into feature/lua-api-q1-parity (a921415e1). pandoc.types.Version userdata in crates/pampa/src/lua/version.rs; quarto.version from quarto_util::cli_version(); PANDOC_VERSION/PANDOC_API_VERSION are Versions; pandoc.utils.type -> Version. 8 new unit tests incl. table.concat on the userdata; pampa lua:: (1033) + conformance ratchet green; clippy clean. GH #942 probe now prints 0.36.0 / Version / comparisons ok. Left open until the epic's PR merges."},"c-b11erzr4":{"id":"c-b11erzr4","author":"claude","created_at":"2026-10-27T21:54:35.502055Z","text":"PR: https://github.com/quarto-dev/q2/pull/949 (feature/lua-api-q1-parity -> main; closes GH #939-#942 on merge). Full cargo xtask verify green incl. hub-client tests on a rebuilt WASM; Windows cross-check not possible from macOS, flagged in the PR."}}}
 {"id":"bd-5qmh5acq","title":"tree-sitter-qmd: citation keys should admit Unicode letters (@é, @Müller2020)","description":"Pandoc's citation keys start with any Unicode letter, digit or _ and allow Unicode letters inside the key, so '@Müller2020' and 'café @élan' are citations there. Ours (grammar.js citation rule '[0-9A-Za-z_]+([:.#$%&+?<>~/-][0-9A-Za-z_]+)*', and the key-start check added in scanner.c by bd-bare-at-literal-w3ytmu8e) are ASCII-only, so these are uncoded parse errors today (table A2 of claude-notes/plans/2026-09-25-bare-at-as-str.md). Widen both to Unicode letters and digits, keeping pandoc's single-internal-punctuation rule.","status":"open","priority":3,"issue_type":"task","created_at":"2026-09-25T21:41:40.519497Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-25T21:41:40.519497Z","labels":["parser"],"dependencies":{"bd-bare-at-literal-w3ytmu8e:discovered-from":{"depends_on_id":"bd-bare-at-literal-w3ytmu8e","type":"discovered-from","created_at":"2026-09-25T21:41:40.519497Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-5qnj","title":"Manage trace size for use as replay/regression-test fixtures","description":"Spun out of bd-45yw (replay engine). Once traces double as replay fixtures (and as user-attached bug-report artifacts), trace size becomes a real constraint — they will be checked into the repo as regression fixtures and posted by users in issues.\n\nCurrent quarto-trace output captures per-stage pipeline state (see crates/quarto-trace/src/lib.rs and crates/quarto-core/src/stage/trace.rs::JsonTraceObserver), which is already on the heavy side for routine diagnostic use.\n\nInvestigate:\n- Where the bulk lives in current traces (per-stage AST snapshots? supporting_files content? something else?)\n- Which content is actually load-bearing for replay vs. diagnostics, and whether the two roles can share one artifact (the design preference noted in bd-45yw's plan)\n- Compression on disk; lazy loading on read\n- Whether per-stage AST snapshots can be elided/diffed when not needed\n- Size budgets — what is reasonable for (a) checked-in CI fixtures and (b) user-attached bug reports\n\nGoal: make 'one trace serves both diagnostic and replay roles' practical. If size cannot be bounded, fall back to two separate artifacts and document why.\n\nReferences:\n- crates/quarto-trace/src/lib.rs (TraceDocument, TraceEntry)\n- crates/quarto-core/src/stage/trace.rs (JsonTraceObserver)\n- claude-notes/plans/2026-05-03-replay-engine.md (open subquestion in Phase 1)","status":"closed","priority":2,"issue_type":"task","created_at":"2026-05-03T21:11:19.859036Z","created_by":"cscheid","updated_at":"2026-05-03T23:29:05.405253Z","closed_at":"2026-05-03T23:29:05.405068Z","close_reason":"Phases 1, 2, 3, and 5 complete. End-to-end verified on bd-5qnj fixtures: big.qmd 16.3 MB pretty -> 62 KB gzipped+deduped (~265x). Unified-artifact promise realized post-merge with bd-45yw. Real-engine supporting_files re-measurement tracked as bd-sr73.","dependencies":{"bd-45yw:related":{"depends_on_id":"bd-45yw","type":"related","created_at":"2026-05-03T21:11:19.859036Z","created_by":"cscheid"}}}
-{"id":"bd-5r97wehd","title":"page-footer: raw HTML <img src> in footer text/items is neither copied to _site nor warned about (Q-5-6)","description":"WHAT HAPPENS. An image written as raw HTML in a _quarto.yml page-footer region, e.g. `<img src=\"/_extensions/posit-docs/assets/images/cookie.svg\" class=\"light-content\">` (the form the posit-dev/product-doc-theme README recommends), is emitted unchanged but the file is never copied into _site/, so it 404s on every page. No Q-5-6 is raised either. The same image written as markdown (`![](/_extensions/.../cookie.svg){.light-content}`) is copied and works.\n\nREPRO (nightly 0.36.0-nightly.20261024): render product-doc-theme example/ (see claude-notes/plans/2026-10-21-product-doc-theme-q2.md, Reproduce). cookie.svg, cookie-copy.svg and posit-logos-2024_horiz-full-color.svg are missing from _site/_extensions/posit-docs/assets/images/; only images that are also referenced by markdown somewhere get copied.\n\nWHERE. crates/quarto-core/src/project/website_post_render.rs copy_footer_images -> collect_config_text_images / collect_items_images only collect Image inlines. RawInline/RawBlock with format html is skipped.\n\nFIX SKETCH. In the same collectors, scan RawInline/RawBlock(html) for <img src> (and probably <source srcset>/<link href> later) with an HTML-tag scanner rather than regex, then go through the existing probe/copy/Q-5-6 path. The span is the raw node span (no per-attribute span). Consider navbar/sidebar text too. Output location follows q2s current layout (_site/<same path>); the Q1 quarto-contrib relocation question is open question 1 in the plan.\n\nQ1: relocates anything under _extensions/ to site_libs/quarto-contrib/quarto-project/ and rewrites the src (projectExtensionPathResolver).","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-10-26T16:26:28.674645Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-26T16:40:57.694746Z","labels":["extensions","websites"],"dependencies":{"bd-0i6fii0k:discovered-from":{"depends_on_id":"bd-0i6fii0k","type":"discovered-from","created_at":"2026-10-26T16:26:28.674645Z","created_by":"Carlos Scheidegger"},"bd-page-footer-items-f4th80mj:related":{"depends_on_id":"bd-page-footer-items-f4th80mj","type":"related","created_at":"2026-10-26T16:26:28.674645Z","created_by":"Carlos Scheidegger"}},"comments":{"c-qca7wqtb":{"id":"c-qca7wqtb","author":"Carlos Scheidegger","created_at":"2026-10-26T16:40:57.694746Z","text":"2026-10-26: user direction is to fix the content, not q2. Verified pure-markdown spellings that work on BOTH q2 and Q1 (fixture footer-raw-html-img-investigation/markdown-spellings/): logo = markdown images + href; cookie item = [imgs]{#open_preferences_center title=...} with item href 'javascript:void(0)' (Q1 only renders footer item markdown when href is set; href '#' gets rewritten to '../#' by Q1). Open: confirm the GTM-injected consent handler still fires with the id on a span inside the <a>. Next: product-doc-theme PR, then close this as working-as-designed."},"c-yismph2r":{"id":"c-yismph2r","author":"Carlos Scheidegger","created_at":"2026-10-26T16:31:34.781709Z","text":"Investigated 2026-10-26 on branch braid/bd-5r97wehd-footer-raw-html-img. Plan skeleton: claude-notes/plans/2026-10-26-footer-raw-html-img.md (repro under footer-raw-html-img-investigation/). Reproduces at 4d966ea05 for footer center/items/bare items AND navbar text AND document body raw <img>. Verdict: needs a policy decision — docs/guides/projects/paths.qmd documents raw-HTML refs as intentionally uncollected (fc5pvkcv decision 3); documented 'project: resources:' workaround verified working; extension could likely ship the images via contributes.project.project.resources."}}}
+{"id":"bd-5r97wehd","title":"page-footer: raw HTML <img src> in footer text/items is neither copied to _site nor warned about (Q-5-6)","description":"WHAT HAPPENS. An image written as raw HTML in a _quarto.yml page-footer region, e.g. `<img src=\"/_extensions/posit-docs/assets/images/cookie.svg\" class=\"light-content\">` (the form the posit-dev/product-doc-theme README recommends), is emitted unchanged but the file is never copied into _site/, so it 404s on every page. No Q-5-6 is raised either. The same image written as markdown (`![](/_extensions/.../cookie.svg){.light-content}`) is copied and works.\n\nREPRO (nightly 0.36.0-nightly.20261024): render product-doc-theme example/ (see claude-notes/plans/2026-10-21-product-doc-theme-q2.md, Reproduce). cookie.svg, cookie-copy.svg and posit-logos-2024_horiz-full-color.svg are missing from _site/_extensions/posit-docs/assets/images/; only images that are also referenced by markdown somewhere get copied.\n\nWHERE. crates/quarto-core/src/project/website_post_render.rs copy_footer_images -> collect_config_text_images / collect_items_images only collect Image inlines. RawInline/RawBlock with format html is skipped.\n\nFIX SKETCH. In the same collectors, scan RawInline/RawBlock(html) for <img src> (and probably <source srcset>/<link href> later) with an HTML-tag scanner rather than regex, then go through the existing probe/copy/Q-5-6 path. The span is the raw node span (no per-attribute span). Consider navbar/sidebar text too. Output location follows q2s current layout (_site/<same path>); the Q1 quarto-contrib relocation question is open question 1 in the plan.\n\nQ1: relocates anything under _extensions/ to site_libs/quarto-contrib/quarto-project/ and rewrites the src (projectExtensionPathResolver).","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-10-26T16:26:28.674645Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T17:28:46.026759Z","closed_at":"2026-10-27T17:28:46.026759Z","close_reason":"Working as designed. q2 deliberately does not look inside raw HTML: per bd-root-relative-paths-design-fc5pvkcv and docs/guides/projects/paths.qmd ('Raw HTML is not rewritten'), files referenced only from raw <img>/<a> tags are neither copied nor warned about, and authors are told to write images in markdown (or list files under project: resources:). Markdown images in footer/navbar regions are copied, rebased per page, and raise Q-5-6 when missing. Fixed on the content side instead: posit-dev/product-doc-theme now has a q2 branch (install: q2 add posit-dev/product-doc-theme@q2, commit 6e4d99c) whose README and example write the footer images as markdown, e.g. text: '![Posit](/images/logo.svg){.light-content width=\"65px\"}' and, for the cookie control, text: '[![..](/images/cookie.svg){.light-content}![..](/images/cookie-copy.svg){.dark-content}](#){#open_preferences_center title=\"Update cookie preferences\"}'. Example renders 6/6, 0 errors/warnings. Record: claude-notes/plans/2026-10-26-footer-raw-html-img.md (on main @ 77cf3cef9).","labels":["extensions","websites"],"dependencies":{"bd-0i6fii0k:discovered-from":{"depends_on_id":"bd-0i6fii0k","type":"discovered-from","created_at":"2026-10-26T16:26:28.674645Z","created_by":"Carlos Scheidegger"},"bd-page-footer-items-f4th80mj:related":{"depends_on_id":"bd-page-footer-items-f4th80mj","type":"related","created_at":"2026-10-26T16:26:28.674645Z","created_by":"Carlos Scheidegger"}},"comments":{"c-qca7wqtb":{"id":"c-qca7wqtb","author":"Carlos Scheidegger","created_at":"2026-10-26T16:40:57.694746Z","text":"2026-10-26: user direction is to fix the content, not q2. Verified pure-markdown spellings that work on BOTH q2 and Q1 (fixture footer-raw-html-img-investigation/markdown-spellings/): logo = markdown images + href; cookie item = [imgs]{#open_preferences_center title=...} with item href 'javascript:void(0)' (Q1 only renders footer item markdown when href is set; href '#' gets rewritten to '../#' by Q1). Open: confirm the GTM-injected consent handler still fires with the id on a span inside the <a>. Next: product-doc-theme PR, then close this as working-as-designed."},"c-yismph2r":{"id":"c-yismph2r","author":"Carlos Scheidegger","created_at":"2026-10-26T16:31:34.781709Z","text":"Investigated 2026-10-26 on branch braid/bd-5r97wehd-footer-raw-html-img. Plan skeleton: claude-notes/plans/2026-10-26-footer-raw-html-img.md (repro under footer-raw-html-img-investigation/). Reproduces at 4d966ea05 for footer center/items/bare items AND navbar text AND document body raw <img>. Verdict: needs a policy decision — docs/guides/projects/paths.qmd documents raw-HTML refs as intentionally uncollected (fc5pvkcv decision 3); documented 'project: resources:' workaround verified working; extension could likely ship the images via contributes.project.project.resources."}}}
 {"id":"bd-5rl408mv","title":"Windows: HTML <link href> emits backslash separators for css/!path theme includes","description":"On Windows, a css/theme include reached from a subdirectory renders as href=\"..\\..\\shared\\styles.css\" instead of the expected forward-slash form ../../shared/styles.css. Surfaced as the one non-CRLF smoke-all failure (metadata/dir-metadata-paths/chapters/intro/doc.qmd asserts link[href=\"../../shared/styles.css\"]). Path resolution is correct; only the separator is wrong. Same family as closed bd-dff27o04 (JSON writer backslash separators, fixed via quarto_util::to_forward_slashes at ingress), different emission site: the HTML <link> href builder for resolved css/!path resources. Fix: apply to_forward_slashes at the href emission (or resolution) site. TDD: render a doc that pulls a css include from an ancestor/sibling dir, assert forward-slash href in the emitted <link>. Emission site still to be located precisely - start from the smoke-all dir-metadata-paths fixture and the HTML resource-link path in quarto-core.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-03T11:45:23.948243Z","created_by":"cderv","updated_at":"2026-07-03T14:36:27.104413Z","closed_at":"2026-07-03T14:36:27.104413Z","close_reason":"Fixed in commit 342c91bf: forward-slash normalization at !path emission + is_rooted guard for POSIX-absolute paths. smoke_all 1->0, directory_metadata_tests 15/15 green, workspace compiles.","dependencies":{"bd-dff27o04:related":{"depends_on_id":"bd-dff27o04","type":"related","created_at":"2026-07-03T11:45:54.278429Z","created_by":"cderv"}},"comments":{"c-1yrye4sx":{"id":"c-1yrye4sx","author":"cderv","created_at":"2026-07-03T14:32:34.089654Z","text":"Fixed on this worktree branch, commit 342c91bf. Root cause exactly as scoped: adjust_paths_to_document_dir (quarto-core/src/project/mod.rs) stringified the pathdiff result with to_string_lossy, leaking native backslashes into the css !path value used verbatim in the HTML <link href>. Fixed by normalizing with quarto_util::to_forward_slashes at the emission site (same helper as the SCSS/JSON-writer separator fixes). Found and fixed a second Windows defect in the same guard (Chris approved folding it in): the relative-path check used Path::is_relative, which on Windows wrongly classifies POSIX-absolute /usr/... as relative (not is_absolute without a drive prefix), so it was rebased into ../../../usr/...; now guarded on quarto_util::is_rooted. RED observed on 4 pre-existing directory_metadata_tests (backslash + absolute-path); all 15 green after. smoke_all 1->0 failures. The 3 residual engine_execution failures are pre-existing (confirmed by stash-revert), different subsystem, out of scope."}}}
 {"id":"bd-5rr4lgj1","title":"Reader does not strip a leading UTF-8 BOM (U+FEFF); it becomes prose content","description":"A file starting with the UTF-8 byte order mark (EF BB BF, U+FEFF) is not special-cased anywhere in the qmd reader. At 7ef59618 it is a hard parse error (U+FEFF is category Cf, which the prose regexes reject). Once bd-wuiu1of7 lands (\\p{Cf} accepted in prose) the file will parse, but the BOM survives as the first character of the first Str — and, when the file starts with YAML front matter, likely breaks the '---' delimiter match. Pandoc strips a leading BOM before parsing (Text.Pandoc.Readers.Markdown / the shared input normalisation). Fix shape: strip a leading U+FEFF once at the reader entry point (readers::qmd::read), before tree-sitter sees the bytes, keeping source offsets honest (the BOM is 3 bytes; source_info must account for it or the reader should record the trim). TDD: fixture with BOM + front matter + heading; assert metadata parsed and first Str has no U+FEFF.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-09-11T14:18:34.042416Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-11T14:18:34.042416Z","labels":["pampa","parity"],"dependencies":{"bd-wuiu1of7:discovered-from":{"depends_on_id":"bd-wuiu1of7","type":"discovered-from","created_at":"2026-09-11T14:18:34.042416Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-5sbguner","title":"Flaky: doc-inventory.test.ts drops the binary-file entry under load/cold caches","description":"quarto-sync-client's 'reports index, text, and binary docs with states and heads' failed twice on 2026-09-18 (inside cargo xtask verify, and once standalone via npm test -w) with the binary-file logo.png entry missing from the inventory; passed 4+ consecutive runs afterward including the identical invocation. Both failures were the first runs after a fresh npm install (automerge-repo 2.6.0-alpha.5), suggesting a timing race in binary-doc registration that loses under cold caches or parallel load. CI green on main. Env: macOS, Node 24.8.","status":"closed","priority":3,"issue_type":"bug","created_at":"2026-09-19T13:11:11.722905Z","created_by":"Andrew Holz","updated_at":"2026-10-10T15:41:22.331857Z","closed_at":"2026-10-10T15:41:22.331857Z","close_reason":"Duplicate of bd-c72wsugj (same doc-inventory flush race; root cause and fix tracked there, plan claude-notes/plans/2026-10-10-flaky-ci-tests-pass.md)","dependencies":{"bd-3fwtdhil:discovered-from":{"depends_on_id":"bd-3fwtdhil","type":"discovered-from","created_at":"2026-09-19T13:11:11.722905Z","created_by":"Andrew Holz"}}}
@@ -395,6 +397,7 @@
 {"id":"bd-706b0ixu","title":"Paste images from clipboard into the Monaco source editor","description":"Analogous to the existing editor drag-and-drop image flow, support pasting an image directly from the clipboard into the source editor: create the binary file in the project with an automatically generated (content-hash-based) filename — no dialog — and insert a markdown image reference at the cursor so the preview updates immediately. Must be safe under concurrent pastes from multiple CRDT peers, and must take a position on SVG (image/svg+xml can carry executable script). Design doc: claude-notes/plans/2026-08-27-paste-image-clipboard.md","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-08-27T14:03:59.423865Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-27T19:41:21.670639Z","closed_at":"2026-08-27T18:07:40.582018Z","close_reason":"Implemented on braid/bd-706b0ixu-paste-image-clipboard (e0e2d2bbc), all gates green, E2E verified in production bundle; plan claude-notes/plans/2026-08-27-paste-image-clipboard.md has the full record. Follow-ups: bd-yspyic32 (mixed payloads), bd-myoj9kp5 (SVG posture).","comments":{"c-6pfeq4wv":{"id":"c-6pfeq4wv","author":"Carlos Scheidegger","created_at":"2026-08-27T14:05:51.409839Z","text":"Design exploration written: claude-notes/plans/2026-08-27-paste-image-clipboard.md. Covers clipboard format landscape (paste event vs async API vs Monaco hooks), SVG security analysis, and design options; recommends capture-phase DOM paste listener + raster-only allowlist + content-hash filenames (pasted-<hash8>.<ext>) for CRDT concurrency safety. Awaiting user iteration before execution."},"c-j45phvym":{"id":"c-j45phvym","author":"Carlos Scheidegger","created_at":"2026-08-27T18:07:38.366267Z","text":"Implemented and verified. Commits on branch braid/bd-706b0ixu-paste-image-clipboard: 630916d87 (design doc), e0e2d2bbc (implementation: fileUpload/pasteImages.ts + pasteImageHandler.ts + Editor.tsx capture-phase listener; 41 new unit tests), a32f8cdb1 (changelog + E2E record). Gates: tsc -b, eslint (no new findings), 1049 unit + 114 integration + 133 wasm tests, npm run build:all — all green. E2E in production bundle via local-prod: content-hash naming, dedup, SVG decline, mixed-payload pass-through, preview rendering all observed live (plan §6). Not yet pushed/merged — awaiting user review."},"c-m6lder6e":{"id":"c-m6lder6e","author":"Carlos Scheidegger","created_at":"2026-08-27T19:41:21.670639Z","text":"Merged to main as PR #623 (https://github.com/quarto-dev/q2/pull/623)."}}}
 {"id":"bd-70krno7a","title":"quarto-yaml: expose content spans for quoted scalars (diagnostics land one column early)","description":"Scalar spans in quarto-yaml deliberately include the surrounding quotes (parser.rs test_quoted_scalar_spans_cover_the_whole_scalar). Any diagnostic that composes 'offset into the scalar CONTENT' with that span — e.g. a re-parsed config-markdown string's image-target span (parse_config_string_as_markdown threading the scalar's SourceInfo; see bd-page-footer-image-items-stmpikgo Phase 4) — renders one column early for quoted scalars, since content starts one byte after the opening quote. Escaped scalars would be off by more (escape sequences make the content-to-source mapping non-linear); block scalars differ again. Robust fix: quarto-yaml exposes a content span (or a content-to-source offset map) alongside the whole-scalar span, and q2's substring-composition sites use it. Observed: footer Q-5-6 for a single-quoted page-footer scalar underlines starting at the paren before the URL instead of the URL's first character — right file, right line, one column off. Repro: claude-notes/plans/page-footer-image-items-investigation/repro-missing/ in q2.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-08-18T21:35:42.869070Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-18T21:35:42.869070Z","dependencies":{"bd-page-footer-image-items-stmpikgo:discovered-from":{"depends_on_id":"bd-page-footer-image-items-stmpikgo","type":"discovered-from","created_at":"2026-08-18T21:35:42.869070Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-72f43umg","title":"Migrate remaining destructive write paths onto validated OutputSink","description":"Phase 3 follow-up from bd-cfl67 (render truncates source images). That fix introduced `OutputSink` (crates/quarto-core/src/output_sink.rs), which validates every destination is absolute, lexically clean, and under `allowed_roots`, and skips Copy ops where src == dest. `write_artifacts`/`enqueue_artifacts`, `flush_site_libs`, `ResourceCollectorTransform`, and the pass-2 WASM renderers were migrated onto it. Three write paths were left unmigrated and were never filed as strands at the time (discovered 2026-07-27 during strand cleanup — the plan doc said 'separate beads issues' but none existed):\n\nA1 (MEDIUM) — Migrate `copy_resources_to_output_dir` onto the sink.\nA2 (LOW) — Migrate `copy_favicon` / `copy_robots_txt` onto the sink.\nA3 (LOW) — Migrate engine intermediate writes onto the sink. Requires a design decision: keep `intermediate_dir` as an allowed root, or lift it under `output_dir/.intermediate/`.\n\nWhy it matters: these are the remaining destructive-write paths that bypass the sink's allowed-roots validation — the same class of unguarded write that truncated user-authored images to 0 bytes in bd-cfl67. This is hardening, not a live bug: no known reproduction truncates source files today.\n\nPlan doc: claude-notes/plans/2026-05-20-render-truncates-source-images.md (Phase 3 checklist, items A1-A3).","status":"open","priority":2,"issue_type":"task","created_at":"2026-07-27T20:53:11.951702Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-27T20:53:11.951702Z","dependencies":{"bd-cfl67:discovered-from":{"depends_on_id":"bd-cfl67","type":"discovered-from","created_at":"2026-07-27T20:53:11.951702Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-7328beha","title":"revealjs logo: emitted raw, never relativized per page or copied (URL-space config path)","description":"crates/quarto-core/src/revealjs/footer_logo.rs:168-176 (logo_slot_html) emits the revealjs logo: string verbatim into <img src>: no page_url_for relativization, no copy intent, and a leading / lands in the HTML as a /-absolute link. A _quarto.yml logo seen from a subdirectory deck is wrong. Found by the 2026-10-28 consumer sweep (claude-notes/issue-reports/945/consumer-sweep-2026-10-28.md); scoped out of bd-lwvumpfk because it is a URL-space emitter needing the resource-resolver exit (ResourceCopyIntent + page_url_for), not a filesystem join. Contract: claude-notes/designs/path-resolution-model.md rule 2 URL space; same shape as the navbar logo (bd-root-relative-paths-design-fc5pvkcv case A).","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-10-28T15:48:09.127928Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T22:51:28.119116Z","closed_at":"2026-10-28T22:51:28.119116Z","close_reason":"Merged in PR #958 (dbd7cebd5): reveal.js logo resolved via config-path resolver into URL space with a copy intent; listing leading-/ image src anchored at the site root with quarto_util::is_external_url as the sole classifier.","labels":["websites"],"dependencies":{"bd-lwvumpfk:related":{"depends_on_id":"bd-lwvumpfk","type":"related","created_at":"2026-10-28T15:48:09.127928Z","created_by":"Carlos Scheidegger"},"bd-oejuizi9:related":{"depends_on_id":"bd-oejuizi9","type":"related","created_at":"2026-10-28T15:48:09.127928Z","created_by":"Carlos Scheidegger"}},"comments":{"c-a39oewfb":{"id":"c-a39oewfb","author":"Carlos Scheidegger","created_at":"2026-10-28T20:59:14.047796Z","text":"PR #958 (with bd-epk8cmdl): resolve_logo in revealjs/footer_logo.rs resolves through config_paths, emits page_url_for and registers a ResourceCopyIntent (skipped in VFS-root mode)."}}}
 {"id":"bd-732bn","title":"Author error-docs pages for theme subsystem (2 codes)","description":"Author stub-quality pages for all 2 theme subsystem error codes under docs/errors/theme/. Follow template in docs/errors/yaml/Q-1-10.qmd. See claude-notes/plans/2026-05-22-error-docs-content.md.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-05-24T21:07:57.831390Z","created_by":"cscheid","updated_at":"2026-05-24T21:15:22.969708Z","closed_at":"2026-05-24T21:15:22.969353Z"}
 {"id":"bd-738mbhpj","title":"hub-mcp Phase 4: Quarto-specific intelligence","description":"Plan §5 Phase 4 (claude-notes/plans/2026-10-05-elevate-quarto-hub-mcp.md). Spike S-1 (qmd parser in Node) gates CAP-11 get_outline + section selectors; CAP-12 render_project/render_file behind opt-in --allow-render; CAP-13 docs tool (absorbs bd-dn81ol95, bd-b6cocsxw); Q-4 clear_capture decision.","status":"closed","priority":2,"issue_type":"epic","created_at":"2026-10-06T08:45:00.155617Z","created_by":"shikokuchuo","updated_at":"2026-10-07T00:08:14.592887Z","closed_at":"2026-10-07T00:08:14.592887Z","close_reason":"Phase 4 complete — see plan completion record (claude-notes/plans/2026-10-05-elevate-quarto-hub-mcp.md) and the comment trail.","dependencies":{"bd-3qe7unp7:blocks":{"depends_on_id":"bd-3qe7unp7","type":"blocks","created_at":"2026-10-06T08:45:51.473525Z","created_by":"shikokuchuo"},"bd-tbuerrg7:parent-child":{"depends_on_id":"bd-tbuerrg7","type":"parent-child","created_at":"2026-10-06T08:45:33.298329Z","created_by":"shikokuchuo"}},"comments":{"c-o5trcrsb":{"id":"c-o5trcrsb","author":"shikokuchuo","created_at":"2026-10-07T00:08:11.512628Z","text":"Phase 4 landed in 9 commits on braid/bd-738mbhpj-hub-mcp-phase-4-quarto-intelligence (stacked on Phase 3). S-1 decided (a) wasm-qmd-parser --target nodejs staged keyring-style (4.6MB raw/1.24MB gz, 3.8ms init); build un-rotted (patch tree-sitter-language, wasm-c-shim, panic=abort last-wins). CAP-11: get_outline + section selectors (line-based splices, CRDT-merge safe). CAP-12: render behind --allow-render, QUARTO_Q2_PATH injection, 3 modes incl. per-file batch (Q-7-4). CAP-13: docs tool over q2 docs llms --full + --json seam (bd-dn81ol95, bd-b6cocsxw closed). Q-4: clear_capture human-only. Gate: cargo xtask verify 14/14, hub-mcp suite 498+3, eval 15/15 median 5 turns, real-binary e2e recorded in the plan. Tool budget at ceiling (24/24 with auth) — noted for Phase 5."}}}
 {"id":"bd-73lilznv","title":"Move sandboxed preview from quarto-dev.github.io/q2 to /q2-sandbox and free /q2 for docs","description":"quarto-dev.github.io/q2 currently serves the q2-sandboxed-preview iframe (deploy-sandboxed-preview.yml, Pages in GitHub Actions mode; the default URL is hardcoded in Q2SandboxedPreviewIframe.tsx:67 and baked into the live quarto-hub.com bundle). Move it to a new repo quarto-dev/q2-sandbox (https://quarto-dev.github.io/q2-sandbox/), repoint hub-client, retire the old workflow, and publish the Q2 docs at /q2/ with a self-unregistering serviceWorker.js so stale sandbox workers registered at scope /q2/ don't 502 docs pages. Plan: claude-notes/plans/2026-10-26-move-sandbox-free-q2-pages.md. Same-origin hardening (sandbox shares quarto-dev.github.io with docs) is out of scope, tracked separately.","status":"closed","priority":1,"issue_type":"task","assignee":"Carlos Scheidegger","created_at":"2026-10-26T19:53:46.089598Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T15:21:33.430725Z","closed_at":"2026-10-27T15:21:33.430725Z","external_ref":"https://github.com/quarto-dev/q2/pull/919","labels":["hub-client","infra"],"comments":{"c-6q1v8102":{"id":"c-6q1v8102","author":"Carlos Scheidegger","created_at":"2026-10-27T15:21:30.479492Z","text":"2026-10-27 Phase 4 done. PR #929 merged as b3302a1a2 (all checks green). deploy-docs.yml run 40526439632 on main: build+deploy success. Live https://quarto-dev.github.io/q2/ serves the Q2 docs (title 'Quarto 2'); about/guides/llms.txt/search.json/staged examples and all landing-page site_libs assets 200; /q2/serviceWorker.js is the kill switch; /q2-sandbox/ still 200. Follow-up worth filing: stage-doc-examples resolves repo_root via --git-common-dir, so from a worktree it stages into the main checkout and rewrites tracked examples/lmd/10-project/index.html."},"c-7t3grta4":{"id":"c-7t3grta4","author":"Carlos Scheidegger","created_at":"2026-10-26T20:12:47.151564Z","text":"Phase 1 GitHub setup + Phase 2 code done (commits be6300e0f + changelog) on braid/bd-73lilznv-move-sandboxed-preview-quarto in .worktrees/bd-73lilznv-move-sandboxed-preview-quarto. q2-sandbox repo, deploy key, Q2_SANDBOX_DEPLOY_KEY secret created; gh-pages seeded manually and verified live at https://quarto-dev.github.io/q2-sandbox/. Not pushed yet — awaiting user OK to push branch / open PR. See plan Progress log."},"c-cc9mq8u3":{"id":"c-cc9mq8u3","author":"Carlos Scheidegger","created_at":"2026-10-26T19:54:38.462432Z","text":"Plan written 2026-10-26: claude-notes/plans/2026-10-26-move-sandbox-free-q2-pages.md (uncommitted on main in room-3 checkout at time of writing). Awaiting user review before execution. Open decision: deploy mechanism for q2-sandbox (recommended: q2 workflow pushes dist to q2-sandbox gh-pages via deploy key)."},"c-vq0rpe64":{"id":"c-vq0rpe64","author":"Carlos Scheidegger","created_at":"2026-10-27T13:47:08.317268Z","text":"2026-10-27 Phase 4 committed locally on braid/bd-73lilznv-docs-pages (04b57670d): deleted deploy-sandboxed-preview.yml; added deploy-docs.yml (source-built q2, stage-doc-examples, render docs, deploy-pages; PRs build only); docs/serviceWorker.js kill switch verified locally end-to-end. Awaiting user go-ahead to push + open PR. Merging it triggers the first docs deploy to /q2/."},"c-w2rxfp1v":{"id":"c-w2rxfp1v","author":"Carlos Scheidegger","created_at":"2026-10-26T20:16:01.275576Z","text":"PR #919 opened (remote branch feature/bd-73lilznv-move-sandboxed-preview-quarto). After merge: confirm first deploy-q2-sandbox.yml run succeeds, then Phase 3."},"c-ynm825o1":{"id":"c-ynm825o1","author":"Carlos Scheidegger","created_at":"2026-10-26T22:09:27.003240Z","text":"2026-10-26 Phase 2+3 done. PR #919 merged as d1522ac09. First CI deploy-q2-sandbox.yml run 40407250561 succeeded (gh-pages commit 'Deploy quarto-dev/q2@d1522ac09'). Production: quarto-hub-deployment build 40407273253 at q2_ref=d1522ac09 (includes #917 samod 0.16 rewire, user-approved), deployed via 'sudo quarto-hub-deploy' at 22:07Z; version d1522ac09c5da84d6830b77ffca6c22309be4f6d-20261026-220610; previous was main-20261016-161148 (rollback target). Hub active, auth_ok/ws_connect flowing, no WARN/ERROR in first 3 min. Live main-*.js references quarto-dev.github.io/q2-sandbox/ (old /q2/ string only in embedded changelog text). Pending: in-browser smoke test of a format: q2-sandboxed-preview doc; then Phase 4 (delete deploy-sandboxed-preview.yml, docs deploy + SW kill switch). Plan Progress log to be updated in the Phase 4 PR."}}}
@@ -543,6 +546,7 @@
 {"id":"bd-9svl","title":"[websites phase 2] Sidebar: data model, generate, render, template","description":"Implement sidebar feature for website projects. Delivers:\n- Sidebar data model in quarto-navigation (Sidebar, SidebarEntry, SidebarContents enums)\n- SidebarGenerateTransform (format-agnostic; reads website.sidebar, resolves auto:, sets active/expanded)\n- SidebarRenderTransform (HTML-specific; rewrites .qmd -> .html hrefs via ProjectIndex)\n- New template slot $rendered.navigation.sidebar$ in full HTML template\n- Active-item highlighting by source-path equality (format-agnostic)\n- Sidebar-for-page selection (Q1 sidebarForHref equivalent by source-path containment)\n- DocumentProfile gains additive 'order: Option<i32>' field for auto-sort\n\nPlan: claude-notes/plans/2026-04-24-websites-phase-2.md\nParent epic: bd-0tr6 (websites epic)\nBlocked by: bd-w5os (phase 1, closed)\n\nFormat-agnostic Generate vs HTML-specific Render split is load-bearing:\nnavigation.sidebar keeps source paths so a future non-HTML per-page\nformat could reuse it with its own rewrite.\n\nDeferred to follow-ups: search, tools, logo/subtitle/header/footer slots,\ninteractive collapse JS (rides with Phase 5), draft-mode option.","status":"closed","priority":1,"issue_type":"task","created_at":"2026-04-24T17:11:48.884350Z","created_by":"cscheid","updated_at":"2026-04-24T17:57:26.598128Z","closed_at":"2026-04-24T17:57:26.597171Z","close_reason":"Phase 2 sidebar complete: data model in quarto-navigation, Generate+Render transforms (format-agnostic/HTML split), sidebar_for_page + active-state helpers, auto expansion, template slot, AstTransformsStage project_index bridge fix, text enrichment from ProjectIndex. 1063 quarto-core tests + 81 quarto-navigation tests + 5 pipeline integration tests + manual 3-page CLI smoke all pass. cargo xtask verify green. Follow-ups filed: bd-6cme (search), bd-fod3 (tools), bd-ht0n (logo/subtitle), bd-49ar (collapse JS), bd-w0o9 (draft-mode), bd-l6f0 (explicit expanded), bd-81x4 (multi-sidebar diagnostic), bd-tfy0 (deep-dir grouping), bd-2quy (bridge audit). Epic follow-ups: bd-n9dr (nav config unification), bd-4g6g (sidebar-left template).","dependencies":{"bd-0tr6:parent-child":{"depends_on_id":"bd-0tr6","type":"parent-child","created_at":"2026-04-24T17:12:21.865157Z","created_by":"cscheid"},"bd-w5os:blocks":{"depends_on_id":"bd-w5os","type":"blocks","created_at":"2026-04-24T17:11:48.884350Z","created_by":"cscheid"}}}
 {"id":"bd-9t5nmq81","title":"hephaestus: .hep plots in the separate-origin sandboxed preview frame (format: q2-sandboxed-preview)","description":"Phase 2 (bd-sxiv2tio) ships .hep rendering for q2-preview and revealjs through the same-origin Q2PreviewIframe (q2 preview SPA + hub-client). The separate-origin sandboxed frame (hub-client/quarto-hub-sandboxed-preview, its own non-workspace npm project) is deferred: it needs hephaestus-svg-wasm added to its own package.json, 'hep' added to isBinaryPath in src/assetPolicy.ts so the SW proxy ships bytes as base64, and a cross-origin check that the wasm + Roboto assets load from q2-preview-assets/. Plan: claude-notes/plans/2026-09-17-hephaestus-hep-integration.md (Phase 2, deferred).","status":"open","priority":3,"issue_type":"feature","created_at":"2026-09-17T19:44:32.226934Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T19:44:32.226934Z","labels":["hephaestus"],"dependencies":{"bd-sxiv2tio:discovered-from":{"depends_on_id":"bd-sxiv2tio","type":"discovered-from","created_at":"2026-09-17T19:44:32.226934Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-9t8zsh98","title":"Hub mints different author IDs for bare vs automerge:-prefixed project param","description":"GET /auth/author (and /auth/actor) HMAC the project query param verbatim. hub-mcp always passes the bare doc id (parseProjectRef), the SPA passes the automerge:-prefixed form for share-joined projects (App.tsx:391) and the bare form at createNewProject (client.ts:2229). Same sub + same doc therefore mints two unrelated author IDs; MCP edits show as an unknown author. Fix: canonicalize at the mint (strip the automerge: prefix) in both endpoints identically (D5). Found while reviewing MCP attribution on braid/bd-1ilystml-hub-mcp-phase-7-tool-consolidation.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-10-13T16:45:48.338049Z","created_by":"shikokuchuo","updated_at":"2026-10-15T10:01:25.900521Z","closed_at":"2026-10-15T10:01:25.900521Z","close_reason":"Superseded by bd-evxckg64. Its MCP-side prefix fix shipped and was correct for share-joined projects, but its canonical-form premise was wrong for SPA-created projects (see comment c-ien0j4vg). The server-side strip in bd-evxckg64 makes any client spelling mint the bare-canonical value, and the MCP prefixing has been removed (commit b005e71c3).","comments":{"c-bbsk750d":{"id":"c-bbsk750d","author":"shikokuchuo","created_at":"2026-10-13T17:47:32.194374Z","text":"Resolution changed per maintainer direction: NOT hub-side normalization (reverted). Canonical project spelling stays the automerge:-prefixed URL form the SPA uses; hub-mcp's fetchAuthorId now prefixes bare ids before calling /auth/author (+ the /auth/actor 404 fallback), so the same sub+document mints the SPA's author ID. Tests: connection-manager.test.ts author-ID fetch block asserts the prefixed query (incl. no-double-prefix passthrough)."},"c-ien0j4vg":{"id":"c-ien0j4vg","author":"shikokuchuo","created_at":"2026-10-15T09:04:54.290787Z","text":"Superseded — recording why the 'canonical = automerge:-prefixed (the SPA form)' assumption was wrong. (1) The SPA is NOT uniformly prefixed: createNewProject resolves the author over the BARE id (quarto-sync-client client.ts:2229-2234, parseAutomergeUrl strips the scheme) and returns/stores it verbatim into IDB (projectStorage.addProject, no normalization) and the project set (addProjectToSet stores entry.indexDocId verbatim; only the map key is stripped). Every SPA-created project therefore resolves bare on every subsequent open; only share-joined projects are prefixed (App.tsx:392 normalization). (2) The author-ID transition's E2E sign-off pinned BARE byte-for-byte as the hub-mcp<->hub-client contract: claude-notes/plans/2026-09-30-automerge-author-id-transition.md lines 685-687 ('both pass the bare id'). (3) Prefixing MCP thus fixed MCP vs share-joined-SPA but left MCP vs SPA-creator mismatched — the common case, since every project's creation history is stamped by its creator. New resolution: canonical = BARE; strip a leading 'automerge:' server-side in sub_to_actor_id_for_project (one point covers /auth/author and /auth/actor, D5 preserved); zero SPA changes; MCP's prefixing is dropped after the server fix ships. This was the strand's original proposed fix — the revert rested on the incomplete premise that the SPA consistently uses the prefixed form."}}}
+{"id":"bd-9u10sb4b","title":"Lua API: os.exit(1) in a filter kills the whole q2 process instead of failing the document (GH #941)","description":"Quarto 1 filters call os.exit(1) to fail a render. In q2 the native Lua state is Lua::new() with the full stdlib, so os.exit is libc exit: the whole process dies, no output is written for any page (including already-rendered ones), there is no render summary, and parse errors on pages not yet reached are never reported. Reproduced 2026-10-27 on main at 3f7d4b0e8 with the issue's a/b/c website: exit 1, empty _site/, c.qmd's Q-2-12 never shown.\n\nFix: replace os.exit on both the native stdlib os table and the WASM synthetic one (os_wasm.rs, shared helper) with a Rust function raising an external LuaExit { code } error (Lua's code mapping: none/true -> 0, false -> 1, n -> n). apply_lua_filter_with maps it to a new LuaFilterError::Exit { code, filter } with the message 'Lua filter <path> called os.exit(<code>); the document was not rendered'. Downstream already fails just that document, continues, prints the summary, and exits non-zero. os.exit(0) also fails the document (Q1 parity: pandoc produced no output). error() is not changed. Update crates/pampa/tests/wasm_lua.rs per .claude/rules/wasm.md. CLI e2e test at crates/quarto/tests/integration/lua_os_exit_cli.rs.\n\nPlan: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md","status":"in_progress","priority":1,"issue_type":"bug","created_at":"2026-10-27T20:22:18.287994Z","created_by":"claude","updated_at":"2026-10-27T21:54:44.345532Z","external_ref":"https://github.com/quarto-dev/q2/issues/941","labels":["lua","pampa"],"dependencies":{"bd-pqj37chw:parent-child":{"depends_on_id":"bd-pqj37chw","type":"parent-child","created_at":"2026-10-27T20:22:18.287994Z","created_by":"claude"}},"comments":{"c-esm27o8l":{"id":"c-esm27o8l","author":"claude","created_at":"2026-10-27T21:35:43.212081Z","text":"Implemented and merged into feature/lua-api-q1-parity: crates/pampa/src/lua/os_exit.rs (LuaExit external error, install_os_exit, requested_exit chain walk), LuaFilterError::Exit in filter.rs, same replacement in the shortcode env, WASM os test extended. CLI e2e: crates/quarto/tests/integration/lua_os_exit_cli.rs. GH #941 repro now renders b.html, reports c.qmd's Q-2-12 and A's os.exit(1), prints 'Rendered 1 of 3 files', exits non-zero. Left open until the epic's PR merges."},"c-jqln7y93":{"id":"c-jqln7y93","author":"claude","created_at":"2026-10-27T21:54:44.345532Z","text":"PR: https://github.com/quarto-dev/q2/pull/949 (feature/lua-api-q1-parity -> main; closes GH #939-#942 on merge). Full cargo xtask verify green incl. hub-client tests on a rebuilt WASM; Windows cross-check not possible from macOS, flagged in the PR."}}}
 {"id":"bd-9uqdoy0e","title":"lua: require is unavailable in Lua filters, but works in shortcodes (GH #587)","description":"GH https://github.com/quarto-dev/q2/issues/587. create_filter_environment (crates/pampa/src/lua/filter.rs) never installs register_scoped_require, so extension Lua filters run with the stock Lua require and an untouched package.path; require('_modules/greet') fails in a filter but works in a shortcode of the same extension. Fix: install the scoped require in the filter environment (runtime Arc already in scope) — AFTER bd-sr0nipl7's stack split lands, so filters inherit the corrected require rather than the resolve_path breakage. Plan: claude-notes/plans/2026-08-24-lua-require-current-path.md; assessment: claude-notes/research/2026-08-24-lua-current-path-gh588.md","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-08-24T22:08:38.501848Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-25T15:26:48.234671Z","closed_at":"2026-08-25T15:26:48.234671Z","close_reason":"Fixed in PR #597 (merged): create_filter_environment installs the scoped require, so extension Lua filters load sibling modules (root-relative and absolute resolve_path forms). Pinned by the contract-filter-require fixture.","dependencies":{"bd-sr0nipl7:blocks":{"depends_on_id":"bd-sr0nipl7","type":"blocks","created_at":"2026-08-24T22:08:45.800587Z","created_by":"Carlos Scheidegger"},"bd-sr0nipl7:related":{"depends_on_id":"bd-sr0nipl7","type":"related","created_at":"2026-08-24T22:08:47.798202Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-9vmlk2md","title":"Three warnings have no Q code: Missing shortcode argument, Shortcode error, YAML string scalar has no content provenance","description":"q2 render --json-errors emits these as kind:warning records with no 'code' field, and the plain output prints them as 'Warning: <title>' without the [Q-x-y] tag. Tooling that groups diagnostics by code (and users grepping for codes) misses them: the claude-notes bracket/shortcode passes skipped ten 'Missing shortcode argument' warnings for exactly this reason. 'Missing shortcode argument' and 'Shortcode error' are user-facing and belong next to Q-16-3/Q-16-5 (shortcode subsystem); the provenance one is an internal self-check (convert.rs:202) and may want a code in the internal range or to be demoted to a debug log once the desync bug is fixed.","status":"open","priority":3,"issue_type":"chore","created_at":"2026-10-08T20:05:37.788262Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-08T20:05:37.788262Z","dependencies":{"bd-uk8zgkha:discovered-from":{"depends_on_id":"bd-uk8zgkha","type":"discovered-from","created_at":"2026-10-08T20:05:37.788262Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-9x3zbuj8","title":"q2 preview: bring hierarchical block navigator to --allow-edit; fix breadcrumb/rich-text toolbar overlap","description":"Two tasks for the q2-preview live block editor (tiptap rich text).\n\nTask 1: The hierarchical block navigator (BreadcrumbChip, ◀ Dv ¶ ▶) currently appears in hub-client/quarto-hub.com but NOT in 'q2 preview --allow-edit'. It is gated solely on PreviewContext.unlockNestingCursor. hub-client defaults that preference ON (schema.ts default true); the q2-preview SPA defaults it OFF and only enables via ?nestingCursor=1 (parseNestingCursorParam in PreviewApp.tsx). Bring it to q2 preview by defaulting nestingCursor ON in the SPA, mirroring the richText default-on / ?richText=0 opt-out pattern. The breadcrumb self-gates on editTarget, so it only appears when actually editing (i.e. under --allow-edit).\n\nTask 2: When both are present (a nested rich-text block), BreadcrumbChip and RichTextToolbar overlap at the block's top-left. Toolbar is abs-positioned at left:-2px/bottom:100% of the edit box; breadcrumb pins its band right-edge at surfaceLeft with ▶ + future-placeholder spilling right past the pivot, over the toolbar. Fix: lay them side by side with the breadcrumb to the RIGHT of the fixed-position rich-text controls (keeping toolbar position independent of the variable-width breadcrumb).\n\nPlan: claude-notes/plans/2026-06-25-q2-preview-hierarchy-nav-and-overlap.md","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-06-25T13:36:39.891405Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-27T19:22:57.657810Z","closed_at":"2026-07-27T19:22:57.657810Z","close_reason":"Landed: PR #345 merged 2026-06-25 (hierarchy navigator default-on + inline breadcrumb). Audit bd-a0eyjshu.","comments":{"c-3vyk86he":{"id":"c-3vyk86he","author":"Carlos Scheidegger","created_at":"2026-06-25T20:29:19.724271Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/345 (feature/bd-9x3zbuj8-preview-hierarchy-nav-overlap → main). cargo xtask verify --skip-hub-build passed before push."},"c-6qp6wjmx":{"id":"c-6qp6wjmx","author":"Carlos Scheidegger","created_at":"2026-06-25T13:38:06.376341Z","text":"Plan written: claude-notes/plans/2026-06-25-q2-preview-hierarchy-nav-and-overlap.md. Task 1 (navigator in q2 preview) is a one-line default flip of parseNestingCursorParam mirroring richText. Task 2 (overlap fix) has a real design fork — Option A (reposition standalone chip right of toolbar, drop left-spill geometry) vs Option B (render breadcrumb inline in toolbar flex row). Awaiting user review before implementing."},"c-oaia8fu3":{"id":"c-oaia8fu3","author":"Carlos Scheidegger","created_at":"2026-06-25T13:41:37.291932Z","text":"Design fork settled: Option B (inline breadcrumb in toolbar flex row for rich blocks; standalone chip unchanged for plain blocks). Key win: standalone left-spill geometry + its big test suite are preserved. Plan updated accordingly. Still awaiting go-ahead to implement."},"c-udmh342h":{"id":"c-udmh342h","author":"Carlos Scheidegger","created_at":"2026-06-25T13:48:01.694191Z","text":"Task 1 implemented (TDD). parseNestingCursorParam flipped to default-on (?nestingCursor=0 opts out), exported. New unit test parseNestingCursorParam.test.ts (RED→GREEN). Updated p3-2-nesting-cursor-spa.integration.test.tsx (no-param now expects unlock=true). Rewrote q2-preview-spa/e2e/nesting-cursor.spec.ts: default boot → unlocked + chip visible; ?nestingCursor=0 → locked. All SPA unit (37) + integration (75) green. Real-binary e2e deferred to combined verification after Task 2. Next: Task 2."},"c-y2lh9z19":{"id":"c-y2lh9z19","author":"Carlos Scheidegger","created_at":"2026-06-25T14:13:46.447801Z","text":"Both tasks implemented + verified end-to-end. Task 1: nesting navigator default-on in q2 preview (parseNestingCursorParam flipped). Task 2: breadcrumb renders inline in the rich-text toolbar (BreadcrumbCrumbs extracted; standalone chip suppressed for rich targets via richTextSupport.ts). Verified via real q2 binary + Chrome (toolbar = 'B I S x2 x2 🔗 ◀ Dv ¶ ▶', standalone chip absent, no overlap) and screenshot. Tests green: preview-renderer 500+515, q2-preview-spa 37+75 + 37 e2e, hub-client 662+76, both production builds. Discovered + filed bd-fpys25b0 (systemic hub-client e2e textarea-baseline breakage from bd-j1nto6eq, out of scope); fixed bd-n4v4phe4 (edit-cell-sizing pin). Committed on branch braid/bd-9x3zbuj8-preview-hierarchy-nav-overlap (5 commits). Awaiting review + push approval; cargo xtask verify pending before push."}}}
@@ -581,7 +585,7 @@
 {"id":"bd-agsgrbfn","title":"cargo xtask verify: tree-sitter test reuses a stale shared ~/.cache/tree-sitter/lib/markdown.dylib across checkouts","description":"tree-sitter CLI caches the compiled grammar at $TREE_SITTER_LIBDIR/<name>.dylib, defaulting to ~/.cache/tree-sitter/lib when the env var is unset (a tree-sitter loader default; q2 configures nothing). The key is only the grammar *name* ('markdown'), and freshness is 'lib mtime newer than src/parser.c / scanner.c'. So every q2 checkout on a machine (rooms/room-*/q2 and every .worktrees/*) shares ONE markdown.dylib, and whichever checkout compiled most recently wins.\n\nObserved 2026-09-25 (room-5, bd-angle-bracket-u27e8-parse-error-r6l55zmh): room-5 regenerated parser.c at 16:52:31. room-2 regenerated its own parser.c at 16:56:37 and ran tree-sitter test, which rewrote ~/.cache/tree-sitter/lib/markdown.dylib at 16:56:37. room-5's verify step 4 then saw a lib newer than its parser.c, reused room-2's grammar, and failed 9 new corpus tests with nonsense trees ((shortcode_name) in plain prose). Even 'tree-sitter test --rebuild' is only a momentary fix: the next foreign run overwrites it again.\n\nConfirmed fix: TREE_SITTER_LIBDIR=<checkout>/target/tree-sitter-lib makes verify step 4 and the CRLF parity run pass 732/732.\n\nProposed implementation:\n- xtask sets TREE_SITTER_LIBDIR=<project_root>/target/tree-sitter-lib on every tree-sitter invocation (verify.rs step 4, treesitter_crlf.rs, and any other xtask that shells out to tree-sitter). target/ is already per-checkout and gitignored, and cargo clean wipes it.\n- Agents and humans also run 'tree-sitter generate; tree-sitter build; tree-sitter test' by hand (AGENTS.md:647). Cover those with a checked-in env mechanism (e.g. .envrc for direnv, or a repo-local wrapper 'cargo xtask ts-test'), or at least document exporting TREE_SITTER_LIBDIR in AGENTS.md.\n- Also applies to the doctemplate grammar (doctemplate.dylib).","status":"closed","priority":3,"issue_type":"bug","created_at":"2026-09-25T21:56:08.535866Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-28T15:18:09.440757Z","closed_at":"2026-09-28T15:18:09.440757Z","close_reason":"Merged in PR #739 (fd0ec32f1): xtask pins TREE_SITTER_LIBDIR to <root>/target/tree-sitter-lib; CRLF run uses a private libdir; new cargo xtask ts-test for hand runs. Doctemplate-in-verify follow-up: bd-0njf3lab.","dependencies":{"bd-angle-bracket-u27e8-parse-error-r6l55zmh:discovered-from":{"depends_on_id":"bd-angle-bracket-u27e8-parse-error-r6l55zmh","type":"discovered-from","created_at":"2026-09-25T21:56:08.535866Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-aim2gqis","title":"hub-client: in-context debugging/diagnostic affordances for the editor SPA","description":"The /debug.html Automerge debugger runs in a separate browsing context with its own ephemeral Repo, so it cannot observe the live in-memory state of the editor SPA (sync-client Repo, DocHandle states, presence, VFS, render pipeline). Design and build debugging/diagnostic affordances that run in the SAME JS context as the editor: (a) machine-readable window.quartoDebug extensions for the Automerge layer, aimed at both humans and LLM agents driving the browser; (b) an in-context inspector UI (lazy-loaded panel/iframe) reusing the src/debug displays against the live repo. Plan doc: claude-notes/plans/2026-07-29-hub-client-in-context-debugging.md (design iteration in progress; no implementation go-ahead yet). Related prior work: bd-2rv8 (window.quartoDebug), claude-notes/plans/2026-04-16-hub-client-automerge-debugger.md (/debug.html).","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-07-29T19:09:54.320337Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-30T19:01:42.628363Z","closed_at":"2026-07-30T18:18:37.137965Z","close_reason":"All four phases shipped and verified: quartoDebug.am machine-readable Automerge API + doctor() + message tap + live inspector panel + embedded server-view debugger. Plan doc claude-notes/plans/2026-07-29-hub-client-in-context-debugging.md holds the full design record and per-phase e2e transcripts. Branch chain ends at braid/bd-09aja9gl-phase-4-iframe-embed.","dependencies":{"bd-2rv8:related":{"depends_on_id":"bd-2rv8","type":"related","created_at":"2026-07-29T19:11:26.236256Z","created_by":"Carlos Scheidegger"}},"comments":{"c-8etbc3c3":{"id":"c-8etbc3c3","author":"Carlos Scheidegger","created_at":"2026-07-30T17:20:26.623313Z","text":"Cross-cutting decisions settled with Carlos (2026-07-30), recorded in the plan doc: single quartoDebug gate shipping in prod behind the flag; quartoDebug.am.unsafe namespace (handle(ref) + Automerge module) as console escape hatch, observation-only rule stands; quartoDebug.help() as the human+agent runtime contract (tested against the API surface) + apiVersion; message tap defaults to summaries with {capture:'full'} params opt-in. Remaining before go-ahead: confirm phasing A→D→B→C and cut per-phase child strands."},"c-f5s6d1bc":{"id":"c-f5s6d1bc","author":"Carlos Scheidegger","created_at":"2026-07-29T19:11:24.880915Z","text":"Source survey done; design plan written to claude-notes/plans/2026-07-29-hub-client-in-context-debugging.md. Four building blocks identified: (A) quartoDebug.am machine-readable Automerge API incl. doctor(); (B) lazy in-context inspector panel reusing src/debug components via RepoContext against the live sync-client Repo; (C) iframe embed of /debug.html for live-vs-server comparison; (D) ring-buffered sync-message tap in quarto-sync-client. Recommended order A → D → B → C. Awaiting design iteration with Carlos before any implementation."},"c-fiuxovhi":{"id":"c-fiuxovhi","author":"Carlos Scheidegger","created_at":"2026-07-30T19:01:42.628363Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/444 (single PR covering all four phases, branch feature/bd-aim2gqis-in-context-debugging)."}}}
 {"id":"bd-ak4cnm44","title":"Release q2 v0.12.0","description":"Cut the v0.12.0 binary release per claude-notes/instructions/release-runbook.md. Scope: origin/main as of 0087da18 (release main as-is; no open PRs pulled in). Contents since v0.11.0: glob-semantics migration (#461), Alpine no-glibc release assertion (#458), listing-glob resolution against declaring file's dir (#460), AGENTS.md, ephemeral hub secrets preview fix (#462), listing diagnostics span fix (#459).","status":"closed","priority":1,"issue_type":"task","created_at":"2026-08-07T17:44:51.928528Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-07T18:36:37.063583Z","closed_at":"2026-08-07T18:36:37.063583Z","close_reason":"v0.12.0 released and verified end-to-end","comments":{"c-r7t8k52b":{"id":"c-r7t8k52b","author":"Carlos Scheidegger","created_at":"2026-08-07T18:36:35.743923Z","text":"Release published: https://github.com/quarto-dev/q2/releases/tag/v0.12.0 (run 31205651943, all jobs success; 18 assets = 5 platforms + checksums.sha256 + standalone hub-mcp bundle).\n\nPost-publish verification (runbook step 6), all against the real released artifact:\n- install.sh one-liner resolved 'latest release: v0.12.0', checksum verified, minisign signature verified, binary prints 'q2 (quarto 2) 0.12.0'.\n- q2 mcp --launcher-info with all three hub env vars unset: QUARTO_HUB_MCP_CLIENT_ID / _SECRET / QUARTO_HUB_SERVER all 'bundled'; bundle-files 16, bundle-hash 0ceae53039f3cde5, zero 'placeholder' markers; build-info gitCommit e8842518 (== tagged commit) gitDirty false, keyring addons darwin-arm64 + darwin-x64.\n- bootstrap_sh::resolves_latest_version_from_github (network, normally #[ignore]d) passes against the live release.\n- End-to-end render with the downloaded binary: 'q2 render smoke.qmd' emitted <meta name=\"generator\" content=\"quarto-rust-0.12.0\"> and correct <h2>Hello</h2>; output inspected."},"c-toeni2ys":{"id":"c-toeni2ys","author":"Carlos Scheidegger","created_at":"2026-08-07T18:09:35.948803Z","text":"Bump PR #463 merged (e8842518, main now version 0.12.0; all 8 PR checks green, main's own CI green at 0087da18). Local gate before merge: cargo xtask verify --skip-hub-build clean on lints/clippy/fmt; cargo nextest run --workspace --no-fail-fast = 10997 passed / 0 failed / 197 skipped. One transient failure on the first verify run was the known macOS-APFS case flake bd-eb2wnxkp (doc id 3qAsj... vs 3QAsj...), not a bump regression; passed on re-run. Tag v0.12.0 pushed at 6456f2a2 -> Release workflow firing."}}}
-{"id":"bd-aliases-redirects-missing-sch7cd1g","title":"aliases: is silently ignored — no redirect stubs written for renamed/moved pages","description":"q2 silently ignores the 'aliases:' document metadata key. No redirect stubs are written and no diagnostic is emitted, so a site that moved or renamed pages serves 404s at every old URL after a port from Quarto 1.\n\n## What happens vs Quarto 1\n\nGiven a page declaring:\n\n    ---\n    title: Current page\n    aliases:\n      - /old-name.html\n      - ../previous/index.html\n    ---\n\nQuarto 1 writes a small HTML redirect stub at each alias path (_site/old-name.html, _site/previous/index.html). The stub is a <title>Redirect</title> document whose inline script does window.location.replace to the canonical page, preserving both hash and query string:\n\n    var redirects = {\"\":\"../index.html\"};\n    var hash = window.location.hash.startsWith('#') ? window.location.hash.slice(1) : window.location.hash;\n    var redirect = redirects[hash] || redirects[\"\"] || \"/\";\n    window.document.title = 'Redirect to  ' +  redirect;\n    if (!redirects[hash]) { redirect = redirect + window.location.hash; }\n    redirect = redirect + window.location.search;\n    window.location.replace(redirect);\n\nq2 writes only the canonical pages. The key is dropped with no warning, which is the part that makes this expensive: a porting project gets no signal at all that its redirects are gone.\n\n## Root-cause area\n\nNothing in crates/ reads 'aliases' — verified at HEAD (1ba0f2ec, the v0.19.0 release merge); every grep hit for the word is an unrelated use (SASS theme aliases, YAML anchor aliases, path aliases). This is unimplemented, not broken.\n\nTwo pieces are needed:\n\n1. DocumentProfile (crates/quarto-core/src/document_profile.rs:351) has no aliases field. Add 'pub aliases: Vec<String>' and extract it next to the sibling frontmatter keys around line 695 — extract_string_list(meta, \"aliases\") is exactly the shape categories: already uses. This bumps DOCUMENT_PROFILE_VERSION (currently 9, line 83).\n\n2. crates/quarto-core/src/project/website_post_render.rs is the natural home for the writer. It already owns exactly this class of work — per-project post-render side files, walking index.profiles(): the favicon copier (line ~66) and the sitemap.xml emitter (line ~150), the latter already iterating profiles and mapping them to output paths.\n\n## Suggested fix direction\n\nAdd a write_alias_redirects pass in website_post_render.rs alongside the sitemap writer: for each profile with a non-empty aliases list, resolve each alias against the output directory (absolute aliases are site-root-relative; relative ones resolve against the page's own output location), compute the relative href from the stub back to the page's output_href, and write the Q1 stub template. Worth deciding explicitly: whether to warn or error when two pages claim the same alias, or when an alias collides with a real output path — Q1 lets the stub overwrite, which on a case-insensitive filesystem produces a genuinely confusing result we have already hit in the Connect docs.\n\n## Real-world hit in the Connect docs\n\n69 source files declare aliases:. The Quarto 1 reference site contains 99 redirect stub HTML files that the q2 render omits entirely — that is the whole 451-vs-352 HTML file-count gap between the two rendered sites. Every external link, bookmark and search-engine result pointing at a pre-reorganization Connect docs URL would 404. This is the single largest remaining structural difference between the Q1 and q2 renders of that project; content-wise the port is otherwise at 352 of 352 pages with zero errors.\n\n## Repro\n\n/Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/aliases-redirects-missing/\n\nRun 'q2 render' there: 2 HTML files are written (index.html and current/index.html) where Quarto 1 writes 4, the two extra being the redirect stubs for the aliases declared in current/index.qmd. Compare with 'quarto render'.\n\nObserved continuously from q2 0.15.0 through 0.19.0. Origin strand in the q2-connect-docs skein: br-aliases-redirects-8gk18exz.","status":"in_progress","priority":2,"issue_type":"feature","created_at":"2026-08-12T14:57:17.674476Z","created_by":"Claude (q2-connect-docs)","updated_at":"2026-08-12T19:38:27.852163Z","labels":["website"],"dependencies":{"bd-hzwecpyk:related":{"depends_on_id":"bd-hzwecpyk","type":"related","created_at":"2026-08-12T15:08:37.250245Z","created_by":"Carlos Scheidegger"},"bd-wch2dotq:related":{"depends_on_id":"bd-wch2dotq","type":"related","created_at":"2026-08-12T15:08:38.787787Z","created_by":"Carlos Scheidegger"}},"comments":{"c-6wpn8y9p":{"id":"c-6wpn8y9p","author":"Carlos Scheidegger","created_at":"2026-08-12T18:48:49.512310Z","text":"Design decisions settled with the user 2026-08-12 (commit 672cb26f, plan § Design decisions):\n\n1. bd-hzwecpyk closed as duplicate; this is the implementation strand.\n2. Collisions are HARD ERRORS, diverging from Q1's warn-and-skip. Three conditions: stub-vs-rendered-page, same alias + same fragment key from two pages, and case-only collisions. New codes Q-5-23+ in the 'project' subsystem.\n3. Case-only collisions error on EVERY platform, not just case-insensitive filesystems, so Linux CI fails the same way macOS does.\n4. OPEN: stub template. Two candidates written out under the investigation dir. Byte-parity with Q1 explicitly not required; recommendation is the improved stub (noscript meta-refresh fallback, rel=canonical, charset, DOCTYPE, body link).\n5. 'aliases:' in a non-website project warns rather than staying silent.\n6. Hash fragments implemented properly in the first cut.\n7. draft: true pages emit no stubs — leaking a draft's existence via a live redirect is worse than over-hiding it. Cross-linked to bd-4zdf so the sitemap resolves the same way.\n8. Stale-stub cleanup deferred to bd-wdhhl0t9.\n\nConsequence worth flagging: decision 2 makes this a breaking change for existing Q1 sites. A project whose alias collisions Q1 tolerated will now fail to render. Phase 6 must check whether the Connect docs' 106 aliases trip it."},"c-dgw6tdmx":{"id":"c-dgw6tdmx","author":"Carlos Scheidegger","created_at":"2026-08-12T19:38:27.852163Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/515 (branch braid/aliases-redirect-stubs -> origin/feature/aliases-redirect-stubs). Awaiting CI + review."},"c-l7unwmt5":{"id":"c-l7unwmt5","author":"Carlos Scheidegger","created_at":"2026-08-12T19:35:45.174273Z","text":"IMPLEMENTED on branch braid/aliases-redirect-stubs (commit 7c30181d, unpushed). cargo xtask verify --skip-hub-build green; 11793 workspace tests pass.\n\nResult: the Connect docs now render 451 HTML files against Q1's 451 — the gap is closed exactly, with all 99 stub paths identical to Q1's.\n\nMAJOR FINDING: the first render of the unmodified Connect docs FAILED with two Q-5-25 case-only-collision errors, and both are real bugs that Q1 has been silently shipping. cookbook/content/integrations/{msgraph,sharepoint}/viewer/r/index.qmd each declare an alias '/…/viewer/R/' (capital R) while rendering to '…/viewer/r/index.html'. On macOS those are one file. Q1's guard is an exact string compare, so it writes the stub on top of the rendered page. Verified in docs-quarto-1/_site: both of those files are redirect stubs whose target is '../r/index.html' — pointing at themselves. Two documentation pages are destroyed in the published Q1 site, replaced by infinite redirect loops.\n\nSo the counts reconcile as Q1 = 350 real pages + 101 stubs, q2 = 352 real pages + 99 stubs. Same total; q2 has the two pages Q1 ate.\n\nACTION NEEDED IN THE PORT REPO (not here): those two aliases must be removed before docs-quarto-2 renders under q2. The 451-vs-451 measurement was taken on a scratch copy with them removed, so the port repo was never modified.\n\nAlso fixed two things found on the way:\n- config_sources::rebase_source_candidates — a collision diagnostic spans two documents and both root front matter at the dense FileId(0), so the existing binder would have rendered the second document's offsets against the first document's text (the bd-m6wmztln mis-pairing, reached from the other direction).\n- orchestrator no longer stringifies post_render errors that already carry diagnostics.\n\nRemaining: user review, then push + PR. Follow-up bd-wdhhl0t9 (stale stubs) still open."},"c-o0vzdywr":{"id":"c-o0vzdywr","author":"Carlos Scheidegger","created_at":"2026-08-12T15:08:21.190507Z","text":"Investigated on main @ db075676. Plan skeleton: claude-notes/plans/2026-08-12-aliases-redirect-stubs.md; artifacts under claude-notes/plans/aliases-redirect-stubs-investigation/.\n\nVerdict: ready to design, pending 7 policy questions in the plan.\n\nCorrections to this strand's description:\n- Q1 does NOT let an alias stub overwrite a real output. It warns and skips (website-aliases.ts, 'would overwrite the target. Skipping.'). What Q1 misses is the case-insensitive-filesystem variant, since its guard is an exact string compare.\n- Duplicate of bd-hzwecpyk (cderv, 2026-06-23), which already carries the 'blocks bd-0tr6' edge. Dedup is the user's call.\n\nNew findings:\n- q2 needs none of Q1's ~60-line incremental workaround: Pass 1 profiles every file regardless of RenderMode, so index.profiles() always carries the full alias set in post_render.\n- Measured the Connect-docs corpus: 106 unique alias entries across 69 files. 76 site-root-relative, 30 page-relative, 77 trailing-slash, 18 .html, 5 with #fragments. Every Q1 fixupHref branch is exercised.\n- Hash fragments are NOT deferrable: cookbook/off-host-execution/creating-execution-environments/index.qmd and cookbook/operations/deploying-content/index.qmd both claim /cookbook/custom-execution-environments/index.html with different fragments pointing at different targets. A fragment-less first cut would emit wrong redirects, not just incomplete ones.\n- Ordering hazard: post_render runs before the resource-copy pass, so a declared resource could silently clobber a stub."}}}
+{"id":"bd-aliases-redirects-missing-sch7cd1g","title":"aliases: is silently ignored — no redirect stubs written for renamed/moved pages","description":"q2 silently ignores the 'aliases:' document metadata key. No redirect stubs are written and no diagnostic is emitted, so a site that moved or renamed pages serves 404s at every old URL after a port from Quarto 1.\n\n## What happens vs Quarto 1\n\nGiven a page declaring:\n\n    ---\n    title: Current page\n    aliases:\n      - /old-name.html\n      - ../previous/index.html\n    ---\n\nQuarto 1 writes a small HTML redirect stub at each alias path (_site/old-name.html, _site/previous/index.html). The stub is a <title>Redirect</title> document whose inline script does window.location.replace to the canonical page, preserving both hash and query string:\n\n    var redirects = {\"\":\"../index.html\"};\n    var hash = window.location.hash.startsWith('#') ? window.location.hash.slice(1) : window.location.hash;\n    var redirect = redirects[hash] || redirects[\"\"] || \"/\";\n    window.document.title = 'Redirect to  ' +  redirect;\n    if (!redirects[hash]) { redirect = redirect + window.location.hash; }\n    redirect = redirect + window.location.search;\n    window.location.replace(redirect);\n\nq2 writes only the canonical pages. The key is dropped with no warning, which is the part that makes this expensive: a porting project gets no signal at all that its redirects are gone.\n\n## Root-cause area\n\nNothing in crates/ reads 'aliases' — verified at HEAD (1ba0f2ec, the v0.19.0 release merge); every grep hit for the word is an unrelated use (SASS theme aliases, YAML anchor aliases, path aliases). This is unimplemented, not broken.\n\nTwo pieces are needed:\n\n1. DocumentProfile (crates/quarto-core/src/document_profile.rs:351) has no aliases field. Add 'pub aliases: Vec<String>' and extract it next to the sibling frontmatter keys around line 695 — extract_string_list(meta, \"aliases\") is exactly the shape categories: already uses. This bumps DOCUMENT_PROFILE_VERSION (currently 9, line 83).\n\n2. crates/quarto-core/src/project/website_post_render.rs is the natural home for the writer. It already owns exactly this class of work — per-project post-render side files, walking index.profiles(): the favicon copier (line ~66) and the sitemap.xml emitter (line ~150), the latter already iterating profiles and mapping them to output paths.\n\n## Suggested fix direction\n\nAdd a write_alias_redirects pass in website_post_render.rs alongside the sitemap writer: for each profile with a non-empty aliases list, resolve each alias against the output directory (absolute aliases are site-root-relative; relative ones resolve against the page's own output location), compute the relative href from the stub back to the page's output_href, and write the Q1 stub template. Worth deciding explicitly: whether to warn or error when two pages claim the same alias, or when an alias collides with a real output path — Q1 lets the stub overwrite, which on a case-insensitive filesystem produces a genuinely confusing result we have already hit in the Connect docs.\n\n## Real-world hit in the Connect docs\n\n69 source files declare aliases:. The Quarto 1 reference site contains 99 redirect stub HTML files that the q2 render omits entirely — that is the whole 451-vs-352 HTML file-count gap between the two rendered sites. Every external link, bookmark and search-engine result pointing at a pre-reorganization Connect docs URL would 404. This is the single largest remaining structural difference between the Q1 and q2 renders of that project; content-wise the port is otherwise at 352 of 352 pages with zero errors.\n\n## Repro\n\n/Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/aliases-redirects-missing/\n\nRun 'q2 render' there: 2 HTML files are written (index.html and current/index.html) where Quarto 1 writes 4, the two extra being the redirect stubs for the aliases declared in current/index.qmd. Compare with 'quarto render'.\n\nObserved continuously from q2 0.15.0 through 0.19.0. Origin strand in the q2-connect-docs skein: br-aliases-redirects-8gk18exz.","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-08-12T14:57:17.674476Z","created_by":"Claude (q2-connect-docs)","updated_at":"2026-10-28T13:40:31.587386Z","closed_at":"2026-10-28T13:40:31.587386Z","close_reason":"Feature shipped: aliases: redirect stubs landed via claude-notes/plans/2026-08-12-aliases-redirect-stubs.md (status: complete); docs at docs/guides/projects/aliases.qmd. Strand was left in_progress by oversight; noticed during triage of GH #944 (bd-ex8qvuwu), which revisits decision 3 of that plan.","labels":["website"],"dependencies":{"bd-hzwecpyk:related":{"depends_on_id":"bd-hzwecpyk","type":"related","created_at":"2026-08-12T15:08:37.250245Z","created_by":"Carlos Scheidegger"},"bd-wch2dotq:related":{"depends_on_id":"bd-wch2dotq","type":"related","created_at":"2026-08-12T15:08:38.787787Z","created_by":"Carlos Scheidegger"}},"comments":{"c-6wpn8y9p":{"id":"c-6wpn8y9p","author":"Carlos Scheidegger","created_at":"2026-08-12T18:48:49.512310Z","text":"Design decisions settled with the user 2026-08-12 (commit 672cb26f, plan § Design decisions):\n\n1. bd-hzwecpyk closed as duplicate; this is the implementation strand.\n2. Collisions are HARD ERRORS, diverging from Q1's warn-and-skip. Three conditions: stub-vs-rendered-page, same alias + same fragment key from two pages, and case-only collisions. New codes Q-5-23+ in the 'project' subsystem.\n3. Case-only collisions error on EVERY platform, not just case-insensitive filesystems, so Linux CI fails the same way macOS does.\n4. OPEN: stub template. Two candidates written out under the investigation dir. Byte-parity with Q1 explicitly not required; recommendation is the improved stub (noscript meta-refresh fallback, rel=canonical, charset, DOCTYPE, body link).\n5. 'aliases:' in a non-website project warns rather than staying silent.\n6. Hash fragments implemented properly in the first cut.\n7. draft: true pages emit no stubs — leaking a draft's existence via a live redirect is worse than over-hiding it. Cross-linked to bd-4zdf so the sitemap resolves the same way.\n8. Stale-stub cleanup deferred to bd-wdhhl0t9.\n\nConsequence worth flagging: decision 2 makes this a breaking change for existing Q1 sites. A project whose alias collisions Q1 tolerated will now fail to render. Phase 6 must check whether the Connect docs' 106 aliases trip it."},"c-dgw6tdmx":{"id":"c-dgw6tdmx","author":"Carlos Scheidegger","created_at":"2026-08-12T19:38:27.852163Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/515 (branch braid/aliases-redirect-stubs -> origin/feature/aliases-redirect-stubs). Awaiting CI + review."},"c-l7unwmt5":{"id":"c-l7unwmt5","author":"Carlos Scheidegger","created_at":"2026-08-12T19:35:45.174273Z","text":"IMPLEMENTED on branch braid/aliases-redirect-stubs (commit 7c30181d, unpushed). cargo xtask verify --skip-hub-build green; 11793 workspace tests pass.\n\nResult: the Connect docs now render 451 HTML files against Q1's 451 — the gap is closed exactly, with all 99 stub paths identical to Q1's.\n\nMAJOR FINDING: the first render of the unmodified Connect docs FAILED with two Q-5-25 case-only-collision errors, and both are real bugs that Q1 has been silently shipping. cookbook/content/integrations/{msgraph,sharepoint}/viewer/r/index.qmd each declare an alias '/…/viewer/R/' (capital R) while rendering to '…/viewer/r/index.html'. On macOS those are one file. Q1's guard is an exact string compare, so it writes the stub on top of the rendered page. Verified in docs-quarto-1/_site: both of those files are redirect stubs whose target is '../r/index.html' — pointing at themselves. Two documentation pages are destroyed in the published Q1 site, replaced by infinite redirect loops.\n\nSo the counts reconcile as Q1 = 350 real pages + 101 stubs, q2 = 352 real pages + 99 stubs. Same total; q2 has the two pages Q1 ate.\n\nACTION NEEDED IN THE PORT REPO (not here): those two aliases must be removed before docs-quarto-2 renders under q2. The 451-vs-451 measurement was taken on a scratch copy with them removed, so the port repo was never modified.\n\nAlso fixed two things found on the way:\n- config_sources::rebase_source_candidates — a collision diagnostic spans two documents and both root front matter at the dense FileId(0), so the existing binder would have rendered the second document's offsets against the first document's text (the bd-m6wmztln mis-pairing, reached from the other direction).\n- orchestrator no longer stringifies post_render errors that already carry diagnostics.\n\nRemaining: user review, then push + PR. Follow-up bd-wdhhl0t9 (stale stubs) still open."},"c-o0vzdywr":{"id":"c-o0vzdywr","author":"Carlos Scheidegger","created_at":"2026-08-12T15:08:21.190507Z","text":"Investigated on main @ db075676. Plan skeleton: claude-notes/plans/2026-08-12-aliases-redirect-stubs.md; artifacts under claude-notes/plans/aliases-redirect-stubs-investigation/.\n\nVerdict: ready to design, pending 7 policy questions in the plan.\n\nCorrections to this strand's description:\n- Q1 does NOT let an alias stub overwrite a real output. It warns and skips (website-aliases.ts, 'would overwrite the target. Skipping.'). What Q1 misses is the case-insensitive-filesystem variant, since its guard is an exact string compare.\n- Duplicate of bd-hzwecpyk (cderv, 2026-06-23), which already carries the 'blocks bd-0tr6' edge. Dedup is the user's call.\n\nNew findings:\n- q2 needs none of Q1's ~60-line incremental workaround: Pass 1 profiles every file regardless of RenderMode, so index.profiles() always carries the full alias set in post_render.\n- Measured the Connect-docs corpus: 106 unique alias entries across 69 files. 76 site-root-relative, 30 page-relative, 77 trailing-slash, 18 .html, 5 with #fragments. Every Q1 fixupHref branch is exercised.\n- Hash fragments are NOT deferrable: cookbook/off-host-execution/creating-execution-environments/index.qmd and cookbook/operations/deploying-content/index.qmd both claim /cookbook/custom-execution-environments/index.html with different fragments pointing at different targets. A fragment-less first cut would emit wrong redirects, not just incomplete ones.\n- Ordering hazard: post_render runs before the resource-copy pass, so a declared resource could silently clobber a stub."}}}
 {"id":"bd-amms9jvg","title":"Phase 1.2: vendor reveal.js 6 dist into resources/revealjs/","description":"Copy reveal.js 6 dist (reveal.js, reveal.css, one theme, needed plugins) into resources/revealjs/ + README documenting source/version (mirror resources/scss/README.md). External Sources Policy: never reference external-sources/ or node_modules/ at compile/runtime. Keep cargo xtask lint green.","status":"closed","priority":1,"issue_type":"task","created_at":"2026-06-08T20:35:33.100941Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-08T21:47:32.725987Z","closed_at":"2026-06-08T21:47:32.725987Z","close_reason":"Vendored reveal.js 6.0.0 (MIT) into resources/revealjs/ (reset/reveal/theme/js + LICENSE + README), embedded via include_str!. xtask lint green. Commit c43c2406.","labels":["revealjs"],"dependencies":{"bd-2m4wanyd:parent-child":{"depends_on_id":"bd-2m4wanyd","type":"parent-child","created_at":"2026-06-08T20:35:33.100941Z","created_by":"Carlos Scheidegger"}},"comments":{"c-fea7h12n":{"id":"c-fea7h12n","author":"Carlos Scheidegger","created_at":"2026-06-08T21:00:39.986194Z","text":"Asset mechanism decided 2026-06-08: vendor reveal.js 6 into resources/revealjs/ and embed via include_dir!/include_str! so q2 stays a single binary. Not copy-to-output. Keep cargo xtask lint (external-sources-in-macro) green."}}}
 {"id":"bd-an6z4","title":"Error-docs content authoring (umbrella): populate pages for the 133 current catalog entries","description":"Author /docs/errors/Q-X-Y.qmd pages for each entry in error_catalog.json. Open one sub-issue per subsystem when ready to work on it; closing a sub-issue requires all pages in that subsystem at status=stub or better.\n\nSubsystem sizes (133 total):\n- markdown: 36\n- yaml: 21\n- writer: 21\n- xml: 16\n- listing: 16\n- template: 7\n- navigation: 7\n- project: 3\n- theme: 2\n- internal: 2\n- lua: 1\n- cli: 1\n\nEach page is hand-authored; tooling (bd-tooling) generates stubs but does NOT generate prose. Per-page quality bar is in the foundation plan.\n\nPlan: claude-notes/plans/2026-05-22-error-docs-content.md","status":"closed","priority":3,"issue_type":"epic","created_at":"2026-05-22T17:09:37.728930Z","created_by":"cscheid","updated_at":"2026-06-10T18:38:09.608132Z","closed_at":"2026-05-24T21:16:46.561908Z","labels":["content","documentation","error-reporting"],"dependencies":{"bd-4ltdl:blocks":{"depends_on_id":"bd-4ltdl","type":"blocks","created_at":"2026-05-24T21:07:57.094765Z","created_by":"cscheid"},"bd-732bn:blocks":{"depends_on_id":"bd-732bn","type":"blocks","created_at":"2026-05-24T21:07:57.904153Z","created_by":"cscheid"},"bd-8k4ny:blocks":{"depends_on_id":"bd-8k4ny","type":"blocks","created_at":"2026-05-24T21:07:57.632775Z","created_by":"cscheid"},"bd-8zim0:blocks":{"depends_on_id":"bd-8zim0","type":"blocks","created_at":"2026-05-24T20:54:12.747207Z","created_by":"cscheid"},"bd-94x8a:parent-child":{"depends_on_id":"bd-94x8a","type":"parent-child","created_at":"2026-05-22T17:09:37.728930Z","created_by":"cscheid"},"bd-c263g:blocks":{"depends_on_id":"bd-c263g","type":"blocks","created_at":"2026-05-24T21:07:58.176873Z","created_by":"cscheid"},"bd-gkqxl:blocks":{"depends_on_id":"bd-gkqxl","type":"blocks","created_at":"2026-05-24T21:07:57.361666Z","created_by":"cscheid"},"bd-nvlxn:blocks":{"depends_on_id":"bd-nvlxn","type":"blocks","created_at":"2026-05-22T17:09:41.059965Z","created_by":"cscheid"},"bd-oqbpi:blocks":{"depends_on_id":"bd-oqbpi","type":"blocks","created_at":"2026-05-24T20:35:56.160185Z","created_by":"cscheid"},"bd-s9emf:blocks":{"depends_on_id":"bd-s9emf","type":"blocks","created_at":"2026-05-24T21:07:56.828284Z","created_by":"cscheid"},"bd-tep4x:blocks":{"depends_on_id":"bd-tep4x","type":"blocks","created_at":"2026-05-24T20:24:43.820419Z","created_by":"cscheid"},"bd-uhmzq:blocks":{"depends_on_id":"bd-uhmzq","type":"blocks","created_at":"2026-05-24T21:07:56.562544Z","created_by":"cscheid"}}}
 {"id":"bd-an9gkxnp","title":"qmd: block-level editorial marks (::: ++ / ::: -- / ::: >>)","description":"Add block-level editorial marking syntax to tree-sitter-qmd, mirroring the inline [++ ], [-- ], [>> ] marks: fenced divs opened with '::: ++', '::: --', '::: >>'. Refactor the fenced-div opener so '::: ^id' note definitions and the new editorial markers share one extensible 'sigil' dispatch. Plan: claude-notes/plans/2026-09-24-block-editorial-marks.md (assessment + phased implementation; awaiting user go-ahead).","notes":"Implemented on braid/bd-an9gkxnp-qmd-block-level-editorial (4 commits: 5337b87ef, 3f8ac499c, 54824ea1b, aad95cb70; cargo xtask verify green). All plan phases done; awaiting review/merge. Follow-ups: bd-t1avfz0n (q2-preview block-comment chrome). Pre-existing bugs found: bd-2281lkrx (fixed here), bd-9fy9p4zl (list-table content not postprocessed, P1), bd-i4xrcdqx (list-item blank line before ::: round-trip).","status":"in_progress","priority":2,"issue_type":"feature","created_at":"2026-09-24T15:17:01.303420Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-24T16:23:40.336444Z","labels":["parser","tree-sitter"]}
@@ -663,7 +667,7 @@
 {"id":"bd-bsut","title":"Match Quarto 1 page-navigation default + styling + icons","description":"Tweak prev/next page-navigation behavior in websites to match Quarto 1:\n\n1. Default visibility: Q1 defaults website page-navigation to false (books default to true). Today PageNavGenerateTransform always populates the strip when a sidebar exists, so links appear unbidden in minimal websites. Add config support for top-level / website.page-navigation, default false for websites, with document-level override.\n\n2. Styling: Q1's .page-navigation is display:flex; justify-content:space-between, putting prev on the left and next on the right of one row. Q2 currently emits the markup but has no horizontal layout rule, so each link wraps onto its own row, left-aligned. Port the Q1 SCSS rules.\n\n3. Arrow icons: Q2 emits <i class=\"bi bi-arrow-left-short\"></i> and bi-arrow-right-short, but Bootstrap Icons CSS is not bundled into Q2 output, so the glyphs render blank. Bundle bootstrap-icons (or a minimal subset) and add the <link rel=\"stylesheet\"> to the head.\n\nReference fixture: examples/websites/06-site-metadata. Plan: claude-notes/plans/2026-04-30-page-nav-q1-parity.md.\n\nPhases:\n- Phase 0: tests + fixture, capture current Q2 output for regression baselines\n- Phase 1: project-config gate (page-navigation default false for websites)\n- Phase 2: Q1 .page-navigation flex SCSS\n- Phase 3: bundle Bootstrap Icons + link in head\n- Phase 4: docs (link bd-nf50)","status":"closed","priority":1,"issue_type":"feature","created_at":"2026-04-30T13:14:56.003176Z","created_by":"cscheid","updated_at":"2026-04-30T13:54:01.336227Z","closed_at":"2026-04-30T13:54:01.336080Z","close_reason":"All four phases complete: project-config gate (default off for websites), Q1 page-nav layout SCSS ported, Bootstrap Icons vendored + auto-emitted, docs + examples updated. Full CI verify passes. End-to-end browser screenshot recorded at /tmp/q2-page-nav-after.png shows single-row strip with visible arrow glyphs. Plan: claude-notes/plans/2026-04-30-page-nav-q1-parity.md"}
 {"id":"bd-bthmzyrc","title":"DECISION: code a:any-link — keep Q2 downlit gray-underline or adopt TS no-underline/hover? (audit row 13g)","description":"TS Quarto _quarto-rules.scss:328-335: 'code a:any-link { color:inherit; text-decoration:none } code a:hover { color:inherit; text-decoration:underline }' (pkgdown/downlit style: no underline, underline on hover). Q2 ALREADY has a divergent rule at _bootstrap-rules.scss:2242: 'code a:any-link { color:inherit; text-decoration-color: $gray-600 }' (keeps underline but grays it). Same feature (downlit code links), different intentional design. Audit row 13g marked PN 'cheap' without noticing the existing rule. Porting TS would override Q2's choice. Needs Carlos's call: (a) keep Q2's gray-underline, mark 13g intentionally-divergent; (b) adopt TS's no-underline+hover; (c) hybrid. Held during bd-u5yvsdgw.","status":"closed","priority":2,"issue_type":"question","created_at":"2026-07-21T20:14:52.571100Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-21T20:46:01.079776Z","closed_at":"2026-07-21T20:46:01.079776Z","close_reason":"Adopted Quarto 1's code a:any-link behavior (text-decoration:none + code a:hover underline), replacing Q2's gray-underline variant, per Carlos. TDD+e2e; phase5 styles.css b06ec21d->a5d32eee. Merged to feature/quarto-rules-scss-parity.","dependencies":{"bd-4doe9lvt:parent-child":{"depends_on_id":"bd-4doe9lvt","type":"parent-child","created_at":"2026-07-21T20:14:52.571100Z","created_by":"Carlos Scheidegger"},"bd-u5yvsdgw:discovered-from":{"depends_on_id":"bd-u5yvsdgw","type":"discovered-from","created_at":"2026-07-21T20:14:52.571100Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-btjkyylx","title":"Title-block bottom margin missing in Q2 HTML (title block flush against content)","description":"Q2 HTML output has no gap between the title block (#title-block-header) and the first article content, because the base rule '#title-block-header { margin-block-end: 1rem; position: relative; margin-top: -1px; }' from TS Quarto's _quarto-rules.scss was never ported into Q2's SCSS bundle (quarto-sass). Q1 shows a 17px gap; Q2 shows 0px. Fix: add the base rule to resources/scss/html/templates/title-block.scss rules section. Affects both 'q2 render' and 'q2 preview' (shared quarto-sass bundle). See plan claude-notes/plans/2026-07-21-title-block-bottom-margin-parity.md","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-21T15:31:53.273233Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-21T17:42:15.019160Z","closed_at":"2026-07-21T17:42:15.019160Z","close_reason":"PR #406 merged as c7523c2b (title-block base rule ported, baseline re-captured)","labels":["bug","css"],"comments":{"c-tutyz5k3":{"id":"c-tutyz5k3","author":"Carlos Scheidegger","created_at":"2026-07-21T15:58:51.406897Z","text":"Fix committed (f0de2005) and PR #406 opened: https://github.com/quarto-dev/q2/pull/406 — branch feature/bd-btjkyylx-title-block-margin. Awaiting review/merge. Note: q2 preview requires a WASM rebuild + preview-server restart to reflect the change (stale-WASM); verified the ported SCSS is embedded in the freshly-built WASM."}}}
-{"id":"bd-btoxaf07","title":"Bump quarto-source-map 0.3.1 / quarto-error-reporting 0.4.1: Concat-shaped config diagnostics coalesce across pages","description":"posit-dev/quarto-error-reporting#10: Q-2-9 (and any diagnostic) located in a folded multi-line YAML scalar printed once per page because the coalescer could not key Concat-shaped locations. Fixed upstream in quarto-source-map 0.3.1 (SourceInfo::resolved_runs, posit-dev/quarto-source-map#9) and quarto-error-reporting 0.4.1 (posit-dev/quarto-error-reporting#11). q2 side: bump both workspace deps, add end-to-end tests (folded and block scalar fixtures over three pages) in crates/quarto/tests/integration/coalesced_diagnostics.rs. No q2 code change. Until both crates publish, the branch builds against their PR branches via [patch.crates-io] (REMOVE BEFORE MERGE).","status":"in_progress","priority":2,"issue_type":"chore","created_at":"2026-10-27T15:49:28.429636Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T15:49:32.481709Z"}
+{"id":"bd-btoxaf07","title":"Bump quarto-source-map 0.3.1 / quarto-error-reporting 0.4.1: Concat-shaped config diagnostics coalesce across pages","description":"posit-dev/quarto-error-reporting#10: Q-2-9 (and any diagnostic) located in a folded multi-line YAML scalar printed once per page because the coalescer could not key Concat-shaped locations. Fixed upstream in quarto-source-map 0.3.1 (SourceInfo::resolved_runs, posit-dev/quarto-source-map#9) and quarto-error-reporting 0.4.1 (posit-dev/quarto-error-reporting#11). q2 side: bump both workspace deps, add end-to-end tests (folded and block scalar fixtures over three pages) in crates/quarto/tests/integration/coalesced_diagnostics.rs. No q2 code change. Until both crates publish, the branch builds against their PR branches via [patch.crates-io] (REMOVE BEFORE MERGE).","status":"closed","priority":2,"issue_type":"chore","created_at":"2026-10-27T15:49:28.429636Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T17:23:36.608564Z","closed_at":"2026-10-27T17:23:36.608564Z","close_reason":"Merged in quarto-dev/q2#937; q2 on quarto-source-map 0.3.1 / quarto-error-reporting 0.4.1"}
 {"id":"bd-bu5y34fp","title":"Rebase/merge bd-ky14a (PR #235 hash-based FileIds) onto current main","description":"PR #235 (branch feature/bd-ky14a-pampa-hash-fileids, single commit eb726334) is 235 commits behind origin/main and CONFLICTING. It originally waited on #231, which was superseded by #260 (feat(provenance): targeted paragraph/header inline editing, merged 2026-06-10). Merging is NOT mechanical:\n\n1. engine_execution.rs has a genuine architectural conflict: main's multi-engine reconciliation loop (bd-5yff4) re-introduced the exact sequential-FileId + quarto_ast_reconcile::remap_file_ids workaround that bd-ky14a removes — now per-engine (FileId(merged_context.filenames.len())). The hash-based approach must be re-expressed inside that loop.\n\n2. Semantic entanglement with #260: crates/pampa/src/apply_node_edit.rs hard-codes FileId(0) as the primary file. lookup_block currently IGNORES that param (it is _file_id), matching on full SourceInfo equality, so the immediate risk is lower than it looks — but the pool 'd' decode (data.as_u64() as usize) and native/wasm usize truncation of the 64-bit hash need an audit.\n\n3. 61 snapshot conflicts are NOT mechanical: main's JSON writer renamed pool keys (sourceInfoPool→p, attrS→a) and changed pool layout (bd-xvdop integration-test move also changed the 'source:' path). Snapshots must be REGENERATED via cargo insta after the code is correct, not hand-merged.\n\nFull context: claude-notes/plans/2026-06-15-merge-bd-ky14a-onto-main.md (on the feature branch). Do NOT start implementation until cscheid gives the go-ahead.","status":"open","priority":2,"issue_type":"task","created_at":"2026-06-15T16:45:16.209239Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-15T17:00:48.057260Z","labels":["merge","refactor","source-location"],"dependencies":{"bd-ky14a:discovered-from":{"depends_on_id":"bd-ky14a","type":"discovered-from","created_at":"2026-06-15T16:45:16.209239Z","created_by":"Carlos Scheidegger"}},"comments":{"c-73hmp3h9":{"id":"c-73hmp3h9","author":"Carlos Scheidegger","created_at":"2026-06-15T16:47:11.878644Z","text":"Context note committed to feature/bd-ky14a-pampa-hash-fileids: claude-notes/plans/2026-06-15-merge-bd-ky14a-onto-main.md (commit 164d33d3). Awaiting cscheid go-ahead before starting the merge."},"c-u2xbb82u":{"id":"c-u2xbb82u","author":"Carlos Scheidegger","created_at":"2026-06-15T17:00:48.057260Z","text":"Reviewed PR #286 (block-editing epic, feature/block-editing-improvements). Findings added to the context doc (commit a35e85f3): #286 is MERGEABLE, draft, freshly rebased on main; Rust side purely additive (regenerate_nested_buffers + WASM binding, touches no FileId machinery). Real entanglement is the q2-preview block-editing source index's 'entry.d === 0' assumption — rooted in #260 (already on main), WIDENED by #286 (more d:0 siKey lookups + hardcoded :0 Rust tests). bd-ky14a turns d into a hash, silently emptying the source index → block editing breaks (no unit test catches it). RECOMMENDATION: wait for #286 to land, then rebase bd-ky14a and fix the d-scheme once. engine_execution.rs conflict is from #260 and unaffected by the wait."}}}
 {"id":"bd-but3","title":"Hub-client math rendering blocked by iframe sandbox","description":"bd-w5ov shipped MathJax + KaTeX for HTML output. Live Chromium smoke confirmed correctness on the native CLI render path. However, a post-implementation regression test in hub-client preview shows raw \\(e=mc^2\\) text — math is NOT typesetting there.\n\nRoot cause: the hub-client preview iframe is sandboxed without 'allow-scripts'. Set in DoubleBufferedIframe.tsx:347,355 and MorphIframe.tsx:564 (sandbox='allow-same-origin allow-popups'). iframePostProcessor.ts:149-167 documents that script inlining is disabled for the same reason. No <script> in the iframe executes today, so MathJax/KaTeX CDN loaders never run, and our inline window.MathJax config block also doesn't execute.\n\nThe WASM pipeline IS correctly wired to MathJsStage and meta.math IS populated; the iframe sandbox just drops everything on the floor.\n\nResolution: blocked on the existing (separate) plan to isolate the preview iframe on a subdomain via service workers, which would let us re-enable allow-scripts safely. Once that lands, math typesetting in hub-client should 'just work' with no changes to MathJsStage.\n\nAlternatives considered (not pursued in v1):\n- Add allow-scripts unconditionally — strictly weaker security model; rejected by user.\n- Server-side pre-typesetting (KaTeX renderToString or MathJax Node) on the WASM pipeline — non-trivial and not needed once scripts run.\n- Switch hub-client default to Temml/MathML (no JS runtime needed) — bd-si8b tracks this independently; would not fix MathJax/KaTeX users.\n\nAcceptance: when service-worker iframe isolation is in place, set sandbox='allow-scripts allow-same-origin allow-popups' (and re-enable script handling in iframePostProcessor.ts), then re-run a hub-client smoke against the user's reproducer doc:\n\n---\nformat: html\ntitle: Math maybe?\n---\nHello, world. Will this work? $e=mc^2$.\n\n— and confirm typeset output in the preview.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-05-05T00:11:28.712023Z","created_by":"cscheid","updated_at":"2026-05-05T00:11:28.712023Z","dependencies":{"bd-w5ov:discovered-from":{"depends_on_id":"bd-w5ov","type":"discovered-from","created_at":"2026-05-05T00:11:28.712023Z","created_by":"cscheid"}}}
 {"id":"bd-buwhvpc2","title":"RevealJS: reveal plugin foundation (menu/notes/line-highlight)","description":"Q2 ships ZERO reveal plugins in the render scaffold (no plugins: array). Q1 bundles menu (default), notes/search/zoom, line-highlight, pdfexport — enabling speaker view (S), menu, code-line stepping. Deferred from Stage D. Needs a vendor+register plugin foundation. Related: bd-0qaarvzx (speaker notes).","status":"open","priority":2,"issue_type":"feature","created_at":"2026-06-16T18:45:51.548809Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-16T18:45:51.548809Z","labels":["revealjs"],"dependencies":{"bd-yown2ts4:related":{"depends_on_id":"bd-yown2ts4","type":"related","created_at":"2026-06-16T18:45:51.548809Z","created_by":"Carlos Scheidegger"}}}
@@ -701,6 +705,7 @@
 {"id":"bd-cig41dcs","title":"CI flake: ts_engine test_race_free_instance_exclusive times out (watchdog DEADLOCK DETECTED)","description":"Failed on PR #906 (ubuntu-latest, run 39851900839 job 120081314475), unrelated to the PR's diff:\n\n  FAIL [15.010s] quarto-core engine::ts_engine::tests::test_race_free_instance_exclusive\n  panicked at crates/quarto-core/src/engine/ts_engine.rs:1367:14: DEADLOCK DETECTED: test timed out: Timeout\n\nSuspected cause (from reading the test, not reproduced): the launch_watcher thread polls sent_messages() for at most 300 x 10 ms = 3 s and then exits. If neither ensure_launched thread has sent LaunchEngine within that window (plausible on a loaded CI runner running ~17k tests in parallel), no response is ever delivered, both ensure_launched calls block forever, and the 15 s watchdog fires. The watcher also uses the message's index in sent_messages() as its response id (sent_messages drops the real ids), which is correct today only by coincidence of id assignment.\n\nFix direction: make the watcher wait on the condition rather than a fixed 3 s budget (loop until it sees LaunchEngine or the watchdog fires; or block on the mock's condvar), and deliver using the recorded request id instead of the index. Then check the same pattern in sibling tests in ts_engine.rs.","notes":"PR: https://github.com/quarto-dev/q2/pull/910 (branch bugfix/bd-cig41dcs-ts-engine-race-flake). Close on merge.","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-10-22T21:26:26.727289Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-23T02:53:41.134678Z","labels":["flake"],"dependencies":{"bd-t3n1nghb:discovered-from":{"depends_on_id":"bd-t3n1nghb","type":"discovered-from","created_at":"2026-10-22T21:26:26.727289Z","created_by":"Carlos Scheidegger"}},"comments":{"c-wzydg9vj":{"id":"c-wzydg9vj","author":"Carlos Scheidegger","created_at":"2026-10-23T00:27:45.575965Z","text":"Rerun of the same commit (786a39957, run 39851900839 attempt 2) passed on ubuntu-latest, so this is a flake, not a regression."}}}
 {"id":"bd-citeproc-cluster-collapse-kns8uggw","title":"Q2 citeproc output differs from pandoc --citeproc: same-author cluster not collapsed, bibliography punctuation outside quotes","description":"Found 2026-10-01 while porting Q1 typst-citeproc.qmd (after PR #770 made citeproc: true work on Typst). Same input through q2 render --to typst (citeproc: true) vs real pandoc --citeproc, default style.\n\n1. In-text, same-author cluster [@Cronbach_1951; @Cronbach_1952] (both 'Cronbach, Lee J. 1951', disambiguated 1951a/1951b): pandoc gives '(Cronbach 1951a, 1951b)'; Q2 gives '(Cronbach 1951a; Cronbach 1951b)' (author repeated, ';' delimiter, no collapse).\n2. Bibliography entry: pandoc gives 'Psychometrika... \"Coefficient Alpha and the Internal Structure of Tests.\"' (period inside the closing quote, en-US punctuation-in-quote); Q2 gives 'Tests\"\\.' with the period outside the quote.\n\nYear-suffix disambiguation itself works (1951a/1951b both appear), so this is not bd-bkcgdee7's original symptom (whose repro used the malformed comma form [@a, @b], which pandoc itself nests). Both are low severity cosmetic divergences; fixture workspace-4 crates/quarto/tests/smoke-all/typst/citations/typst-citeproc.qmd accepts both forms until fixed. Starting point: crates/quarto-citeproc (cite-group-delimiter / collapse handling, locale punctuation-in-quote).","status":"open","priority":3,"issue_type":"bug","created_at":"2026-10-01T22:48:35.983432Z","created_by":"Gordon Woodhull","updated_at":"2026-10-01T22:48:35.983432Z"}
 {"id":"bd-cjf31ps7","title":"Heading-id deduplication can still emit duplicate ids","description":"The `seen_ids` counter in the reader's header filter (crates/pampa/src/pandoc/treesitter_utils/postprocess.rs:946-952) never records the ids it *generates*, only the base ids it derives. A generated `base-N` can therefore collide with a later (or earlier) heading, and q2 emits the same `id` twice — invalid HTML, and a deep link to the anchor silently resolves to whichever section comes first.\n\n```rust\nlet final_id = if let Some(count) = seen_ids.get_mut(&base_id) {\n    *count += 1;\n    format!(\"{}-{}\", base_id, count)   // <- never inserted into seen_ids\n} else {\n    seen_ids.insert(base_id.clone(), 0);\n    base_id\n};\n```\n\n## Measured 2026-08-13\n\nq2 at commit 6af97135 (branch braid/bd-heading-id-drops-inline-content-fl84n3ql) vs pandoc 3.9.0.2 (`pandoc -f markdown+smart -t html --section-divs`). Two independent collisions, one in each direction:\n\nSource `## a` / `## a` / `## a-1`:\n\n| | Pandoc | q2 |\n|---|---|---|\n| 1 | `a` | `a` |\n| 2 | `a-1` | `a-1` |\n| 3 | `a-1-1` | `a-1`  **(duplicate)** |\n\nSource `## b` / `## b-1` / `## b`:\n\n| | Pandoc | q2 |\n|---|---|---|\n| 1 | `b` | `b` |\n| 2 | `b-1` | `b-1` |\n| 3 | `b-2` | `b-1`  **(duplicate)** |\n\nCase 1: a heading whose *literal* text slugs to `a-1` collides with the id generated for a repeated `## a`. Case 2: the generated `b-1` for the second `## b` collides with an *earlier* literal `b-1`. Note the counter is not even consulted for the literal ones, so the order does not matter.\n\n## Why Pandoc does not have this\n\n`uniqueIdent` keeps a `Set` of every identifier already handed out and searches for the first free `base-N`, rather than keeping a per-base counter. Note the resulting rule in case 1: `a-1` is treated as its own base, so the third heading becomes `a-1-1`, not `a-2`.\n\n## Suggested fix direction\n\nReplace the `HashMap<String, usize>` counter with a used-set (or keep the counter as a search hint but verify membership before emitting, and insert the emitted id). Matching Pandoc exactly means: given `base`, if `base` is free use it, else try `base-1`, `base-2`, ... until free, and record whatever is emitted.\n\nWatch for snapshot churn: unlike bd-heading-id-drops-inline-content-fl84n3ql (which touched zero snapshots and zero docs/ anchors) this changes ids only in documents that already have a collision, but any such document in the fixture corpus will move.\n\n## Provenance\n\nFound by reading `postprocess.rs` while fixing bd-heading-id-drops-inline-content-fl84n3ql, then confirmed by rendering. Pre-existing and independent of that fix — the collector fix makes empty-derived ids much rarer, which if anything makes this *less* likely to be hit, not more. Filed at the user's request rather than folded in, since it is a different defect in a different function.\n\np2 rather than p3 because the failure mode is duplicate `id` attributes in shipped HTML, which is invalid and breaks anchors silently.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-08-13T22:16:13.069397Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-13T22:16:13.069397Z","labels":["markdown"],"dependencies":{"bd-heading-id-drops-inline-content-fl84n3ql:discovered-from":{"depends_on_id":"bd-heading-id-drops-inline-content-fl84n3ql","type":"discovered-from","created_at":"2026-08-13T22:16:13.069397Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-cjqhcaz6","title":"q2 add / q2 use brand: non-interactive --dry-run of a remote source is impossible (--trust rejected with --dry-run)","description":"## Symptom\n\nWithout a terminal (`--no-prompt`, `--json`, CI, an agent's shell), you cannot dry-run `q2 add` or `q2 use brand` against a remote source. Each of the two errors tells you to do what the other one forbids:\n\n```\n$ q2 add posit-dev/product-doc-theme@q2 --dry-run          # no TTY\nError: Refusing to download posit-dev/product-doc-theme@q2 on GitHub without confirmation\n... There is no terminal to ask on, so nothing was fetched.\nRe-run with --trust to confirm you trust the authors of this extension.\n\n$ q2 add posit-dev/product-doc-theme@q2 --dry-run --trust\nError: --trust cannot be combined with --dry-run\n--dry-run writes nothing, so there is nothing for --trust to authorize. Run with\n--dry-run alone to preview, then re-run with --trust to proceed.\n```\n\nSeen with q2 0.36.0 while installing posit-dev/product-doc-theme into a downstream website port. Any remote target reproduces it (`<org>/<repo>`, an archive URL). Local folders and archives are unaffected because they skip the trust gate.\n\n## Cause\n\n- `crates/quarto/src/commands/add/mod.rs` (`execute`, the loop over `(\"--force\", ...), (\"--trust\", ...)`) rejects `--trust` together with `--dry-run`. The same loop is in `crates/quarto/src/commands/use_cmd/mod.rs` (`execute_brand`).\n- `crates/quarto/src/commands/add/resolve.rs` `confirm_trust` (called from the `Target::Remote` arm) runs whatever `--dry-run` says, and with no prompter it refuses unless `req.trust` is set. `use_cmd/brand.rs` has the same gate.\n\nThe guard's rationale (\"--dry-run writes nothing, so there is nothing for --trust to authorize\") holds for `--force`, but not for `--trust` on a remote source. A dry run still downloads the archive to compute the `[Install]/[Update]/[No Change]` table, and the trust gate controls that download, not the write. So `--trust` does authorize something during a dry run: the fetch. The rejection is deliberate and tested: `crates/quarto/tests/integration/use_brand.rs` `trust_with_dry_run_is_rejected`. For `add`, the end of `dry_run_shows_the_table_and_writes_nothing` in `tests/integration/add.rs` tests only `--force`.\n\n## Proposed fix\n\nAllow `--trust` with `--dry-run`, and keep rejecting `--force` with `--dry-run` (that pairing really is contradictory: `--force` only authorizes overwriting local changes, which a dry run never does). Apply the change in both `q2 add` and `q2 use brand` so the two commands stay in step, as their comments promise.\n\n- Drop `\"--trust\"` from the two guard loops; `confirm_trust` already does the right thing with `req.trust`.\n- Change the no-prompter refusal hint so it is correct in both modes, e.g. \"Re-run with --trust (it also works with --dry-run)…\".\n- Update the docs page `guides/authoring/extensions.md` (\"Remote sources must be trusted\" bullet) to say that a non-interactive dry run of a remote source needs `--trust`.\n\nThe other option, letting a dry run fetch without trust, is weaker. The `use brand` gate treats the download itself as the sensitive step (\"Downloading a brand runs someone else's content through this machine\"), so the fetch should keep needing consent in every mode. `--trust` is how a non-interactive caller gives it.\n\n## Tests\n\n- `add.rs`: a dry run of a remote archive URL with `--trust` (reuse the local archive server from `a_trusted_archive_url_installs`) succeeds, prints the `[Install]` table, and leaves `_extensions/` absent. `--dry-run --force` stays rejected.\n- `use_brand.rs`: flip `trust_with_dry_run_is_rejected` to the same shape: `--trust --dry-run` on a remote target previews and writes nothing.\n- No-TTY `--dry-run` without `--trust` on a remote source is still refused before any request, as `a_remote_source_without_trust_is_refused_before_any_request` checks today.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-10-27T16:24:25.968687Z","created_by":"Claude Opus 5.5","updated_at":"2026-10-27T22:37:49.541367Z","closed_at":"2026-10-27T22:37:49.541367Z","close_reason":"Fixed in PR #948 (merged 2026-10-27, 06460329f)","labels":["cli","extensions"],"dependencies":{"bd-0i6fii0k:discovered-from":{"depends_on_id":"bd-0i6fii0k","type":"discovered-from","created_at":"2026-10-27T16:24:25.968687Z","created_by":"Claude Opus 5.5"},"bd-rfz4e3bw:related":{"depends_on_id":"bd-rfz4e3bw","type":"related","created_at":"2026-10-27T16:24:25.968687Z","created_by":"Claude Opus 5.5"}},"comments":{"c-rycsmcdq":{"id":"c-rycsmcdq","author":"Carlos Scheidegger","created_at":"2026-10-27T21:42:16.444783Z","text":"Implemented per plan claude-notes/plans/2026-10-27-trust-with-dry-run.md; PR #948 (https://github.com/quarto-dev/q2/pull/948). --trust now combines with --dry-run in q2 add and q2 use brand; --force still rejected; mode-aware refusal hint; docs updated."}}}
 {"id":"bd-cjxlb","title":"Keyring error messages must not leak credential blob","description":"Plan §Phase 5 calls this out as a specifically-flagged risk: when CredentialStore.write/read/clear surfaces a backend error, the error message must NOT contain the credential blob (or any token-shaped substring).\n\nTest name: keyring_error_does_not_leak_blob_in_message in ts-packages/quarto-hub-mcp/test/auth/credential-store.test.ts.\n\nImplementation: wrap keyring errors via redact(err.message) at every site that touches a blob.\n\nPlan §Phase 5: claude-notes/plans/2026-05-05-hub-mcp-device-flow-implementation.md","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-05-20T14:27:50.747041Z","created_by":"shikokuchuo","updated_at":"2026-07-27T23:03:09.356871Z","closed_at":"2026-07-27T23:03:09.356871Z","close_reason":"Done. credential-store.ts redacts via redactTokens(errMessage(err)) at both KeyringUnavailableError throw sites (:187, :200). Tests: 'keyring error message does not leak credential bytes' (credential-store.test.ts:399) and 'unavailable-backend warning on read does not leak credential values' (:460). Shipped in 7599a32f (#234).","dependencies":{"bd-sfr5l:parent-child":{"depends_on_id":"bd-sfr5l","type":"parent-child","created_at":"2026-05-20T14:27:50.747041Z","created_by":"shikokuchuo"}}}
 {"id":"bd-ckbqmupi","title":"q2 render --json-errors: warning records lack a file path; pass-1 records carry a redundant ANSI text blob","description":"Repro: `q2 render <project> --json-errors 2>out.jsonl` on a multi-file project with parse errors and warnings (e.g. claude-notes with `_quarto.yml` `render: [\"**/*.md\"]`: 917 pass-1 failure records, 131 warning records).\n\n1. Standalone warning records ($schema json-diagnostic.json: code, title, problem, hints, start/end line/col, details, rendered) have NO source file field. In a multi-file render they cannot be attributed to a file without scraping the ANSI `rendered` text. Primary bug.\n\n2. Pass-1 failure records ($schema json-pass1-failure.json: {source_file, diagnostics, error}) DO carry structured diagnostics in `diagnostics` (correct). But `error` is the ariadne-rendered ANSI-colored text (escape codes + OSC-8 hyperlinks), and it appears to render only a subset of `diagnostics`, so a consumer reading `error` undercounts. Under --json-errors it should be plain text, or be dropped. Minor.\n\n(Originally filed claiming pass-1 failures had no structured data. That was wrong: I had read only `error` and missed the `diagnostics` array.)","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-09-23T21:15:53.291771Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-07T16:29:46.224923Z","closed_at":"2026-10-07T16:29:46.224923Z","close_reason":"Merged: source_file in PR #795, plain rendered/error in PR #800","dependencies":{"bd-gnw9asuo:parent-child":{"depends_on_id":"bd-gnw9asuo","type":"parent-child","created_at":"2026-10-06T16:56:22.887734Z","created_by":"t"}},"comments":{"c-0q3u4psp":{"id":"c-0q3u4psp","author":"t","created_at":"2026-10-06T19:30:57.644199Z","text":"R1 (ANSI/OSC-8 in JSON `rendered`) filed upstream in posit-dev/quarto-error-reporting's skein as qe-hal9cc7b: an opt-in plain rendering mode for diagnostic_to_json. Once released, q2 adopts it under --json-errors. The pass-1 `error` blob is built on the q2 side and stays here."},"c-109u6h7l":{"id":"c-109u6h7l","author":"t","created_at":"2026-10-06T18:55:53.580862Z","text":"Item 1 (warning records lack a file path) is fixed on braid/bd-gnw9asuo-q2-render-json-errors @ 70707f1ba: every located record now carries an absolute source_file derived from its own location. Item 2 (ANSI rendered/error blobs, R1 in the plan) is deferred and still open here."},"c-adn503df":{"id":"c-adn503df","author":"t","created_at":"2026-10-07T14:51:55.320529Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/800 (remote branch bugfix/bd-ckbqmupi-q2-render-json-errors)."},"c-jo2ba195":{"id":"c-jo2ba195","author":"t","created_at":"2026-10-07T14:50:46.781201Z","text":"Item 2 implemented on braid/bd-ckbqmupi-q2-render-json-errors @ 874b9ac66 (off main 5c61810b1): bump to quarto-error-reporting 0.4.0; --json-errors rendered and pass-1 error are now plain. Plan: claude-notes/plans/2026-10-07-json-errors-plain-rendered.md. claude-notes: 0 escape-bearing strings; stderr 6.48 MB -> 1.45 MB. Full verify green. Not pushed."}}}
 {"id":"bd-ckra329s","title":"live-share P5: spike — join a hosted quarto-hub share URL","description":"Design note + prototype, go/no-go deliverable (not committed implementation). Parse https share URL (provide_hub parse_index_doc_id precedent); serve Phase-4 hub-client embed locally; local /ws reverse proxy dialing wss://quarto-hub.com/ws with Authorization: Bearer (hub-mcp PKCE/keyring + BearerDialer precedents); proxy /auth/actor. Open questions: bearer expiry mid-session (hub validates once at upgrade), browserless login flow, presence identity. Plan Phase 5.","status":"open","priority":2,"issue_type":"task","created_at":"2026-08-03T10:28:46.304959Z","created_by":"shikokuchuo","updated_at":"2026-08-03T10:28:46.304959Z","dependencies":{"bd-6y0p1bne:blocks":{"depends_on_id":"bd-6y0p1bne","type":"blocks","created_at":"2026-08-03T10:28:46.304959Z","created_by":"shikokuchuo"},"bd-jt1etjbn:blocks":{"depends_on_id":"bd-jt1etjbn","type":"blocks","created_at":"2026-08-03T10:28:46.304959Z","created_by":"shikokuchuo"},"bd-yyoyvx91:parent-child":{"depends_on_id":"bd-yyoyvx91","type":"parent-child","created_at":"2026-08-03T10:28:46.304959Z","created_by":"shikokuchuo"}}}
@@ -715,7 +720,7 @@
 {"id":"bd-commentblock-geometry-flake-d45dj81u","title":"Flaky: CommentBlock.geometry 'nudges two bubbles' reads the previous relayout pass on macOS CI","description":"Seen on PR #835 (TS Test Suite run 38590196251, macos-latest, 'Run test suite' job), in ts-packages/preview-renderer's integration tier:\n\n  CommentBlock.geometry.integration.test.tsx > nudges two bubbles with the same anchor apart, earlier up and later down\n  AssertionError: expected 24 to be +0   (at line 133: expect(nudge(c1)).toBe(0))\n\nThe PR only touches .github/workflows/release-pipeline.yml and quarto-hub-mcp's build-wasm-host.mjs, and the same suite passed on main 8 times that day including the branch's base (bf8d1780b). Received nudge(c1) = 24 is the result of the FIRST pass (p0 hovered: c0 pinned, c1 pushed down 24px), i.e. the assertion ran before the second hover's relayout pass executed.\n\nMechanism (CommentBlock.tsx): the test dispatches a native mousemove -> onDocumentMouseMove -> record.onPointer -> setIsHovered() (React state set outside React's event system, so the render is deferred to React's scheduler) -> effect at ~L1355 writes entryRef.current.hovered -> scheduleBubbleRelayout() -> requestAnimationFrame. The test's flushRelayout() awaits two rAFs. jsdom's rAF is a ~16ms interval; if React's render lands after the test's first rAF callback (which registers the second), the pass's rAF is queued behind the assertion. A loaded macOS runner makes that window easy to miss.\n\nLikely fix: make the hover state flush deterministic in the test — wrap the dispatchEvent in act() (or await waitFor on the transform) so the relayout rAF is registered before flushRelayout's frames; alternatively have flushRelayout poll until bubbleRelayoutScheduled has cleared. Other tests in the file use the same pattern and are exposed the same way.\n\nSeparately logged in the same job output, probably teardown noise: \"Uncaught TypeError: Cannot read properties of undefined (reading 'querySelector') at MutationObserver.<anonymous> (http://localhost:3000/:221:36)\".","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-14T16:09:46.318171Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-14T16:09:46.318171Z","labels":["flaky","preview-renderer"]}
 {"id":"bd-commonmark-reader-fileid-xtvc23oz","title":"commonmark reader registers its input twice and stamps nodes with the content-less FileId(0) (GH #883)","description":"readers/commonmark.rs::read builds ASTContext::with_filename (add_file(path, None) -> FileId(0)), then add_file(path, Some(input)) -> FileId(1), then stamps every node with FileId(0). Works by accident for on-disk paths (FileId(0) re-reads the file); for in-memory input no node maps to line/col. Fix mirrors qmd reader: single registration, use the returned FileId. test_diagnostic_path_normalization.rs pins FileId(1) and must flip. Plan: claude-notes/plans/2026-10-22-commonmark-reader-single-fileid.md","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-10-22T13:58:17.210603Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-22T20:31:54.229204Z","closed_at":"2026-10-22T20:31:54.229204Z","close_reason":"Fixed in PR #901 (merged to main as 8ddbf8b17, 2026-10-22).","labels":["gh-883","source-map"],"comments":{"c-syqwk3dr":{"id":"c-syqwk3dr","author":"Carlos Scheidegger","created_at":"2026-10-22T14:57:36.273851Z","text":"Fixed on braid/bd-commonmark-reader-fileid-xtvc23oz-commonmark-reader-registers-its (merged into feature/pandoc-read-source-locations); verify green. Not pushed / no PR yet."}}}
 {"id":"bd-cp3em","title":"CaptureSpliceStage drops includes/filters/supporting_files from recorded ExecuteResult","description":"crates/quarto-core/src/stage/stages/capture_splice.rs explicitly only consumes `result.markdown` from the captured ExecuteResult JSON. Comment in the code:\n\n> We don't need the rest of the ExecuteResult shape here (filters, includes, supporting_files) — those are engine-side concerns the splice doesn't reproduce in the AST.\n\n**Verified still present in PR #238 (`feature/multi-engine` branch)** — the new fold-over-Vec<EngineCapture> path drops aux fields per-iteration, same comment. So this is a multi-engine bug as well as a single-engine bug.\n\nEngines emitting side effects via `ExecuteResult.includes` (e.g. CSS/JS injection) work on `q2 render` but silently fail on `q2 preview`. Same shape as the CodeHighlightStage incident in CLAUDE.md.\n\nFor the mermaid first ship (bd-gwfdo), the plan v2 routes the jsdelivr <script> via the engine's `markdown` output (Q-C → C1) which dodges this gap. Fixing bd-cp3em would let us migrate to the cleaner ExecuteResult.includes path (Q-C → C2/C3) in a follow-up.\n\nDiscovered during mermaid engine design — see claude-notes/plans/2026-05-28-mermaidjs-engine-design.md gap G6.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-05-28T13:46:02.824396Z","created_by":"cscheid","updated_at":"2026-05-28T16:16:32.036257Z","dependencies":{"bd-c6h96:discovered-from":{"depends_on_id":"bd-c6h96","type":"discovered-from","created_at":"2026-05-28T13:46:02.824396Z","created_by":"cscheid"}}}
-{"id":"bd-cp9wzu6o","title":"hub-client useEffect refactor Phase 3: derived state out of effects","description":"Phase 3 of claude-notes/plans/2026-10-24-hub-client-useeffect-refactor.md: Editor.tsx diagnostics split (:748) + default file selection render-adjust (:765) + comparingWithMain render-adjust (:334); ReactAstSlideRenderer clamp during render (:90); dialog form resets via remount (NewFileDialog, NewAssetDialog + delete :79 suppression, ShareDialog); ProjectsHome openingId render-adjust (:697). Stacked on PR #925.","status":"in_progress","priority":2,"issue_type":"task","created_at":"2026-10-27T14:09:01.119097Z","created_by":"shikokuchuo","updated_at":"2026-10-27T14:38:10.597732Z","comments":{"c-2j6e5n3z":{"id":"c-2j6e5n3z","author":"shikokuchuo","created_at":"2026-10-27T14:38:10.597732Z","text":"Open as PR #931 (https://github.com/quarto-dev/q2/pull/931), stacked on #925, base refactor/bd-06s9sufx-useeffect-phase2."},"c-m5n88g2o":{"id":"c-m5n88g2o","author":"shikokuchuo","created_at":"2026-10-27T14:30:40.805222Z","text":"Phase 3 landed on refactor/useeffect-phase3 (commit c52965b7e), stacked on the Phase 2 branch. All six plan items done: Editor diagnostics split (useMemo) + default-selection render-adjust + compare-with-main render-adjust; slide-index clamp during render (controlled = display-clamp only); dialog resets via conditional mounts (NewFileDialog/NewAssetDialog/ShareDialog at all call sites, NewAssetDialog :79 suppression deleted); ProjectsHome openingId render-adjust. Editor.characterization.test.tsx added (reusable by Phases 4-5). test:ci all legs green (1728u/229i/434w), npm run build green, eslint net -10 pre-existing errors. No changelog entry (refactor; Phase 2 precedent). Not browser-smoked. Awaiting push approval + stacked PR."}}}
+{"id":"bd-cp9wzu6o","title":"hub-client useEffect refactor Phase 3: derived state out of effects","description":"Phase 3 of claude-notes/plans/2026-10-24-hub-client-useeffect-refactor.md: Editor.tsx diagnostics split (:748) + default file selection render-adjust (:765) + comparingWithMain render-adjust (:334); ReactAstSlideRenderer clamp during render (:90); dialog form resets via remount (NewFileDialog, NewAssetDialog + delete :79 suppression, ShareDialog); ProjectsHome openingId render-adjust (:697). Stacked on PR #925.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-10-27T14:09:01.119097Z","created_by":"shikokuchuo","updated_at":"2026-10-27T17:35:33.930478Z","closed_at":"2026-10-27T17:35:33.930478Z","close_reason":"Merged to main as PR #931 (part of the merged useEffect stack)","comments":{"c-2j6e5n3z":{"id":"c-2j6e5n3z","author":"shikokuchuo","created_at":"2026-10-27T14:38:10.597732Z","text":"Open as PR #931 (https://github.com/quarto-dev/q2/pull/931), stacked on #925, base refactor/bd-06s9sufx-useeffect-phase2."},"c-m5n88g2o":{"id":"c-m5n88g2o","author":"shikokuchuo","created_at":"2026-10-27T14:30:40.805222Z","text":"Phase 3 landed on refactor/useeffect-phase3 (commit c52965b7e), stacked on the Phase 2 branch. All six plan items done: Editor diagnostics split (useMemo) + default-selection render-adjust + compare-with-main render-adjust; slide-index clamp during render (controlled = display-clamp only); dialog resets via conditional mounts (NewFileDialog/NewAssetDialog/ShareDialog at all call sites, NewAssetDialog :79 suppression deleted); ProjectsHome openingId render-adjust. Editor.characterization.test.tsx added (reusable by Phases 4-5). test:ci all legs green (1728u/229i/434w), npm run build green, eslint net -10 pre-existing errors. No changelog entry (refactor; Phase 2 precedent). Not browser-smoked. Awaiting push approval + stacked PR."}}}
 {"id":"bd-cpyq99ps","title":"Flaky: preview-renderer integration run fails on an unhandled elementFromPoint error","description":"cargo xtask verify step 11 (shared preview-* package tests) intermittently fails with ALL TESTS PASSING: 'Test Files 48 passed (48) / Tests 565 passed | 1 skipped (566) / Errors 1 error'. Vitest treats an unhandled error as a run failure, so the whole verify goes red on a green test suite.\n\nThe error: TypeError: (intermediate value).elementFromPoint is not a function, thrown from prosemirror-view's posAtCoords (node_modules/prosemirror-view/dist/index.js:465) via placeCaretFromClick (ts-packages/preview-renderer/src/q2-preview/richtext/caretFromClick.ts:32) called from RichTextEditor.tsx:235. Originates in src/q2-preview/p3-4-inline-breadcrumb.integration.test.tsx, last test 'renders the breadcrumb INSIDE the toolbar row and suppresses the standalone chip (rich Para)'.\n\nRoot cause shape: jsdom does not implement document.elementFromPoint, and RichTextEditor.tsx:235 schedules the caret placement in a requestAnimationFrame. When the rAF callback happens to fire AFTER the test that set it up has finished, it throws outside any test, so no test catches it and vitest reports an unhandled error. Timing-dependent, hence intermittent.\n\nReproduction: intermittent. Observed once on 2026-07-28 during bd-dofxhzaj's pre-push verify; 3 immediate consecutive re-runs of 'npm run test:integration' in ts-packages/preview-renderer were all clean.\n\nLikely fixes (pick one, don't do all): (a) stub document.elementFromPoint in the integration-test setup file — jsdom will never have it, and every consumer of posAtCoords has this exposure, not just this test; (b) cancel the pending rAF in RichTextEditor's cleanup so no callback can outlive the component; (c) guard placeCaretFromClick on typeof document.elementFromPoint === 'function'. (a)+(b) together is probably the honest fix: (a) removes the crash, (b) removes the leak that makes it escape the test.\n\nNOT caused by the bd-dofxhzaj musl work — that branch's diff is .yml + .md only. Filed because a flake that reddens verify on an all-green suite will waste someone's afternoon.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-28T16:05:26.351402Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-28T17:27:06.493+00:00","closed_at":"2026-07-28T17:27:06.492+00:00","dependencies":{"bd-dofxhzaj:discovered-from":{"depends_on_id":"bd-dofxhzaj","type":"discovered-from","created_at":"2026-07-28T16:05:26.351402Z","created_by":"Carlos Scheidegger"}},"comments":{"c-dxc5bisx":{"id":"c-dxc5bisx","author":"Carlos Scheidegger","created_at":"2026-07-28T16:51:49.586191Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/431 (bugfix/bd-cpyq99ps-elementfrompoint-flake -> main, 1 commit). cargo xtask verify --skip-hub-build green exit 0.\n\nROOT CAUSE, and it is NOT what this strand originally guessed: one unguarded line in ProseMirror's posAtCoords — 'let elt = (view.root.elementFromPoint ? view.root : doc).elementFromPoint(...)' — guards view.root but not the doc fallback, and jsdom implements elementFromPoint on neither. The strand's guesses (b) 'cancel the pending rAF in cleanup' and (c) 'guard placeCaretFromClick' were both wrong: the mount effect ALREADY does 'return () => cancelAnimationFrame(raf)' AND guards editor.isDestroyed, and caretFromPoint (the sibling helper on the same path) already guards both of its own APIs. Only (a) — stub elementFromPoint in the test setup — was right, and for a better reason than I first gave.\n\nTwo things made it a flake rather than a failure: the call runs in a requestAnimationFrame (deliberately, so the editor box is laid out first), so machine load decides whether the throw is attributed to a test or escapes it; and NOTHING ever exercised the real path — caretFromClick.test.ts uses a fake editor with a vi.fn() posAtCoords, RichTextEditor.caret.integration.test.tsx vi.mocks the whole module, and the file that tripped it mocks nothing and reached real ProseMirror by accident. So caretFromClick.ts's own comment 'jsdom returns null from posAtCoords' was FALSE and had never been checked.\n\nFix: stub Document.prototype.elementFromPoint -> null in src/test-utils/setup.ts. Harness, not production — every real browser implements it, so a guard in caretFromClick.ts would be dead code papering over a jsdom gap; null is the honest answer (no layout, nothing under any point) and makes posAtCoords return the miss callers already handle. Follows a precedent already in the same file (getClientRects stubs on Text/Range for ProseMirror coordsAtPos, same rAF, same remedy).\n\nNew caretFromClick.integration.test.ts drives a REAL tiptap editor, nothing mocked, closing the coverage gap: 5/5 fail before with the exact prosemirror-view:465 stack, 5/5 pass after. Full integration suite x6 before (fails every run), x4 after (49 files, 570 tests, 0 errors). Unit suite unaffected.\n\nHONEST LIMIT: reproduced and fixed the root cause deterministically; did NOT reproduce the original intermittent unhandled-error form on demand (needs the rAF to lose its race against teardown, i.e. machine load). Does not weaken the fix — posAtCoords now has nothing to throw from regardless of when the frame fires.\n\nPlan: claude-notes/plans/2026-07-28-preview-renderer-elementfrompoint-flake.md"}}}
 {"id":"bd-cpzp","title":"qmd writer: implicit-figure path drops trailing newline, collapses next block (issue #180)","description":"Triage doc: claude-notes/issue-reports/180/triage.md on branch issue-180.\n\nRoot cause: write_figure (crates/pampa/src/writers/qmd.rs:759) implicit-figure branch returns directly from write_image without emitting the trailing \\n that every block writer is expected to produce. When such a Figure is followed by another block (top-level or inside a Div), the inter-block separator collapses to a single \\n instead of a blank line, and the re-parser glues the two blocks into one Para. Affects every layout/subfigure div in the docs corpus.\n\nReports covered (both same root cause):\n- Bug A: top-level Figure followed by Para -> one Para after round-trip\n- Bug B: Div with multiple Figure children + caption -> one Para inside the Div after round-trip\n\nFix (TDD-first per crates/pampa/CLAUDE.md):\n1. Add failing roundtrip tests under tests/roundtrip_tests/qmd-json-qmd/\n   - figure_implicit_then_para.qmd\n   - layout_div_subfigures.qmd\n   - figure_implicit_then_figure.qmd (extra coverage)\n2. Verify they fail.\n3. Fix write_figure: replace 'return write_image(...)' with 'write_image(...)?; writeln!(buf)?; Ok(())'.\n4. Verify tests pass.\n5. cargo nextest run --workspace to catch downstream snapshot updates.\n\nNot a duplicate of bd-emr4 — that is about non-implicit Figure shapes hitting the fallback fenced-div path. This bug is in the implicit-figure code path; different code, different fix.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-05-12T18:39:12.847875Z","created_by":"cscheid","updated_at":"2026-05-12T18:39:15.853892Z","dependencies":{"bd-emr4:related":{"depends_on_id":"bd-emr4","type":"related","created_at":"2026-05-12T18:39:15.853594Z","created_by":"cscheid"}}}
 {"id":"bd-cpzr71jr","title":"Windows: cache_set persist onto a just-rewritten file intermittently fails with os error 5 (cache_lru concurrent test)","description":"quarto-system-runtime cache_lru::tests::concurrent_get_and_set_lru_do_not_lose_the_set fails intermittently on Windows (1 of 40 reps on unchanged code, HEAD 8202128d of bugfix/bd-1klbq2zd-path-audit):\n\n  called `Result::unwrap()` on an `Err` value: CacheError(\"failed to persist cache entry sass/_lru_index: failed to persist temporary file: Access is denied. (os error 5)\")\n\nat cache_lru.rs:532 (the writer thread's cache_set_lru unwrap).\n\nTraced cause: not a q2 race. Every in-process access to _lru_index (load_index, cache_list reconcile, store_index) runs under the process-wide INDEX_LOCK; the reader thread's unlocked fs::read touches the value file \"hot\", not the index. A single-threaded PowerShell probe with no concurrency at all (write a temp file in %TEMP%, then [IO.File]::Move(tmp, target, overwrite) onto the same target, 3000 times) also hit \"Access to the path is denied\" once. So on this host, a replace-rename onto a file that was just rewritten occasionally fails because something outside the process (antivirus real-time scan or the search indexer, most likely) still holds a handle on it. NativeRuntime::cache_set (native.rs ~442) does exactly that rename via NamedTempFile::persist, and the test rewrites the index 128 times in a tight loop, so it hits the window now and then.\n\nProduct impact: small. The only production caller, compile_theme_css.rs ~382, ignores the cache_set_lru result (`let _ =`), cache_get_lru ignores its index-store error, and the next write's reconcile adopts entries the index missed. The worst case is a lost index update, not a failed render.\n\nOptions (not decided): retry persist a few times on PermissionDenied on Windows (a common pattern for this AV/indexer case; check ecosystem precedent, e.g. how cargo or rustup handle rename-over on Windows, before choosing), or make the test tolerate a failed index store the way production does. Do not mark the test flaky or skip it on Windows (CLAUDE.local.md).\n\nLiveness: repeat the test 40 times on Windows: cargo nextest run -p quarto-system-runtime --lib -E 'test(=cache_lru::tests::concurrent_get_and_set_lru_do_not_lose_the_set)'. Any \"os error 5\" on _lru_index means it is still live.\n","status":"open","priority":3,"issue_type":"bug","created_at":"2026-09-29T17:12:50.632927Z","created_by":"cderv","updated_at":"2026-09-29T17:12:50.632927Z","dependencies":{"bd-1klbq2zd:discovered-from":{"depends_on_id":"bd-1klbq2zd","type":"discovered-from","created_at":"2026-09-29T17:12:50.632927Z","created_by":"cderv"}}}
@@ -727,7 +732,7 @@
 {"id":"bd-cursor-slide-untitled-heading-p2wuuc13","title":"hub-client revealjs: cursor→slide sync lands on the previous slide after an untitled slide (empty `##` or image-only heading)","description":"In hub-client's split view for format: revealjs, moving the editor cursor into any section that follows an untitled slide (an empty '##', or '## ![alt](img.png)', or '## {background-image=...}') navigates the preview to the slide BEFORE the one the cursor is in. Reproduced 2026-10-20 in a local hub-client (localhost:5173) with a 4-slide deck whose second heading is a bare '##': cursor in slide 2's body shows slide 1, cursor in slide 3 shows slide 2, cursor in slide 4 shows slide 3. The outline panel also omits the untitled slide. Root cause: useCursorToSlide (hub-client/src/hooks/useCursorToSlide.ts) pairs reveal slides with LSP header symbols BY INDEX, but quarto-lsp-core's header_to_symbol returns None for a header whose text is empty (inlines_to_text ignores Image), so one symbol is missing and every later slide inherits the NEXT header's start line. Plan: claude-notes/plans/2026-10-20-cursor-slide-untitled-heading.md","status":"closed","priority":2,"issue_type":"bug","assignee":"claude","created_at":"2026-10-20T14:04:43.871592Z","created_by":"claude","updated_at":"2026-10-20T16:59:31.156772Z","closed_at":"2026-10-20T16:59:31.156772Z","close_reason":"Fixed in PR https://github.com/quarto-dev/q2/pull/868 (branch bugfix/bd-cursor-slide-untitled-heading-p2wuuc13): slide mapping read off the transformed AST; thumbnail gate and mid-capture rerun fixed alongside. Follow-ups bd-outline-untitled-slides-d525xvxv, bd-cursor-slide-vertical-index-wykcyjxb.","labels":["hub-client","revealjs"],"comments":{"c-itk9473u":{"id":"c-itk9473u","author":"claude","created_at":"2026-10-20T16:54:23.050596Z","text":"Implementation on branch braid/bd-cursor-slide-untitled-heading-p2wuuc13 (worktree). Mapping now read off the transformed AST (new hub-client/src/hooks/slideStartLines.ts); useCursorToSlide and useSlideThumbnails no longer take symbols. Browser-verified on :5174: bare ##, image-only and background-image headings, preamble, # Section stack, --- split all map to their own slide; last slide reachable. Two pre-existing thumbnail defects found and fixed in passing (plan: 'Discovered during execution'): the enable gate compared against 'q2-slides' while revealjs decks resolve to 'revealjs' (thumbnails were never generated), and a capture requested mid-capture was dropped (outline could show stale thumbnails). Follow-ups filed: bd-outline-untitled-slides-d525xvxv (outline), bd-cursor-slide-vertical-index-wykcyjxb (h,v). Remaining: cargo xtask verify, commit, PR."},"c-v08gm5lv":{"id":"c-v08gm5lv","author":"claude","created_at":"2026-10-20T14:06:48.251278Z","text":"Reproduced in the browser (DevTools-driven Chrome against npm run local-dev) with a 4-slide deck, second heading a bare '##': line 8→First ✓, line 12 (untitled body)→First ✗, line 16 (Third)→untitled ✗, line 20 (Fourth)→Third ✗. Same table with '## ![alt](missing.png)' as the heading. Outline panel omits the untitled slide in both cases. Diagnosis + recommended fix (derive slide start lines from the transformed AST's 'l' locations per top-level block instead of pairing parseSlides output with LSP symbols by index; same pairing exists in useSlideThumbnails) written up in claude-notes/plans/2026-10-20-cursor-slide-untitled-heading.md. Status: draft, awaiting review before implementation."}}}
 {"id":"bd-cursor-slide-vertical-index-wykcyjxb","title":"hub-client revealjs: mirror the vertical slide index (h, v) inside # Section stacks","description":"The cursor->slide channel (SET_SLIDE / SLIDE_CHANGED, bd-mwbsdmel) only carries the horizontal index h. A '# Section' heading (level below slide-level) becomes a reveal stack whose vertical children are the '##' slides under it; placing the cursor on any of those children navigates to the stack's first slide, and navigating vertically in the preview does not move the editor. Extend the message to (h, v): derive per-child start lines from the stack Div's children in the transformed AST (same helper as the fix for bd-cursor-slide-untitled-heading-p2wuuc13), post both indices, and have RevealNavSync call reveal.slide(h, v) and read getIndices().v on slidechanged. Carlos (2026-10-20): wanted, but later.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-10-20T16:34:59.326801Z","created_by":"claude","updated_at":"2026-10-20T16:34:59.326801Z","labels":["hub-client","revealjs"],"dependencies":{"bd-cursor-slide-untitled-heading-p2wuuc13:discovered-from":{"depends_on_id":"bd-cursor-slide-untitled-heading-p2wuuc13","type":"discovered-from","created_at":"2026-10-20T16:34:59.326801Z","created_by":"claude"}}}
 {"id":"bd-custom-template-not-templated-e5t6m0i0","title":"A custom listing template that uses no doctemplate directives is spliced into the page verbatim, with no diagnostic","description":"A `type: custom` listing whose `template:` file contains no doctemplate directives is copied byte-for-byte into the rendered page, with zero diagnostics and exit 0. The reader sees the raw template. Quarto 1, whose custom listings are EJS, renders real cards from the same file.\n\n## Symptom\n\nlisting: {type: custom, template: welcome-card.ejs, contents: [\"item-*.qmd\"]}\n\n_site/index.html contains, verbatim:\n\n    <div class=\"custom-card-grid<%= items[0].columns === 2 ? ' columns-2' : '' %>\">\n      <% for (const item of items) { %>\n        <a href=\"<%= item.link %>\" class=\"custom-card-wrapper\">\n\nRendered N of N files, no warnings, exit 0. A link checker then reports pages linking to the literal target `<%= item.link %>`.\n\n## Root cause\n\nThree verified links, all at v0.27.0:\n\n1. The doctemplate grammar treats `$` as its ONLY special character — crates/tree-sitter-doctemplate/grammar/grammar.js: `text: ($) => /[^$]+/`. A template containing no `$` therefore parses as a single literal-text node and compiles \"successfully\"; rendering it emits the input unchanged.\n\n2. load_custom_template (crates/quarto-core/src/transforms/listing_render.rs:375) does a plain fs::read_to_string with no content validation. The only diagnostic on that path is Q-12-8, for I/O failure.\n\n3. Nothing anywhere checks that a compiled template actually used a directive. This is an absent check, not a weak one.\n\nQ-12-7 structurally cannot cover this: it fires only when `l.template.is_some() && l.kind != ListingType::Custom` (crates/quarto-core/src/project/listing/config.rs:593). Here `type: custom` is set correctly, so the config is valid and the template is simply never templated.\n\n## The exact silence boundary\n\nTwo existing guards accidentally catch some bad templates. Measured:\n\n  template body                                    result\n  \"HAS A BARE $ SIGN AND NO DIRECTIVES\"            Q-12-10 \"template failed to compile: ParseError ... unexpected ' SIGN AND NO '\"\n  bare EJS markup, no $, does not re-parse         Q-12-10 \"Re-parsing rendered listing `c` failed ... first: Parse error\"\n  the same EJS wrapped in a ```{=html} block       SILENCE, dumped verbatim\n\nSo the trigger is not merely \"contains no bare `$`\" — it is \"contains no bare `$` AND the unchanged text re-parses cleanly as markdown\". Wrapping the template in a raw-HTML block satisfies both, and that is exactly how a well-formed HTML card template is naturally written. The better the template, the likelier the silence.\n\nA stray dollar sign gets you a diagnostic; a carefully written HTML template gets you none.\n\n## Suggested fix direction\n\nCheapest reliable signal: compare the compiled output against the template source — byte-identical means no directive was ever applied. That catches all three rows above uniformly and needs no grammar change. More thorough: have the compiler report whether any directive node was present, and warn when a custom template compiled to pure literal text.\n\nEither way the diagnostic should name the template path and say the file appears not to be a doctemplate, since the common cause will be a Quarto 1 EJS template carried over unconverted.\n\n## Real-world hit\n\nPositron website docs: 11 listings across 8 pages use templates/welcome-card.ejs or templates/feature-card.ejs, carried over from Quarto 1. Affected pages: welcome, features, assistant, assistant-getting-started, ai-features, language-guides, older-releases, tutorials — the site's entire landing and navigation surface. Every one renders raw EJS to the reader with no warning of any kind.\n\n## Repro\n\n/Users/gordon/src/q2-positron-docs/llms-info/repros/ejs-template-dumped/\n\n`q2 render`, then compare _site/repro.html against _site/control.html. Both declare the same listing over the same file-glob contents; only the template differs.\n\n  repro.qmd    template: welcome-card.ejs (copied byte-for-byte from the Positron site)\n               -> 9 literal '<%' in the output, ZERO warnings\n  control.qmd  template: a valid q2 doctemplate\n               -> 0 literal '<%', real cards, correct href=\"item-one.html\"\n\nQuarto 1 on the same repro renders real cards and exits 0. The repro uses a file-glob `contents:` deliberately, so it is independent of inline-record handling.\n\nOrigin strand in the Positron-docs porting skein: br-ejs-template-dumped-e7491tmf. Observed with q2 0.27.0 (official release build); origin/main == v0.27.0 == d05e96ee8 as of 2026-08-25.\n\nNOTE ON SCOPE: this is not a request for EJS support. Converting these templates to doctemplates is the porting project's own work and is tracked there. The ask here is only that q2 not emit an untemplated template into the page silently.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-08-25T15:15:05.077879Z","created_by":"Gordon Woodhull","updated_at":"2026-09-03T20:16:14.058991Z","closed_at":"2026-09-03T20:15:19.132861Z","close_reason":"Fixed and shipped: Q-12-24 (untemplated/EJS custom template warns and skips) plus Q-12-9 widened to .ejs, merged as PR #603, verified in q2 0.28.0","labels":["listings"],"dependencies":{"bd-61cd:parent-child":{"depends_on_id":"bd-61cd","type":"parent-child","created_at":"2026-08-25T17:09:29.805128Z","created_by":"Gordon Woodhull"},"bd-hzsi:related":{"depends_on_id":"bd-hzsi","type":"related","created_at":"2026-08-25T17:09:32.065423Z","created_by":"Gordon Woodhull"},"bd-lu16jgxq:related":{"depends_on_id":"bd-lu16jgxq","type":"related","created_at":"2026-08-25T17:09:36.742943Z","created_by":"Gordon Woodhull"},"bd-oywyaouf:supersedes":{"depends_on_id":"bd-oywyaouf","type":"supersedes","created_at":"2026-08-25T17:09:27.543049Z","created_by":"Gordon Woodhull"},"bd-u4ow:related":{"depends_on_id":"bd-u4ow","type":"related","created_at":"2026-08-25T17:09:34.325787Z","created_by":"Gordon Woodhull"}},"comments":{"c-1u6d0nxt":{"id":"c-1u6d0nxt","author":"Gordon Woodhull","created_at":"2026-08-25T19:25:44.053727Z","text":"PR: https://github.com/quarto-dev/q2/pull/603 (branch feature/bd-custom-template-not-templated-e5t6m0i0-custom-template-not-templated, 4 commits squashed from the working history; pre-squash history kept locally on history/braid/…)."},"c-84exmcqu":{"id":"c-84exmcqu","author":"Gordon Woodhull","created_at":"2026-09-03T20:15:16.307763Z","text":"Re-verified against the shipped q2 0.28.0 binary on 2026-09-03, repro /Users/gordon/src/q2-positron-docs/llms-info/repros/ejs-template-dumped/: Q-12-9 fires with a source span on the template: value, Q-12-24 names the listing and the template path, identifies <% ... %> as Quarto 1 EJS and ends 'Listing skipped.'; repro.html has 0 literal '<%' (was 9), control.html still renders real cards. PR #603 merged 2026-08-25; e633fcd2a..fcd76aebd are on main and contained in v0.28.0 and release/v0.29.0. The leftover local branches braid/bd-custom-template-not-templated-e5t6m0i0-custom-template-not-templated and its history/ twin were duplicates (git cherry: all four commits already upstream) and have been deleted."},"c-b2jqlkyo":{"id":"c-b2jqlkyo","author":"Gordon Woodhull","created_at":"2026-08-25T17:09:43.231607Z","text":"Supersedes bd-oywyaouf (same bug, filed 2026-08-06). Design settled with Gordon 2026-08-25: (1) Q-12-24 fires when a compiled custom template has no directive nodes OR its source contains `<%`; listing is SKIPPED (not defaulted); warning severity; comment-only templates count as doctemplate. (2) Q-12-9 widened to `.ejs`, reframed as 'Quarto 1 EJS template'. (3) Fix the Q-12-9/Q-12-10 error pages, which currently say Q2 listing templates are EJS. (4) Real custom-template docs section in docs/guides/projects/listings.qmd (links bd-hzsi, bd-u4ow). Plan: claude-notes/plans/2026-08-25-custom-template-not-templated.md"},"c-cd1h1urz":{"id":"c-cd1h1urz","author":"Gordon Woodhull","created_at":"2026-08-25T18:57:34.068525Z","text":"Implementation complete on branch braid/bd-custom-template-not-templated-e5t6m0i0-custom-template-not-templated (worktree workspace-2), 8 commits 5ee27ab57..954add367 on top of origin/main 05b6fd75c. Q-12-24 (untemplated / EJS custom template → warn + skip), Q-12-9 widened to .ejs with Quarto-1-EJS framing, e2e integration test, Listings guide 'Custom templates' section, five error pages corrected. Verified: workspace nextest 13389 passed / 199 skipped (+9 vs baseline, all accounted), full cargo xtask verify green, binary e2e on the Positron repro: 9 → 0 literal '<%', both warnings emitted. Follow-ups filed: bd-owflmojl (sniff escape hatch + partials), bd-o1meelim (absolute template: path). Awaiting Gordon's push/merge decision."},"c-hbieks45":{"id":"c-hbieks45","author":"Gordon Woodhull","created_at":"2026-09-03T20:16:14.058991Z","text":"Correction to the previous comment: only the duplicate branch braid/bd-custom-template-not-templated-e5t6m0i0-custom-template-not-templated was deleted. history/braid/bd-custom-template-not-templated-e5t6m0i0-custom-template-not-templated is kept, since this strand records it as the deliberately-preserved pre-squash working history."},"c-kpccral0":{"id":"c-kpccral0","author":"Gordon Woodhull","created_at":"2026-08-25T16:14:52.488949Z","text":"ADDITIONAL FINDING: q2 already has a diagnostic for precisely this situation, and it misses the more common file extension.\n\nQ-12-9 \"Deprecated .ejs.md Template Extension\" — \"The `.ejs.md` template extension is a Q1 idiom and is not supported by Q2 doctemplate. See the Q1 → Q2 migration guide.\" That is exactly the message an author carrying a Q1 custom listing template needs.\n\nIts trigger is a literal suffix test, crates/quarto-core/src/project/listing/config.rs:518:\n\n    if path.ends_with(\".ejs.md\") {\n\nSo `template: card.ejs.md` warns, and `template: card.ejs` does not. Verified end to end on 0.27.0 with two otherwise identical projects, same template body, differing only in the file extension:\n\n    template: card.ejs     -> Rendered 2 of 2, NO diagnostic, template dumped verbatim\n    template: card.ejs.md  -> Rendered 2 of 2, Q-12-9, template dumped verbatim\n\nBoth dump. Only one warns. Q1's own listing templates ship under both spellings (item-default.ejs.md, sidebar.ejs), and the Positron site — the real-world hit on this strand — uses plain `.ejs`: templates/welcome-card.ejs and templates/feature-card.ejs. So the eight affected pages sail past the one warning that exists for their exact situation.\n\nThis strengthens the case for the output-vs-source check proposed above rather than weakening it: an extension-based test cannot be the mechanism, because the extension is a naming convention and the failure is a content property. But widening Q-12-9 to `.ejs` as well is a cheap, independent improvement worth taking regardless — it costs one condition and would have caught the Positron case at config-parse time, before any rendering."},"c-kwwce5qj":{"id":"c-kwwce5qj","author":"Gordon Woodhull","created_at":"2026-08-25T18:52:55.072687Z","text":"VERIFIED against the fix branch, 2026-08-25. Built binary from worktree .worktrees/workspace-2, branch braid/bd-custom-template-not-templated-e5t6m0i0-custom-template-not-templated, tip 900cfdccc. Rendered the full Positron site on main (templates still .ejs, port not applied).\n\nThe fix does everything asked, and the extension widening as well:\n\n  Q-12-9   11 occurrences — widened from `.ejs.md` to also match `.ejs`, which is exactly the\n           additional finding recorded on this strand. Blames the `template:` value and says\n           Quarto 2 listing templates use doctemplate syntax.\n  Q-12-24  11 occurrences — new code. Names the listing id AND the template path, identifies\n           `<% … %>` as Q1 EJS, states that Quarto 2 does not evaluate EJS, points at the\n           Listings guide \"Custom templates\" section, and ends \"Listing skipped.\"\n\n11 and 11 is exactly the number of listings on this site using an .ejs template, so both codes fire on every affected listing and nowhere else.\n\nBehavioural change, which is the important half: RAW EJS NO LONGER REACHES THE PAGE. welcome.html contains 0 occurrences of `<%` (previously the template body was spliced in verbatim). The listing renders as an empty `<div id=\"guide-sections\"></div>` instead. site-asset-check's `<%= item.link %>` broken-reference group is gone entirely — 214 broken refs across 8 distinct names, down from 221 across 9.\n\nSo the silent-and-reader-visible failure is now loud-and-safe. Note the diagnostics are warnings, not errors: the render summary is unchanged at 97 of 98 files, 2 errors (both the unrelated download.qmd), with warnings 1717 -> 1739 (+22 = 11+11).\n\nSide effect worth knowing: the visible-text diff against Q1 goes slightly UP on the affected pages (features 47->51, ai-features 12->13, welcome 12->13, tutorials 11->12, assistant-getting-started 6->7). That is not a regression — it is the numbers becoming honest. Previously the dumped EJS contributed junk text lines that coincidentally offset some of Q1's real card lines; with the listing skipped the diff is now the clean statement \"Q1 has N cards here, q2 has none\". The same observation was made independently while porting the templates."}}}
-{"id":"bd-cvk7jb11","title":"xtask: stage-doc-examples (and other repo_root() callers) target the main checkout when run from a worktree","description":"repo_root() in crates/xtask/src/create_worktree.rs resolves the root via `git rev-parse --git-common-dir`'s parent, i.e. the MAIN checkout. That is right for create-worktree (worktrees live under the main repo's .worktrees/), but stage-doc-examples uses it too: run from .worktrees/<x>/, it renders the main checkout's examples/ and stages into the main checkout's docs/examples/, so the worktree's own `q2 render docs` then fails with 'Declared resource .../docs/examples does not exist on disk'. It also dirties the main checkout: the render rewrites the TRACKED examples/lmd/10-project/index.html. Found during bd-73lilznv Phase 4 (2026-10-27). Fix: audit every repo_root() caller; commands that operate on the current checkout's files should resolve `git rev-parse --show-toplevel` instead. Also consider why a rendered output file under examples/ is tracked.","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-10-27T15:34:42.746994Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T15:46:55.300626Z","dependencies":{"bd-73lilznv:discovered-from":{"depends_on_id":"bd-73lilznv","type":"discovered-from","created_at":"2026-10-27T15:34:42.746994Z","created_by":"Carlos Scheidegger"}},"comments":{"c-iho0y73f":{"id":"c-iho0y73f","author":"Carlos Scheidegger","created_at":"2026-10-27T15:39:40.255768Z","text":"Fix committed on braid/bd-cvk7jb11-xtask-stage-doc-examples: 4 callers (stage-doc-examples, check-noop-rebuild, gen-math-spec, braid-snapshot) -> current_worktree_root(); repo_root renamed main_checkout_root. Also fixed switch-task --from branching off stale local branch (resolve_start_point). Verified: stage-doc-examples from a worktree stages into the worktree, main checkout untouched; xtask nextest 249/250 (the 1 failure, check_feature_sets::discovers_the_wasm_consumers_in_this_repo, is pre-existing and worktree-only: pampa/fuzz cargo metadata walks up to the main checkout's Cargo.toml); clippy host + windows-msvc clean."},"c-skd2uqq0":{"id":"c-skd2uqq0","author":"Carlos Scheidegger","created_at":"2026-10-27T15:46:55.300626Z","text":"Second commit 63a2629f0 on PR #932: empty [workspace] in crates/pampa/fuzz, wasm-qmd-parser, tree-sitter-language-wasm-shim (cargo skips an excluding workspace root and walked into the main checkout's from .worktrees/). xtask nextest now 250/250 from a worktree; build:wasm OK."}}}
+{"id":"bd-cvk7jb11","title":"xtask: stage-doc-examples (and other repo_root() callers) target the main checkout when run from a worktree","description":"repo_root() in crates/xtask/src/create_worktree.rs resolves the root via `git rev-parse --git-common-dir`'s parent, i.e. the MAIN checkout. That is right for create-worktree (worktrees live under the main repo's .worktrees/), but stage-doc-examples uses it too: run from .worktrees/<x>/, it renders the main checkout's examples/ and stages into the main checkout's docs/examples/, so the worktree's own `q2 render docs` then fails with 'Declared resource .../docs/examples does not exist on disk'. It also dirties the main checkout: the render rewrites the TRACKED examples/lmd/10-project/index.html. Found during bd-73lilznv Phase 4 (2026-10-27). Fix: audit every repo_root() caller; commands that operate on the current checkout's files should resolve `git rev-parse --show-toplevel` instead. Also consider why a rendered output file under examples/ is tracked.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-10-27T15:34:42.746994Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T16:49:36.838749Z","closed_at":"2026-10-27T16:49:36.838749Z","dependencies":{"bd-73lilznv:discovered-from":{"depends_on_id":"bd-73lilznv","type":"discovered-from","created_at":"2026-10-27T15:34:42.746994Z","created_by":"Carlos Scheidegger"}},"comments":{"c-4o4jz55o":{"id":"c-4o4jz55o","author":"Carlos Scheidegger","created_at":"2026-10-27T16:49:34.111592Z","text":"PR #932 merged (all 12 checks green)."},"c-iho0y73f":{"id":"c-iho0y73f","author":"Carlos Scheidegger","created_at":"2026-10-27T15:39:40.255768Z","text":"Fix committed on braid/bd-cvk7jb11-xtask-stage-doc-examples: 4 callers (stage-doc-examples, check-noop-rebuild, gen-math-spec, braid-snapshot) -> current_worktree_root(); repo_root renamed main_checkout_root. Also fixed switch-task --from branching off stale local branch (resolve_start_point). Verified: stage-doc-examples from a worktree stages into the worktree, main checkout untouched; xtask nextest 249/250 (the 1 failure, check_feature_sets::discovers_the_wasm_consumers_in_this_repo, is pre-existing and worktree-only: pampa/fuzz cargo metadata walks up to the main checkout's Cargo.toml); clippy host + windows-msvc clean."},"c-skd2uqq0":{"id":"c-skd2uqq0","author":"Carlos Scheidegger","created_at":"2026-10-27T15:46:55.300626Z","text":"Second commit 63a2629f0 on PR #932: empty [workspace] in crates/pampa/fuzz, wasm-qmd-parser, tree-sitter-language-wasm-shim (cargo skips an excluding workspace root and walked into the main checkout's from .worktrees/). xtask nextest now 250/250 from a worktree; build:wasm OK."}}}
 {"id":"bd-cwk7l4dl","title":"xtask lint: flag unwrap_or_default() on SourceInfo-typed expressions","description":"quarto-source-map deprecates SourceInfo::default() via an inherent method that shadows Default::default(), and its doc comment says the residual unwrap_or_default() / #[derive(Default)] hole is 'caught by separate grep tooling'.\n\nThat tooling does not exist. crates/xtask/src/lint/ has exactly two rules (external_sources.rs, metadata_as_str.rs), neither related, and there is no CI grep either. The hole was load-bearing in bd-2mxo: materialize.rs reached for unwrap_or_default() at all three container-span sites, producing Original{FileId(0), 0..0} -- a well-formed span indistinguishable from a real location at the first byte of file 0.\n\nThose specific sites are now fixed, so this is prevention rather than an active bug.\n\nDeferred from bd-2mxo's Phase B because doing it accurately needs type information a grep-level rule lacks: the receiver of .unwrap_or_default() has to be known to be Option<SourceInfo>. Options worth weighing: (a) heuristic grep on identifiers named *source_info*/*key_source*, accepting false negatives; (b) a clippy lint via dylint; (c) skip the lint and instead remove/redefine the Default impl upstream (see claude-notes/scratch/2026-08-06-memo-quarto-source-map-default-sourceinfo.md), which makes the lint unnecessary. (c) is probably the better buy.\n\nRelated: claude-notes/designs/provenance-contract.md sec.10 already declares SourceInfo::default() a bug everywhere outside the JSON reader and test scaffolding.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-06T17:11:20.481170Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-06T17:11:30.156735Z","dependencies":{"bd-2mxo:discovered-from":{"depends_on_id":"bd-2mxo","type":"discovered-from","created_at":"2026-08-06T17:11:20.481170Z","created_by":"Carlos Scheidegger"},"bd-61cd:parent-child":{"depends_on_id":"bd-61cd","type":"parent-child","created_at":"2026-08-06T17:11:30.156727Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-cx1det1y","title":"Backfill error catalog entries + docs pages for Q-2-36, Q-2-37, Q-2-38","description":"The error-corpus entries Q-2-36 (old-style knitr chunk options), Q-2-37 (line break in link destination), and Q-2-38 (unclosed attribute specifier) fire at runtime but are absent from both crates/quarto-error-catalog/error_catalog.json and docs/errors/markdown/ (pages exist for Q-2-1..Q-2-35, Q-2-39, Q-2-40). docs/errors/README.md says the catalog defines which codes exist and every code gets a per-subsystem page, so these three are lapses. Q-2-36 skipped docs deliberately at ship time (claude-notes/plans/2026-05-14-q-2-36-knitr-style-chunk-options.md Phase 4) before the convention firmed up. Fix: add three catalog entries (subsystem markdown, title matching the corpus JSON title, message_template from the corpus message, docs_url per convention, since_version 99.9.9 placeholder) and three docs pages per the README template, status stub. Discovered while adding Q-2-41 (bare-brace escape hint), which registers catalog+docs from the start.","status":"closed","priority":2,"issue_type":"chore","created_at":"2026-08-09T23:38:26.725276Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-10T14:57:26.576234Z","closed_at":"2026-08-10T14:57:26.576234Z","close_reason":"Backfilled catalog entries + docs pages for Q-2-36/37/38, with TDD presence tests and cross-ref promotions. PR #485 merged.","labels":["diagnostics"],"dependencies":{"bd-brace-escape-hint-0tmemkyt:discovered-from":{"depends_on_id":"bd-brace-escape-hint-0tmemkyt","type":"discovered-from","created_at":"2026-08-09T23:38:26.725276Z","created_by":"Carlos Scheidegger"}},"comments":{"c-810yqbpe":{"id":"c-810yqbpe","author":"Carlos Scheidegger","created_at":"2026-08-10T14:29:32.541700Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/485 (branch chore/bd-cx1det1y-error-docs-backfill @ 2ef8b493, full verify green, audit clean). Close when merged."}}}
 {"id":"bd-cxara","title":"Phase 9 — end-to-end verification","description":"CRITICAL per CLAUDE.md: tests passing alone is insufficient. Phase 9 runs the full verification matrix and records exact invocations + observed output in the plan's Verification log.\n\nCoverage: full cargo xtask verify; SPA cookie-path regression; clean-machine device-flow E2E (with per-platform credential-store inspection); no-auth-hub regression + explicit-authenticate short-circuit + observation-requirement check; allowlist parity for cookie 403 and bearer 403 (byte-identical detail); insecure-bearer dev-mode interaction (loopback + non-loopback with/without env flag); force-refresh; force-reauth; dual-credential CVE + WS-upgrade-with-no-Origin sanity check; audit-log spot-check; plaintext-leak grep.\n\nPlan §Phase 9: claude-notes/plans/2026-05-05-hub-mcp-device-flow-implementation.md","status":"closed","priority":1,"issue_type":"task","created_at":"2026-05-20T14:27:31.218633Z","created_by":"shikokuchuo","updated_at":"2026-07-27T19:43:30.967+00:00","closed_at":"2026-07-27T19:43:30.967+00:00","dependencies":{"bd-cmp48:parent-child":{"depends_on_id":"bd-cmp48","type":"parent-child","created_at":"2026-05-20T14:27:31.218633Z","created_by":"shikokuchuo"}}}
@@ -764,6 +769,7 @@
 {"id":"bd-dhtw","title":"Phase 1: gh-pages provider end-to-end","description":"Phase 1 of bd-t3ny. Implement the gh-pages provider end-to-end on top of the Phase 0 scaffolding: common::git wrappers, common::github context discovery, _publish.yml reader, GhPagesProvider (prepare/commit/verify with .nojekyll poll), --dry-run cleanup with no residue, mkdocs-style summary, and an end-to-end test against a bare local remote (dry-run + real-run + json-run).\n\nPlan: claude-notes/plans/2026-05-03-publish-command-and-gh-pages.md (Phase 1 section)\n\nBlocked on Phase 0 (bd-068k).","status":"closed","priority":1,"issue_type":"task","created_at":"2026-05-03T14:37:39.274734Z","created_by":"cscheid","updated_at":"2026-05-03T15:18:38.648426Z","dependencies":{"bd-068k:blocks":{"depends_on_id":"bd-068k","type":"blocks","created_at":"2026-05-03T14:37:39.274734Z","created_by":"cscheid"},"bd-t3ny:parent-child":{"depends_on_id":"bd-t3ny","type":"parent-child","created_at":"2026-05-03T14:37:39.274734Z","created_by":"cscheid"}}}
 {"id":"bd-dia333nt","title":"Document q2's metadata merging discipline (dev-doc + user-facing guide)","description":"q2 resolves metadata by merging layers uniformly per document — project _quarto.yml < extension contributions < _metadata.yml chain (root down to the document's dir) < front matter < -M runtime metadata — in stage/stages/metadata_merge.rs over metadata_layers::collect_metadata_layers, with format flattening (format.<fmt>.* over top-level within each layer) and the tag-based merge semantics of claude-notes/plans/2025-12-07-config-merging-design.md (maps merge field-wise, arrays concatenate, !prefer etc.). The design intent: a top-level key in _quarto.yml is indistinguishable from every document declaring it, so directory/page scoping via _metadata.yml / front matter works for any key without special cases. Neither the discipline nor its practice is written down.\n\nDeliverables:\n1. Dev-doc (dev-docs/ or claude-notes/instructions/): the layer order and merge semantics; the rule for consumers — read per-document merged metadata (ast.meta, or a value folded into DocumentProfile in Pass 1 for cross-page consumers), never raw project.config.metadata, which silently ignores directory/front-matter overrides (several post-render readers in project/website_config.rs still do; audit them); path values resolve relative to the declaring YAML file via SourceInfo (bd-qor9a, navigation_href::resolve_metadata_path); website keys with two spellings go through quarto_config::resolve_website_value.\n2. User-facing guide under docs/guides/projects/: how _quarto.yml, _metadata.yml, front matter and -M combine; precedence; arrays vs maps; !prefer; relative paths.\n\nKnown wrinkle to document (surfaced in bd-2hy80x1j's plan, Q-a): resolve_website_value lets the top-level spelling beat website.<key> AFTER layers merge, so a top-level key in _quarto.yml beats the website.<key> spelling in a higher-precedence _metadata.yml or front matter. Existing behaviour for navbar/sidebar/page-footer; document it, and decide whether it should be fixed (e.g. normalize spellings per layer before merging).","status":"open","priority":2,"issue_type":"docs","created_at":"2026-10-22T12:59:29.638655Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-22T12:59:29.638655Z","labels":["metadata"],"dependencies":{"bd-2hy80x1j:discovered-from":{"depends_on_id":"bd-2hy80x1j","type":"discovered-from","created_at":"2026-10-22T12:59:29.638655Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-digit-line-splits-paragraph-w6tod0gh","title":"Continuation line starting with a digit terminates the paragraph (fatal inside link text)","description":"WHAT HAPPENS. A soft-wrapped paragraph line whose first character is a digit terminates the paragraph. Two paragraphs come out where the author wrote one, splitting a sentence mid-clause:\n\n  input                                            q2 0.17.0                              Quarto 1\n  \"...license leases last\\n30 minutes you...\"      <p>...leases last</p><p>30 minutes...   one <p>, text intact\n  \"...greater than\\n1000, the base filter...\"      <p>...greater than</p><p>1000, the...   one <p>, text intact\n  \"...leases last\\nthirty minutes you...\"          one <p> (correct)                       one <p>\n\nIt fires on ANY digit-leading continuation line, whatever follows. Verified against '30 minutes', '1000,', '1 apple', '9', '3.14', '0', and a 14-digit run — longer than the 9-digit maximum CommonMark allows in an ordered-list marker, so no marker could possibly form. Only a non-digit first character ('x30') stays in the paragraph. This is a line-start character class, not a mis-parse of a list marker.\n\nFATAL VARIANT. When the wrap falls inside link text, the block ends before the closing ']' and the render fails outright:\n\n  The set of characters that must be encoded can be found in\n  [Section 3.2.1 of RFC\n  3986](https://www.rfc-editor.org/rfc/rfc3986#section-3.2.1).\n\n  Error: [Q-2-1] Unclosed Span\n    I reached the end of the block before finding a closing ']'\n\nAuthors cannot reasonably be expected to avoid wrapping before a number, and the failure gives no hint that the line break is what caused it.\n\nWHERE IT LIVES. crates/tree-sitter-qmd/tree-sitter-markdown/src/scanner.c, in the two SOFT_LINE_ENDING gates of the main scan function (first gate ~line 2739, second ~line 2878). Both conditions end with:\n\n    first_lookahead > ' ' && !(first_lookahead >= '0' && first_lookahead <= '9')\n\nA digit-leading continuation line fails both gates, falls through to the LINE_ENDING branch just below them, and LINE_ENDING terminates the paragraph.\n\nThe asymmetry with '*' in the very same condition is the tell. For '*' the scanner peeks first — first_starts_with_star_block / second_starts_with_star_block (~2690-2708, ~2840-2862) measure the delimiter run and require trailing whitespace or EOL, so the soft break is blocked only when a list marker or thematic break really forms. The digit branch has no equivalent peek; it is a blanket exclusion of the character class. ('-' and '+' are blanket-excluded too, but harmlessly: a '-'-leading continuation line genuinely can open a list.)\n\nNote that parse_ordered_list_marker (~line 1195) is already correct and already implements the CommonMark interruption rule — it requires digits followed by '.' or ')', caps at 9 digits, and sets dont_interrupt for anything but a leading '1'. It just never gets consulted, because the paragraph has already been ended by the time it would matter.\n\nSUGGESTED FIX DIRECTION. Mirror the star handling: add a first_starts_with_ordered_marker / second_starts_with_ordered_marker peek that consumes the digit run (bailing past 9 digits), then requires '.' or ')' followed by a space or EOL, and — to match parse_ordered_list_marker's own dont_interrupt rule — only counts as paragraph-interrupting when the run is a single '1'. Gate the digit exclusion on that flag instead of on the raw character class. Everything else keeps working, since the marker path downstream is unchanged.\n\nWorth settling in the same pass: section D of the repro is a genuine '1. apples' interrupting a paragraph after 'Buy these:'. q2 opens a real <ol>; Quarto 1 keeps one paragraph, because Pandoc's markdown reader forbids list interruption entirely while CommonMark permits it for '1.'. Defensible either way, but the two behaviors come from the same rule and a fix here will touch it.\n\nREAL-WORLD HIT (Posit Connect docs, 8+ pages). Sentences split mid-clause wherever a version number, port, timeout or RFC number happened to land at the start of a wrapped line: admin/licensing ('...to make license leases last' / '30 minutes you would use the following syntax', and again for lease renewal), admin/nodejs (twice, on Node version numbers), admin/process-management ('e.g. Red Hat Enterprise Linux' / '7)'), admin/appendix/advanced-ldap ('greater than' / '1000,'), admin/appendix/cli (RFC-3339 timestamp formats), user/troubleshooting (Python version numbers). admin/database/postgres hit the fatal variant inside a callout body; the page only builds today because the porting session hand-escaped the number as '\\3986', which then renders as a visible backslash — it has since been rewrapped instead.\n\nREPRO. /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/digit-line-splits-paragraph/ — run 'q2 render' there and compare with 'quarto render'. index.qmd has sections A (digits), B (digits + punctuation), C (word-leading control) and D (a real ordered list); link-across-lines.qmd is the fatal variant and is the file that errors. README.md carries expected vs actual output and the character-class evidence above. That repo is local-only, hence the absolute path.\n\nOrigin strand in the connect-docs porting skein: br-mtg8ts05. Observed with q2 0.16.0 and re-verified failing on 0.17.0 / origin/main @ 001cb6a5, where nothing has touched the soft-break gates.","status":"in_progress","priority":1,"issue_type":"bug","created_at":"2026-08-11T15:29:29.291635Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-11T16:26:52.994717Z","labels":["bug","parity"],"comments":{"c-ilyz8vr9":{"id":"c-ilyz8vr9","author":"Carlos Scheidegger","created_at":"2026-08-11T15:42:35.993605Z","text":"Investigated at fc2895b2 (pre-flight `cargo xtask verify --skip-hub-build` clean). Plan skeleton + committed repro fixtures: claude-notes/plans/2026-08-11-digit-line-splits-paragraph.md (commit 813ad059). Verdict: ready to design, pending 4 questions in the plan.\n\nRoot cause confirmed exactly as filed — scanner.c gates at 2750 and 2889-2890. Four things the investigation added:\n\n1. THE REPO ALREADY CARRIES A HAND-ESCAPED WORKAROUND FOR THIS BUG. test/corpus/new-spec.txt:2039 is CommonMark Example 284 with the input rewritten from '14.' to '\\14.' and the title \"qmd: start your soft breaks with \\1, i'm sorry.\" A prior session hit this, couldn't fix it, and encoded the workaround into the conformance suite — the same escape hatch the connect-docs porting session independently invented ('\\3986'). Fed the real spec input, q2 doesn't merely split the paragraph, it invents <ol start=\"14\">. Restoring that test to unescaped spec input is the natural acceptance criterion.\n\n2. The \"worth settling in the same pass\" item (section D, '1.' interrupting) needs no new decision. q2 already matches CommonMark and diverges from pandoc-markdown on '-', '1.' AND '1)'. Adopting Q1's never-interrupt rule would mean changing '-' too — a separate, larger call. The proposed fix lands all six interruption-matrix rows on CommonMark.\n\n3. The fix repairs '2. apples' for free. Today it opens a spurious <ol start=\"2\"> — wrong under BOTH specs — because the blanket exclusion closes the paragraph before parse_ordered_list_marker runs, making its correct dont_interrupt logic dead code on this path. Worth calling out in the eventual commit as a behavior change beyond the reported symptom.\n\n4. The '-' and '+' exclusions are NOT harmless, contra the description. \"Temperature dropped to\" / \"-5 degrees overnight.\" splits identically, as does \"+5 percent\". Same bug, different character class, same fix shape. Open question in the plan is whether to fold them into this strand.\n\nAnd ':' does something worse than any of the above — it silently DELETES the colon. \"Defined at\" / \":host scope.\" parses to [Str \"at\", SoftBreak, Str \"host\"]; the character is gone with no diagnostic. Since one paragraph survives, it is being consumed beyond the gate (likely the ':::' fenced-div or definition-list path), so the gate fix will not necessarily cover it. Filed as bd-cxiopjw7 (discovered-from this strand).\n\nBlast radius on existing tests is small: scanning all 50 corpus files for a digit-leading line preceded by a text line yields only Example 285 (must stay interrupting) and a display-math case in issues.txt. Main risk is downstream snapshot churn, since this changes block structure. Full `cargo xtask verify` (not --skip-hub-build) required — the grammar feeds the WASM parser."},"c-kwys52a9":{"id":"c-kwys52a9","author":"Carlos Scheidegger","created_at":"2026-08-11T16:26:52.994717Z","text":"Implemented on branch braid/bd-digit-line-splits-paragraph-w6tod0gh-continuation-line-starting-digit, commit 92737cdd. Plan updated with the full implementation record. NOT pushed — awaiting review.\n\nWHAT LANDED. scanner.c only; no grammar change, so parser.c is untouched and there are no generated diffs. Two peek helpers mirroring the existing '*' peek: peek_dash_plus_opens_block, and peek_ordered_marker returning {well_formed, may_interrupt} with a bail past 9 digits.\n\nTHE SUBTLE PART (two failed attempts, both recorded in code comments so nobody retries them):\n\nThe two gates need DIFFERENT answers from the same peek. Gate 1 runs before match_line, so it cannot distinguish a sibling list item from a paragraph continuation and has no all_will_be_matched guard — it may only ask well_formed. Gate 2 runs after match_line, where all_will_be_matched proves the line belongs to the open blocks, and is the only place the CommonMark interruption rule applies.\n\nAttempt 1 applied the \"only a bare 1 interrupts\" rule in both gates. That broke six existing tests — issues.txt #72, list.txt 4/15/16, and CommonMark Examples 261/282 — all sibling list items like \"1. a / 2. b\", where gate 1 swallowed '2.' into item 1's paragraph.\n\nAttempt 2 consulted valid_symbols[LIST_MARKER_*] the way parse_ordered_list_marker does. That fails because at line-ending scan time the parser is asking for LINE_ENDING / SOFT_LINE_ENDING, so the marker symbols are not in the valid set and read false for every marker. It broke '1.' interruption (Example 285 and two new tests).\n\nAlso: the mark_end guard after gate 2 was keyed on the character (!= '`' && != '*'). With three more peeking branches that proxy no longer holds, so it is now an explicit second_peeked boolean — otherwise a peeked run gets swallowed into the SOFT_LINE_ENDING token range.\n\nA THIRD ESCAPED WORKAROUND, found during implementation. Beyond new-spec.txt Example 284, block_quote.txt test 4 carried '> 2 / \\1' with the title \"softbreaks ignore block structure; but don't start continuation line with 1\" — a lazy-continuation case. Restored to '> 2 / 1'. All three escaped inputs (284, block_quote 4, and 284's expected tree) failed before the fix and pass after.\n\nBEHAVIOR CHANGES BEYOND THE REPORTED SYMPTOM, batched per the user's instruction that such changes are better learned once:\n- '2.' / '14.' after a paragraph no longer open a spurious <ol start=N> (was wrong under BOTH specs).\n- '-5 degrees', '+5 percent', '-- em dash' stay in the paragraph.\n\nRESULTS. tree-sitter test 572/572 (12 confirmed failing first, per TDD). cargo nextest run --workspace: 11626 passed, 197 skipped. cargo xtask verify full including the WASM leg: green. ZERO snapshot files changed — the churn flagged as the main risk did not materialize.\n\nEND-TO-END. cargo run --bin q2 -- render doc.qmd, HTML inspected: the digit continuation is one <p>, the link that previously failed with Q-2-1 renders intact, a real ordered list still works. pampa -t native now matches pandoc -f commonmark -t native EXACTLY on the committed interrupt.qmd (all six matrix rows) and dashplus.qmd — including Str \"+5\", confirming the two pandoc_str CST nodes for '+5' (pre-existing '+' tokenization, present mid-line too) merge into one Str at AST level.\n\nOUT OF SCOPE, left alone as agreed: the ':host' colon-dropping bug (bd-cxiopjw7) still reproduces.\n\nINCIDENTAL, NOT COMMITTED: running the WASM build refreshed crates/wasm-quarto-hub-client/Cargo.lock from 0.16.0 to 0.17.0 version strings — pre-existing staleness after the 0.17.0 bump, unrelated to this fix. Reverted to keep the commit focused; worth a separate cleanup. npm install also rewrote package-lock.json peer-dependency metadata; also reverted."}}}
+{"id":"bd-diu7rtes","title":"Lua API: quarto.doc.input_file / quarto.doc.output_file are missing (GH #940)","description":"Quarto 1 exposes quarto.doc.input_file and quarto.doc.output_file (absolute paths) to Lua filters; q2's quarto.doc has neither, PANDOC_STATE is nil, and the working directory is the project root for every page, so a filter cannot learn which file it is processing. Reproduced 2026-10-27 on main at 3f7d4b0e8: both fields nil in a website render.\n\nFix: register_quarto_doc reads the new LuaOptions.host (HostContext: input_file, output_file) that quarto-core fills from ctx.document.input and ctx.output_path() in UserFiltersStage and ShortcodeResolveTransform. Set the fields in both the filter and the shortcode environments. Also (agreed 2026-10-27) a minimal PANDOC_STATE global in both environments: { input_files = pandoc.List{input_file}, output_file = output_file } (absolute path, which is what Q1's outputFile() produced after joining anyway); the other pandoc fields stay nil deliberately so a filter that needs one fails at its own line instead of getting a plausible wrong default.\n\nDepends on the HostContext plumbing landing with bd-ip1lrgra (#939).\n\nPlan: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-10-27T20:22:10.405327Z","created_by":"claude","updated_at":"2026-10-27T21:54:41.309998Z","external_ref":"https://github.com/quarto-dev/q2/issues/940","labels":["lua","pampa","parity"],"dependencies":{"bd-ip1lrgra:blocks":{"depends_on_id":"bd-ip1lrgra","type":"blocks","created_at":"2026-10-27T20:22:58.301004Z","created_by":"claude"},"bd-pqj37chw:parent-child":{"depends_on_id":"bd-pqj37chw","type":"parent-child","created_at":"2026-10-27T20:22:10.405327Z","created_by":"claude"}},"comments":{"c-9v8nm4qg":{"id":"c-9v8nm4qg","author":"claude","created_at":"2026-10-27T21:54:41.309998Z","text":"PR: https://github.com/quarto-dev/q2/pull/949 (feature/lua-api-q1-parity -> main; closes GH #939-#942 on merge). Full cargo xtask verify green incl. hub-client tests on a rebuilt WASM; Windows cross-check not possible from macOS, flagged in the PR."},"c-lz676vpp":{"id":"c-lz676vpp","author":"claude","created_at":"2026-10-27T21:31:47.527568Z","text":"Implemented and merged into feature/lua-api-q1-parity: quarto.doc.input_file / output_file and the minimal PANDOC_STATE {input_files, output_file}, set from the HostContext in crates/pampa/src/lua/host_context.rs for both the filter and shortcode environments. Unit tests + the lua_host_context_cli e2e test extended. GH #940 repro now prints both absolute paths. Left open until the epic's PR merges."}}}
 {"id":"bd-div-attr-no-space-ne0fudkw","title":"Fenced div with no whitespace after the colons is silently dropped (and with a closing fence becomes a cascading uncoded parse error)","description":"A fenced div whose colons are not followed by whitespace is not parsed as a div. Quarto 1 and Pandoc both accept the tight form. What q2 does next depends only on whether a closing fence is present, and both outcomes are bad.\n\nWITHOUT a closing fence, q2 renders CLEAN, exit 0, ZERO diagnostics — and emits no div at all. Given ':::{.myclass}' followed by body text, 'myclass' is absent from the output entirely and the body survives as bare prose. The opening line is consumed: it does not appear as text, which it would if the parser had given up and treated it as a paragraph. It is consumed, and then the div and its whole attribute block are discarded silently. A callout becomes plain prose with no wrapper, no classes, and no warning. This is the serious half — silent content loss is worse than a fatal error, because an error at least tells the author the page is wrong.\n\nWITH a closing fence, it surfaces as an uncoded 'Error: Parse error … unexpected character or token here' reported at the CLOSING ':::' rather than at the opening line that actually differs, so following the error leads to the wrong place in the source. It also cascades: the committed repro produces 3 errors from one cause, the later ones landing inside a json code fence on '{' characters. Those are not code-block bugs — the control file has the same fence after a spaced div and parses cleanly. In the Positron port this cascade produced errors pointing at JSON and JavaScript inside code fences, which reads like 'q2 parses markdown inside code blocks' and is not that at all.\n\nIT IS NOT ABOUT THE ATTRIBUTE BLOCK. A bare info string with no braces fails identically, so the requirement is whitespace after the colons regardless of what follows:\n  :::{.foo}    q2: 1 error              Q1: <div class=\"foo\">\n  :::foo       q2: 1 error              Q1: <div class=\"foo\">\n  ::: {.foo}   q2: <div class=\"foo\">    Q1: same\n  ::: foo      q2: <div class=\"foo\">    Q1: same\nQuarto 1 (99.9.9) accepts all four.\n\nAND q2 ALREADY DOES THIS CORRECTLY FOR CODE FENCES, which is the strongest single piece of evidence here. The same lexical job — fence marker, optional whitespace, then either an attribute block or a bare word — on a code fence in the same grammar file: three-backtick forms with an attribute block and no space, with an attribute block and a space, and with a bare word and no space are ALL accepted and all emit <pre class=\"foo\">. So 'no space after the marker' is not intrinsically hard for this parser; it is solved a few hundred lines away in the same file. There appears to be no design reason for divs to differ.\n\nWHERE THE CAUSE IS NOT, recorded so it is not re-derived. The grammar: pandoc_div (crates/tree-sitter-qmd/tree-sitter-markdown/grammar.js:975-983) declares optional($._whitespace) between $._fenced_div_start and the info-string-or-attribute choice — the tight form appears permitted. The scanner: crates/tree-sitter-qmd/tree-sitter-markdown/src/scanner.c:644-652 consumes spaces and tabs after the colons and then decides FENCED_DIV_END vs FENCED_DIV_START purely on whether the next character is a newline, logic that is whitespace-count-independent and behaves the same with zero spaces. Both appear to permit the tight form, so the failure is downstream of both. Where exactly has NOT been traced and no mechanism claim is made.\n\nSUGGESTED FIX DIRECTION. The plausible reading is that choice(alias($._commonmark_naked_value, $.info_string), alias($._pandoc_attr_specifier, $.attribute_specifier)) at grammar.js:978 resolves differently when no whitespace precedes it, which would explain both the lost attributes and the desynchronised closer — that is a hypothesis to check first, not a spec. Whatever the mechanism, the acceptance criteria are: all four spellings above produce <div class=\"foo\">; the silent-drop path stops existing; and if any input is still to be rejected it must be rejected with a coded diagnostic pointing at the OPENING line, not the closer. The code-fence path is the reference implementation to compare against.\n\nDO NOT FOLD THIS TOGETHER WITH bd-shortcode-no-space-gspu6l5s. The two look like the same bug and are not. There _shortcode_sep (grammar.js:650) is a choice with NO empty alternative, so the separator is mandatory by construction and rejecting the tight shortcode form is deliberate — that strand asks only for a better diagnostic and explicitly must not loosen the grammar. Here the grammar says optional and content is being lost silently. Different defects, different fixes.\n\nReal-world hit in the Positron docs: 7 of the 19 pages that failed to render on the first full pass — guide-r.qmd, guide-r-debugging.qmd, guide-r-air.qmd, extensions.qmd, extension-development.qmd, quarto.qmd, data-explorer-summary-panel.qmd. They render today only because the source was migrated by hand, so the cost now falls on other Quarto 1 projects during migration.\n\nMinimal repro: /Users/gordon/src/q2-positron-docs/llms-info/repros/div-attr-no-space/ — run 'sh check.sh' there (it takes an optional path to a q2 binary, defaulting to ~/.local/bin/q2). Five fixtures: repro.qmd (tight opener plus closer, 3 errors), control.qmd (spaced, renders the callout), infostring.qmd (bare ':::foo', 1 error, no braces involved), silent.qmd (tight opener, no closer — clean exit 0 and NO BLOCK EMITTED), codefence.qmd (the same shape on a code fence, all three variants accepted).\n\nOrigin strand in the Positron-docs porting skein: br-div-attr-no-space-itqfay4p. Observed with q2 0.28.0 (official release build); re-checked against q2 HEAD 103af4445, where neither grammar.js nor scanner.c has changed since v0.28.0 and pandoc_div still declares optional($._whitespace).","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-09-01T20:38:24.641999Z","created_by":"Gordon Woodhull","updated_at":"2026-09-03T20:39:14.394753Z","closed_at":"2026-09-03T20:39:14.394753Z","close_reason":"Shipped in q2 0.29.0. Verified downstream by re-running llms-info/repros/div-attr-no-space/check.sh: all five fixtures render clean and emit their block, including :::{.callout-note} with a closer (was 3 errors), :::foo (was 1 error) and the silent-loss case :::{.myclass} with no closer (was clean-exit with the block missing).","labels":["diagnostics","parser"],"comments":{"c-4un7iydu":{"id":"c-4un7iydu","author":"Gordon Woodhull","created_at":"2026-09-01T22:46:42.687436Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/639 (branch bugfix/bd-div-attr-no-space-ne0fudkw-div-attr-no-space, commit 5c815d158)."},"c-c0zht23e":{"id":"c-c0zht23e","author":"Gordon Woodhull","created_at":"2026-09-01T21:32:33.018184Z","text":"ROOT CAUSE (traced, not hypothesised). Not the grammar and not parse_fenced_div_marker — both were correctly cleared in the strand. `tree-sitter parse` shows :::{.foo} parsing to a clean pandoc_div when it is the FIRST thing in the document, and failing only when a paragraph or metadata block precedes it. That narrowed it to the soft-line-break path in scanner.c's scan().\n\nThe `:::`-peek at scanner.c:2858 (added for issue #206, pipe-table termination) walks the whole colon run with advance() and never restores the lexer. The two SOFT_LINE_ENDING gates below it then read `lexer->lookahead` — which by then is the character AFTER the colons. For `::: {.foo}` that is a space, so `first_lookahead > ' '` fails and the gate correctly declines, falling through to LINE_ENDING (whose token range is still at post-newline, so the colons survive for the next scan). For `:::{.foo}` it is `{`: every gate test passes, `first_peeked` is false, so the gate's `lexer->mark_end(lexer)` fires at the CURRENT position — past the colons. The debug trace shows it exactly: `lexed_lookahead sym:_soft_line_ending, size:4` for a token that should be size 1. The `:::` is inside the soft-break token, so the grammar never sees a div opener, only a stray attribute_specifier mid-paragraph.\n\nThe hypothesis named in the strand (info_string vs attribute_specifier choice at grammar.js:978) is DISCARDED — that choice resolves fine; the tokens never reach it.\n\nSECOND BUG, same cause: the swallow also ate colon runs that open nothing. `:hello` as a paragraph continuation rendered as `hello`, `::world` as `world` (verified against the 0.28.0 release binary). Same silent-content-loss class, wider blast radius than divs.\n\nFIX (committed on braid/bd-div-attr-no-space-ne0fudkw-div-attr-no-space, 5c815d158, NOT pushed). The peek records whether it consumed a colon run and whether that run opens a block (>=3 colons = fenced div whatever follows; 1 colon + whitespace = caption start; anything else = ordinary text). Both gates test that flag instead of comparing the corrupted lookahead against ':', and take the PEEKED emission path so mark_end cannot absorb the run. The fenced-div test also loses its \"followed by whitespace\" requirement.\n\nVERIFIED. check.sh in the committed repro: all five fixtures clean, repro/control/infostring/silent all emit their div (was 3 errors / clean / 1 error / NO BLOCK EMITTED). Also checked by hand: nested tight divs, tight div in a blockquote, tight div in a list item, tight div immediately after a pipe-table row. tree-sitter test 613/613. cargo nextest run --workspace 13490 passed / 0 failed (baseline on main lists 13483; delta is exactly the 7 tests added). clippy -D warnings clean, cargo xtask lint clean, cargo xtask verify (incl. WASM + hub-client) all steps passed.\n\nOUT OF SCOPE, unchanged, pre-existing: a genuinely malformed attribute block (`:::{.foo` with no closing brace) still produces an uncoded \"unexpected character or token here\" pointing at the line after the opener. That is identical for `::: {.foo` and so is not a whitespace divergence — worth its own strand if we want it coded.\n\nDid not touch the shortcode path (bd-shortcode-no-space-gspu6l5s)."},"c-z5h5zbal":{"id":"c-z5h5zbal","author":"Gordon Woodhull","created_at":"2026-09-01T20:39:36.177184Z","text":"Dispatched to agent fenced-div-no-spaces on workspace-7 (branch braid/bd-div-attr-no-space-ne0fudkw-div-attr-no-space, created fresh from main at 103af4445). Origin strand br-div-attr-no-space-itqfay4p in the Positron-docs porting skein now carries external_ref q2:bd-div-attr-no-space-ne0fudkw.\n\nThe agent was pointed at codefence.qmd as the reference implementation — the code-fence path in the same grammar accepts all three shapes this one rejects — and told the grammar.js:978 choice-resolution reading is a hypothesis to verify or discard, not a diagnosis. It was also warned not to fold this with bd-shortcode-no-space-gspu6l5s (in progress on workspace-6), since the correct fixes are opposite: that one must not loosen its grammar, this one must."}}}
 {"id":"bd-dje564or","title":"Sidebar auto: expansion is one level deep; Q1 nests arbitrarily","description":"When a sidebar 'auto:' spec (or the equivalent 'contents: <dir>' shorthand) matches documents nested more than one directory deep, q2 emits them as flat links inside the top directory's section. Q1 builds an arbitrarily deep tree of nested sections.\n\nExample: with docs/top.qmd and docs/deep/nested.qmd, 'auto: docs' yields a single 'Docs' section whose children are two flat links — one of them labelled from docs/deep/nested.qmd — rather than a 'Docs' section containing a nested 'Deep' sub-section.\n\nCause: group_with_subdirs (crates/quarto-core/src/transforms/sidebar_auto.rs) partitions candidates by their FIRST path component only, and section_for_dir builds its children with link_entry, never recursing. Q1's nodesToEntries (external-sources/quarto-cli/src/project/types/website/website-sidebar-auto.ts) recurses over the full split path.\n\nThis predates bd-sidebar-contents-dir-shorthand-z7arvhx8 — that fix routes bare directories through the same group_with_subdirs machinery, so it inherits the depth limit rather than introducing it. Recorded there as a deliberate known limitation and pinned by the test auto_bare_directory_still_matches_nested_documents.\n\nFix would make section_for_dir recursive (build a sub-section per nested directory, using that directory's index.qmd for title/href as the one-level case already does). Priority 3: only bites projects whose auto-generated sidebar directories have sub-directories. Not needed for the Posit Connect docs port, which motivated the parent strand.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-08-13T16:08:48.277616Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-13T16:08:48.277616Z","labels":["navigation"],"dependencies":{"bd-sidebar-contents-dir-shorthand-z7arvhx8:discovered-from":{"depends_on_id":"bd-sidebar-contents-dir-shorthand-z7arvhx8","type":"discovered-from","created_at":"2026-08-13T16:08:48.277616Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-djpt","title":"Ship Bootstrap Icons font + CSS with Quarto 2 HTML output","description":"Quarto 2 HTML renders Bootstrap-icon markup (`<i class=\\\"bi bi-github\\\"></i>`) for navbar tools and footer nav items, but the Bootstrap Icons package (font file + icon CSS that maps class names to glyphs) is not shipped, so the icons render as empty boxes. Quarto 1 handles this via a separate bootstrap-icons dependency.\n\nScope: vendor vs CDN decision (prefer vendor, to match SCSS strategy); where the font file lives in artifact storage; template link (`<link rel=\\\"stylesheet\\\" href=\\\"...bootstrap-icons.css\\\">`); interaction with the planned JS-deps mechanism (bd-ulgr).\n\nCompanion to bd-ulgr (Bootstrap JS). Both issues concern static Bootstrap-ecosystem resources that Q2 needs to ship to match Q1 UX; worth designing them together.\n\nSurfaced during bd-imiw visual testing — navbar/footer DOM is correct, icons are invisible.","status":"open","priority":2,"issue_type":"feature","created_at":"2026-04-18T20:00:13.337791Z","created_by":"cscheid","updated_at":"2026-04-18T20:00:13.337791Z","dependencies":{"bd-imiw:discovered-from":{"depends_on_id":"bd-imiw","type":"discovered-from","created_at":"2026-04-18T20:00:13.337791Z","created_by":"cscheid"},"bd-ulgr:related":{"depends_on_id":"bd-ulgr","type":"related","created_at":"2026-04-18T20:00:13.337791Z","created_by":"cscheid"}}}
@@ -829,7 +835,7 @@
 {"id":"bd-entbg6x3","title":"quarto-math: TeX math AST with mitex parser front end + source maps; OMML and Typst writers","description":"First step of the native docx writer epic (bd-pq9k90z2). A micro-pandoc for math: mitex-parser (Apache-2.0, external-sources/mitex) as the only reader, a q2-owned command spec (JSON: arg shape + OMML semantics + Typst alias, replacing mitex-spec-gen's Typst-generated rkyv artifact), a normalization pass (nested attachments -> sub/sup, env cell splitting, coarse-word splitting for MathML), and writers for OMML (docx) and Typst. Every node carries a quarto_source_map::SourceInfo built by substring from the Math inline's text; macro-expanded tokens keep original spans because mitex tokens borrow &str slices of the input (verified with a probe: expansions point at the definition site, macro args at the use site). Prerequisite in pampa: Math inline text is not a plain substring of its node span on multi-line math (soft-break collapse, gutter strip), so the reader must record a Substring/Concat mapping for the text. Related experiments: pandoc-hybrid leg (bd-fzqykm0n, bd-ymkkrn64) reaches docx/typst via Lua filters + Pandoc. Plan: claude-notes/plans/2026-09-21-quarto-math-and-native-docx.md","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-09-21T15:11:14.221672Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-23T19:28:11.162743Z","closed_at":"2026-09-23T19:28:11.162743Z","close_reason":"Merged to main in PR #706 (e276f50c) on 2026-09-23","dependencies":{"bd-fzqykm0n:related":{"depends_on_id":"bd-fzqykm0n","type":"related","created_at":"2026-09-21T15:11:14.221672Z","created_by":"Carlos Scheidegger"},"bd-pq9k90z2:parent-child":{"depends_on_id":"bd-pq9k90z2","type":"parent-child","created_at":"2026-09-21T15:11:14.221672Z","created_by":"Carlos Scheidegger"},"bd-ymkkrn64:related":{"depends_on_id":"bd-ymkkrn64","type":"related","created_at":"2026-09-21T15:11:14.221672Z","created_by":"Carlos Scheidegger"}},"comments":{"c-29cfh8me":{"id":"c-29cfh8me","author":"Carlos Scheidegger","created_at":"2026-09-21T18:51:20.515703Z","text":"Renumbered the math error subsystem to Q-22 (PR #704 feature/pandoc-writer-hybrid uses Q-20 pandoc, Q-21 typst). Trial merge of #704 into this branch: source files clean incl. shared json.rs/readers/json.rs/test.rs/crossref_render.rs; only Cargo.lock, pampa snapshots/json/math-with-attr.snap and annotated-qmd academic-paper.json conflict, all regenerable. Recipe in the plan header."},"c-56nny7vo":{"id":"c-56nny7vo","author":"Carlos Scheidegger","created_at":"2026-09-21T15:13:43.964817Z","text":"2026-09-21: research done, plan written at claude-notes/plans/2026-09-21-quarto-math-and-native-docx.md on branch braid/bd-entbg6x3-quarto-math. Next: Phase 0 (fixture corpus, pampa Math text mapping, mitex vendoring decision), then Phase 1 crate with tests first."},"c-730mvqad":{"id":"c-730mvqad","author":"Carlos Scheidegger","created_at":"2026-09-21T18:43:22.197969Z","text":"Session 2026-09-21 progress: Phase 0 complete (corpus 263 fixtures; pampa Math.text_source via bd-ieldbghj, merged locally; mitex vendored as 4 crates + spec JSON dump; OOXML schema closure + xmllint helper). Phase 1: crate skeleton and reader with span side table done (vendored patches #5 span table, #6 expansion budget). Phase 1 design for spec semantics / MathAst / OMML conventions written into the plan. Next: cargo xtask gen-math-spec (codex-backed symbol rows + overrides.json), MathAst normalization with per-fixture snapshots, diagnostics (subsystem 20), OMML writer, Typst writer. Branch feature/bd-entbg6x3-quarto-math at cd58eca37; nothing pushed."},"c-buomxl8q":{"id":"c-buomxl8q","author":"Carlos Scheidegger","created_at":"2026-09-21T19:57:01.854020Z","text":"Typst writer + split pass landed (9b606949f); whole corpus compiles under typst 0.15. Remaining in Phase 1: pampa integration seam (quarto_math::convert with verbatim fallback) and the full cargo xtask verify."},"c-f2w7uptc":{"id":"c-f2w7uptc","author":"Carlos Scheidegger","created_at":"2026-09-21T19:46:22.377916Z","text":"OMML writer landed (86a1c01a0): total writer, 263 snapshots, every fixture validates against shared-math.xsd. Phase 1 remaining: Typst writer, word-splitting pass (flag), pampa integration seam (convert + verbatim fallback), full verify."},"c-kgmbt4ay":{"id":"c-kgmbt4ay","author":"Carlos Scheidegger","created_at":"2026-09-21T21:48:21.829100Z","text":"Pushed origin/feature/bd-entbg6x3-quarto-math; draft PR https://github.com/quarto-dev/q2/pull/706 stacked on #705 (gh stack link 705 706). Held with #705."},"c-uobhs8tv":{"id":"c-uobhs8tv","author":"Carlos Scheidegger","created_at":"2026-09-21T19:33:27.916383Z","text":"Phase 1 progress: spec (commands.json + generator + xtask gen-math-spec) and MathAst normalization landed on feature/bd-entbg6x3-quarto-math (4fc6ad81b). 99 quarto-math/vendored tests green, 263 per-fixture AST snapshots. Next: diagnostics (Q-22 catalog codes + docs pages + sidebar), then the OMML writer."},"c-zndklf81":{"id":"c-zndklf81","author":"Carlos Scheidegger","created_at":"2026-09-21T20:08:42.308877Z","text":"Phase 1 complete on feature/bd-entbg6x3-quarto-math at 1464e0a50: reader, spec, MathAst, Q-22 diagnostics, OMML + Typst writers, split pass, convert() seam with an end-to-end test through pampa. Full cargo xtask verify green. Not pushed. Next: Phase 2 (OPC package layer + validation harness) as a new strand under epic bd-pq9k90z2."}}}
 {"id":"bd-environment-files-372u9qbs","title":"Project _environment files are not loaded (env shortcode resolves against process env only)","description":"Quarto 1 loads project environment-variable files before rendering: _environment, _environment.local, _environment.required, and profile variants (_environment-<profile>); already-set process variables win over file values. q2 0.15.0 never reads these files, so '{{< env NAME >}}' resolves only against the process environment — with the variable defined solely in _environment, every use renders the unresolved marker '?env:NAME' plus a Q-16-5 warning per use.\n\nRoot cause: EnvShortcodeHandler (crates/quarto-core/src/transforms/shortcode_resolve.rs, ~line 213) calls std::env::var directly, and nothing in project discovery/load (crates/quarto-core/src/project/mod.rs) looks for _environment files — 'rg _environment crates/' has no hits.\n\nSuggested fix: load _environment(.local/.required/-<profile>) during project load and inject into the process env (or a project env map consulted by EnvShortcodeHandler), with Q1 precedence: real environment > _environment.local > _environment; _environment.required lists variables that must be defined.\n\nReal-world hit: the Posit Connect docs set CONNECT_VERSION/CONNECT_SHORT_VERSION/CURRENT_YEAR in _environment; under q2 every processed '{{< env CONNECT_VERSION >}}' (page subtitles, body text, news page) renders '?env:CONNECT_VERSION' markers site-wide unless the caller exports the variables manually. (Uses inside website.title/page-footer/HTML includes fail differently — see bd strand filed alongside this one, slug shortcodes-in-metadata.)\n\nMinimal repro: /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/environment-file-ignored/ (see README.md; run 'q2 render' there WITHOUT exporting REPRO_VERSION). Origin strand in the connect-docs porting skein: br-environment-files-n0joqkug. Observed with q2 0.15.0; checked origin/main (8518ac79): no _environment handling.","status":"closed","priority":1,"issue_type":"feature","created_at":"2026-08-10T00:20:53.036683Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-10T15:38:56.926293Z","closed_at":"2026-08-10T15:38:56.926293Z","close_reason":"Merged in PR #486 (https://github.com/quarto-dev/q2/pull/486): _environment/_environment.local loaded with Q1 precedence into a project env map (no process-env mutation); env shortcode, knitr/jupyter subprocesses, and render scripts consume it; _environment.required gets real span-carrying diagnostics; docs at guides/projects/environment.qmd. Profiles follow-up remains open as bd-ev8mk1rp.","labels":["parity"],"comments":{"c-1793ir6y":{"id":"c-1793ir6y","author":"Carlos Scheidegger","created_at":"2026-08-10T14:59:25.143033Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/486 (feature/bd-environment-files-372u9qbs-load-environment-files). Local main reset to origin/main; work lives on the branch."},"c-jhsv0r6i":{"id":"c-jhsv0r6i","author":"Carlos Scheidegger","created_at":"2026-08-10T14:57:11.437997Z","text":"Implementation complete on main (staged, awaiting commit approval): span-annotated dotenv parser (project/environment.rs, 48 unit tests), project_env on StageContext, env shortcode consults map after real env, subprocess propagation (knitr/jupyter/render-scripts, real-env-wins via pre-filter), docs page guides/projects/environment.qmd. E2E verified with real R + jupyter (engine-sees-me / real-wins) and the repro (from-env-file). Full cargo xtask verify green."},"c-l7drvw2q":{"id":"c-l7drvw2q","author":"Carlos Scheidegger","created_at":"2026-08-10T14:18:00.068590Z","text":"Design decisions recorded in claude-notes/plans/2026-08-09-environment-files-loading.md (commit 315271cf): env map on StageContext, no process-env mutation (policy); subprocess propagation in scope; profiles scoped out to bd-ev8mk1rp; .required gets diagnostics; hand-rolled span-annotated parser (dotenvy rejected: not event-based, no spans, reads env during $VAR substitution). Q1 dialect = @std/dotenv 0.225.x."}}}
 {"id":"bd-enzra5mw","title":"q2 preview: 'Download compilable typst project' (zip) download type","description":"Idea from q2-preview hub-parity Plan 6. The typst source download (.typ) is a bare file: images and other resources it references are not included, so it will not compile outside the project. Add a separate download type that is a compilable zip: the .typ plus bundled images. Open question: should it also include .quarto/ resources, fonts and @preview packages for full reproducibility (possibly yes)? Scope: server route (zip of a self-contained request directory), client format table (extension/MIME zip), hub-client wasm path parity. Not part of the hub-parity epic; the epic ships the bare .typ and documents the limitation. Plan 6 had an uncommitted collector prototype that copies images next to the .typ (typst + non-pdf output-ext, out of place); it is dropped from Plan 6 and can be recovered from the Plan 6 branch handoff note.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-10-08T23:01:03.541354Z","created_by":"Gordon Woodhull","updated_at":"2026-10-08T23:01:03.541354Z","labels":["q2-preview","typst"]}
-{"id":"bd-epk8cmdl","title":"Listing code paths treat leading-/ src as external instead of site-root-relative","description":"Decision 4 of bd-root-relative-paths-design-fc5pvkcv: a URL-space leading / means site-root-relative, uniformly. Two listing sites still deviate (conservative defaults, no principled reason):\n\n1. crates/quarto-core/src/project/listing/helpers.rs is_external_src() classifies src.starts_with('/') as external, so listing preview hydration/rebase/copy-intent registration (bd-qv2lsab0) skips root-relative images.\n2. crates/quarto-core/src/project/listing/post_render_upgrade/substitute.rs resolve_preview_url() passes /-prefixed preview srcs through verbatim instead of rebasing them relative to the host page.\n\nUnder the decree both should resolve against the project root and relativize per page, like body images (link_rewrite.rs Image arm) and config assets (resolve_root_relative_resource_href) now do. See claude-notes/plans/2026-08-13-site-root-relative-paths.md § Q4 survey.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-08-13T18:50:37.786329Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-13T18:50:37.786329Z","labels":["websites"],"dependencies":{"bd-root-relative-paths-design-fc5pvkcv:discovered-from":{"depends_on_id":"bd-root-relative-paths-design-fc5pvkcv","type":"discovered-from","created_at":"2026-08-13T18:50:37.786329Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-epk8cmdl","title":"Listing code paths treat leading-/ src as external instead of site-root-relative","description":"Decision 4 of bd-root-relative-paths-design-fc5pvkcv: a URL-space leading / means site-root-relative, uniformly. Two listing sites still deviate (conservative defaults, no principled reason):\n\n1. crates/quarto-core/src/project/listing/helpers.rs is_external_src() classifies src.starts_with('/') as external, so listing preview hydration/rebase/copy-intent registration (bd-qv2lsab0) skips root-relative images.\n2. crates/quarto-core/src/project/listing/post_render_upgrade/substitute.rs resolve_preview_url() passes /-prefixed preview srcs through verbatim instead of rebasing them relative to the host page.\n\nUnder the decree both should resolve against the project root and relativize per page, like body images (link_rewrite.rs Image arm) and config assets (resolve_root_relative_resource_href) now do. See claude-notes/plans/2026-08-13-site-root-relative-paths.md § Q4 survey.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-13T18:50:37.786329Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T22:51:28.119116Z","closed_at":"2026-10-28T22:51:28.119116Z","close_reason":"Merged in PR #958 (dbd7cebd5): reveal.js logo resolved via config-path resolver into URL space with a copy intent; listing leading-/ image src anchored at the site root with quarto_util::is_external_url as the sole classifier.","labels":["websites"],"dependencies":{"bd-root-relative-paths-design-fc5pvkcv:discovered-from":{"depends_on_id":"bd-root-relative-paths-design-fc5pvkcv","type":"discovered-from","created_at":"2026-08-13T18:50:37.786329Z","created_by":"Carlos Scheidegger"}},"comments":{"c-0w3yvcp5":{"id":"c-0w3yvcp5","author":"Carlos Scheidegger","created_at":"2026-10-28T20:13:49.174069Z","text":"Fixed together with bd-7328beha on branch braid/bd-7328beha-revealjs-logo-emitted-raw (both URL-space leading-/ emitters): listing helpers/item/listing_generate/substitute now use one classifier (quarto_util::is_external_url); a leading / rebases to the bare project-relative path at hydration and relativizes per host page; resolve_preview_url anchors /x at the output root."}}}
 {"id":"bd-epufp4ys","title":"Perf: box Link/Image source infos and Table to shrink Inline (~360 -> ~250) and Block (~824 -> ~330)","description":"After bd-1c085k3a, Inline is 360 bytes (set by Link/Image: Attr 104 + target 48 + content 24 + SourceInfo 32 + AttrSourceInfo 80 + TargetSourceInfo 64) and Block is 824 (set by Table). Boxing attr_source/target_source on Link/Image brings Inline to about Span/Code size (~250); boxing Table (then Figure) in Block brings Block to ~330. quarto-pandoc-types only; serde sees through Box so the wire shape is unchanged. Measure with the ast_node_size_ceilings test (lower the ceilings) and hyperfine on docs-quarto-2 per claude-notes/research/2026-09-19-topdown-traverse-alloc-perf.md; walk cost scales roughly linearly with element size (microbenchmark in the note).","status":"open","priority":3,"issue_type":"task","created_at":"2026-09-19T22:06:45.642624Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-19T22:06:45.642624Z","labels":["perf"],"dependencies":{"bd-5yektmwt:related":{"depends_on_id":"bd-5yektmwt","type":"related","created_at":"2026-09-19T22:06:45.642624Z","created_by":"Carlos Scheidegger"},"bd-w0x91nmh:discovered-from":{"depends_on_id":"bd-w0x91nmh","type":"discovered-from","created_at":"2026-09-19T22:06:45.642624Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-ersobfbt","title":"Website header: headroom.js scroll-away + fixed-top parity (deferred from secondary-nav)","description":"Deferred from bd-26bf3j1y (Carlos's decision 2, 2026-08-17): that strand introduces <header id=\"quarto-header\"> as a STATIC wrapper — no 'headroom' class, no 'fixed-top', and no body.nav-fixed. This strand is the Q1-parity follow-up.\n\nQ1 emits <header id=\"quarto-header\" class=\"headroom fixed-top\"> and ships headroom.js, which pins/unpins the header on scroll direction. Q1 SCSS consumers:\n- quarto-nav.scss ~103-115: .headroom-target / header.headroom transition rules and header.headroom--pinned transform.\n- quarto-nav.scss:822: body.nav-fixed { padding-top: navbar-default-offset($theme-name) } — pure compensation for the fixed header, set by the postprocessor when '#quarto-header.fixed-top nav.navbar' is present.\n\nIMPORTANT: those three go together. Adding 'fixed-top' without body.nav-fixed and its padding rule makes the header overlap page content. Ship them in one change or not at all.\n\nAlso in scope: the inline onclick hooks Q1 puts on the secondary-nav toggle and flex-grow link — onclick=\"if (window.quartoToggleHeadroom) { window.quartoToggleHeadroom(); }\". They are guarded, so bd-26bf3j1y omits them safely; this strand adds the function and (optionally) the hooks.\n\nPrerequisite: bd-26bf3j1y must land first (it creates the #quarto-header wrapper this strand modifies).\n\nPlan context: claude-notes/plans/2026-08-17-website-secondary-nav-mobile.md (resolved decision 2).","status":"closed","priority":3,"issue_type":"feature","created_at":"2026-08-17T20:43:57.735378Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-24T18:34:22.926597Z","closed_at":"2026-08-24T18:34:22.926597Z","close_reason":"Implemented: fixed scroll-away header (headroom.js + fixed-top + body.nav-fixed) for q2 render and q2 preview/hub-client, pinned: opt-out, docs page. Branch braid/bd-ersobfbt-headroom-fixed-top; PR to follow. Follow-up: bd-pt1wxeq2 (sticky redesign).","dependencies":{"bd-26bf3j1y:blocks":{"depends_on_id":"bd-26bf3j1y","type":"blocks","created_at":"2026-08-17T20:44:14.476739Z","created_by":"Carlos Scheidegger"},"bd-26bf3j1y:discovered-from":{"depends_on_id":"bd-26bf3j1y","type":"discovered-from","created_at":"2026-08-17T20:43:57.735378Z","created_by":"Carlos Scheidegger"},"bd-e7b7:related":{"depends_on_id":"bd-e7b7","type":"related","created_at":"2026-08-21T21:09:36.594884Z","created_by":"Carlos Scheidegger"}},"comments":{"c-1gw7foy7":{"id":"c-1gw7foy7","author":"Carlos Scheidegger","created_at":"2026-08-21T21:09:40.317530Z","text":"Investigated 2026-08-21; plan skeleton at claude-notes/plans/2026-08-21-headroom-fixed-top-header.md (verdict: ready to design). Key finding: the preview iframe is persistent and already injects Bootstrap JS (entry.tsx, Phase F.1), so the 'preview reinitializes its iframe' rationale behind the wasm32 cfg gates is stale."},"c-6txu14sh":{"id":"c-6txu14sh","author":"Carlos Scheidegger","created_at":"2026-08-24T14:37:49.050513Z","text":"Design aligned 2026-08-24: all 8 questions resolved (plan §Resolved decisions). Option A here, structured so bd-pt1wxeq2 (Option B sticky redesign, filed, blocked-by this) can swap the machinery cheaply. nav-fixed via accumulating body-class list (7b) — snapshot churn to be measured and reported. Secondary-nav wasm gate lifts here; pinned: ships here."},"c-ibvackzh":{"id":"c-ibvackzh","author":"Carlos Scheidegger","created_at":"2026-08-24T15:48:55.720472Z","text":"All four phases complete on braid/bd-ersobfbt-headroom-fixed-top (5 commits, c9c5d5e5 tip + plan commit). Final full cargo xtask verify exit=0. Not pushed; bd-2yd37vuk close and pushing await Carlos."},"c-nn3tepbc":{"id":"c-nn3tepbc","author":"Carlos Scheidegger","created_at":"2026-08-24T15:44:00.468391Z","text":"Phases 2-3 done: preview header + headroom (same vendored JS injected in entry.tsx; pinned honored via data-headroom-pinned), secondary-nav wasm gate lifted, docs page guides/projects/header-scrolling.qmd. Browser e2e both surfaces passed; it caught and fixed a real bug (quarto-nav.js DOMContentLoaded never firing in the dynamically-imported preview entry). Final full verify running."},"c-stwky5vu":{"id":"c-stwky5vu","author":"Carlos Scheidegger","created_at":"2026-08-24T15:12:00.062479Z","text":"Phases 0-1 done (commit 2d9107e9 on braid/bd-ersobfbt-headroom-fixed-top): native q2 render ships the fixed scroll-away header end-to-end. TDD red-first throughout; 13003/13003 workspace tests; snapshot churn: 0 .snap, 1 baseline hash (styles.css, note appended). Phase 2 (preview) next."}}}
 {"id":"bd-escaped-executable-fence-uuvv37pk","title":"Quarto 1's doubled-brace escape for showing a cell is unsupported: literal in a fence, bare parse error in prose","description":"Quarto 1's way of *showing* an executable cell rather than running it — doubling the braces in the fence opener — has no q2 equivalent, and the two contexts fail differently. Both matter to anyone porting docs that document Quarto syntax.\n\n## What happens vs Quarto 1\n\n### Inside a displayed code fence — silently wrong output\n\n    ````markdown\n    ```{{python}}\n    1+1\n    ```\n    ````\n\nQuarto 1 collapses the doubled pair and shows the reader ```{python}```. q2 0.19.0 leaves it doubled: the reader is shown ```{{python}}```, which is not valid Quarto syntax, presented as the thing to copy. No diagnostic.\n\n### In prose — a bare parse error\n\n    X {{python}} Y\n\nfails the whole file with\n\n    Error: Parse error\n     1 │ X {{python}} Y\n       │    ┬\n       │    ╰── unexpected character or token here\n\nNo error code, no hint. The single-brace form \"X {python} Y\" is much better served: it gets [Q-2-41] with an actionable message (\"To write literal braces, escape them as \\{...\\}\"). So the form a porter's source actually contains — the Q1 escape — produces the least helpful failure q2 has.\n\n## What already works (so this is narrower than it looks)\n\n- A fence opener with SINGLE braces inside an outer ````markdown fence renders correctly in q2. Fence bodies are verbatim; there is nothing to escape there.\n- In prose, both \\{python\\} and a code span render {python} cleanly.\n\nSo q2 already has expressive coverage for every case. What is missing is (a) accepting or rejecting Q1's spelling deliberately, and (b) not falling off a cliff when it appears.\n\n## Root-cause area\n\nFence bodies are assembled verbatim from source bytes in process_code_fence_content (crates/pampa/src/pandoc/treesitter_utils/code_fence_content.rs) — it splices around block_continuation markers and does no unescaping, which is why the doubled pair survives to output. The prose path is the attribute-syntax reader that raises Q-2-41 (catalog entry in crates/quarto-error-catalog/error_catalog.json:450); the doubled form evidently reaches the parser before that check and dies as an uncoded parse error instead.\n\n## Suggested fix direction\n\nThe design question is the important half, and it should be answered before any code moves: does q2 adopt {{...}} as an escape at all?\n\nRecommendation is NO for a blanket collapse, for a reason we hit in practice — see the counter-case below — and instead:\n\n1. In prose, make {{...}} produce Q-2-41 (or a sibling code) rather than a bare parse error, with a hint naming the forms that work (backslash-escaped braces, or a code span). Cheap, and it converts the worst failure into the best one.\n2. For fences, decide explicitly and document it. If q2's answer is \"write single braces, fence bodies are verbatim\", that is a perfectly good answer and only needs a migration note plus, ideally, a qmd-syntax-helper rule that rewrites a doubled-brace fence opener to the single-brace form inside displayed fences.\n\n## Counter-case that should shape the decision\n\nuser/fastapi/index.qmd in the Connect docs shows a Jinja template containing a genuine {{ request.headers[...] }}. Quarto 1 unescapes it to a single brace pair, which is WRONG for that page — the reader needs both braces. q2's literal output is the correct one there. A blanket collapse of every doubled brace would regress that page; whatever q2 does should be an explicit, documented escape rather than a global rule.\n\n## Real-world hit in the Connect docs\n\nuser/quarto/index.qmd (5 blocks), user/rmarkdown/index.qmd (2), how-to/publish-databricks-quarto-notebook/index.qmd (1) — all pages whose subject IS how to write Quarto cells, all currently showing the doubled-brace form as the example to copy. These are 5 of the ~47 pages that still differ from the Q1 reference render.\n\n## Repro\n\n/Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/escaped-executable-fence/\n\nRun \"q2 render\" there and compare with \"quarto render\". The repro covers the fence case; the prose parse error is the one-liner \"X {{python}} Y\" above.\n\nObserved from q2 0.16.0 through 0.19.0 (re-verified against the released 0.19.0 binary; HEAD 1ba0f2ec). Origin strand in the q2-connect-docs skein: br-xa5oawj0.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-12T19:13:44.976781Z","created_by":"Claude (q2-connect-docs)","updated_at":"2026-08-17T20:52:07.424970Z","closed_at":"2026-08-17T20:52:07.424970Z","close_reason":"Implemented as Q-2-50 (prose merr-corpus error + fence-opener warning in postprocess), plus docs page and migration note. No {{...}} escape adopted; no qmd-syntax-helper rule (design decision). Commit 297f5cc2; plan claude-notes/plans/2026-08-17-doubled-brace-escape.md","labels":["parser"],"comments":{"c-01qagnu8":{"id":"c-01qagnu8","author":"Carlos Scheidegger","created_at":"2026-08-17T20:41:17.162295Z","text":"Implementation done pending full verify: Q-2-50 corpus cases (prose + link-text; states 2601/2604, no cross-code collisions), catalog entry, docs page docs/errors/markdown/Q-2-50.qmd, fence-side warning via with_code_block arm in postprocess (class starting with '{{' — structurally unambiguous), migration note in docs/guides/authoring/computations.qmd. 9/9 targeted tests pass, 12237 workspace tests green, verified end-to-end through q2 render (warning at 1:5 on the info string; HTML renders literal block)."},"c-69bevrbr":{"id":"c-69bevrbr","author":"Carlos Scheidegger","created_at":"2026-08-17T20:22:00.991049Z","text":"Design aligned with Carlos 2026-08-17: (1) q2 does NOT adopt {{...}} as an escape — fence bodies stay verbatim; (2) mint sibling code Q-2-50 (not Q-2-41) so messaging names the Q1 idiom; (3) deliberately NO qmd-syntax-helper rule — Q1's nested/mixed-brace treatment is heuristic, don't accumulate brittle rules; (4) render-time diagnostic for top-level doubled-brace fence openers (AST-level check on the CodeBlock class — verified {{python}} parses to a literal class, so no false-positive surface). Plan updated with firm phases: claude-notes/plans/2026-08-17-doubled-brace-escape.md"},"c-tyew1sd3":{"id":"c-tyew1sd3","author":"Carlos Scheidegger","created_at":"2026-08-17T20:12:19.294481Z","text":"Investigated on main @ 2760390b. Both failure modes reproduce at HEAD; plan skeleton at claude-notes/plans/2026-08-17-doubled-brace-escape.md, repro fixtures + mechanism notes in claude-notes/plans/doubled-brace-escape-investigation/. Verdict: ready to design — the policy call ({{...}} escape yes/no) is pending user alignment."}}}
@@ -839,6 +845,7 @@
 {"id":"bd-evxckg64","title":"Canonical author-ID HMAC input: bare doc id, stripped server-side","description":"Canonical project input to the /auth/author + /auth/actor HMAC is the BARE index doc id — the spelling the author-ID transition E2E pinned byte-for-byte (plan 2026-09-30 lines 685-687). bd-9t8zsh98's MCP-side prefix fix rested on the false premise that the SPA resolves authors over the prefixed form; it does so only for share-joined projects, while every SPA-created project resolves bare. Fix: strip a leading 'automerge:' in sub_to_actor_id_for_project (auth.rs) — one point covers both endpoints (D5). Then drop hub-mcp's prefix canonicalization in connection-manager.fetchAuthorId and rewrite its comment. Zero SPA changes. One-time attribution-key flip for share-attached collaborators + MCP users; connect() re-writes setIdentity so name/color follow the new key (client.ts:1296). Deploy hub first; MCP cleanup ships anytime after. Plan: claude-notes/plans/2026-10-15-bare-canonical-author-id.md","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-10-15T09:05:02.726588Z","created_by":"shikokuchuo","updated_at":"2026-10-15T10:01:25.892883Z","closed_at":"2026-10-15T10:01:25.892883Z","close_reason":"Done: hub strips a leading 'automerge:' in sub_to_actor_id_for_project (commit fdc1ec20b) so both spellings mint the bare-canonical author/actor ID on /auth/author and /auth/actor (D5 preserved); hub-mcp fetchAuthorId sends the id as given (commit b005e71c3). TDD red/green at the unit boundary and through the real axum routes (session_auth); cargo xtask verify --skip-hub-build all 14 steps green; MCP package suite 621 passed. Deploy order: hub first — the strip is backward-compatible with all client spellings; the MCP change against a pre-strip hub would shift (not fix) the mismatch cohort.","dependencies":{"bd-9t8zsh98:supersedes":{"depends_on_id":"bd-9t8zsh98","type":"supersedes","created_at":"2026-10-15T09:05:02.726588Z","created_by":"shikokuchuo"}}}
 {"id":"bd-ew0vak6b","title":"hub-client q2-preview: editable/read-only toggle in the bottom status bar","description":"format: q2-preview in hub-client is always editable: a mouse-up on any editable paragraph activates the block editor, so clicking a link inside a paragraph also opens an editor (links are hard to follow). q2 preview already has this distinction via --allow-edit, and preview-renderer already honors an editingDisabled flag carried in the UPDATE_AST payload; hub-client never sets it. Add an Edit pill to the replay/status bar at the bottom of the document view (next to Authors and Comments) that turns block editing on/off, persist it as a user preference, and thread it to the iframe as editingDisabled. Plan: claude-notes/plans/2026-09-09-q2-preview-edit-toggle.md","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-09-09T17:27:59.878625Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-09T19:11:33.850419Z","closed_at":"2026-09-09T19:11:33.850419Z","close_reason":"Merged in PR #667 (Edit pill in the bottom bar; previewEditing preference; renderer close-on-disable).","labels":["hub-client","q2-preview"],"comments":{"c-1dzpk1o1":{"id":"c-1dzpk1o1","author":"Carlos Scheidegger","created_at":"2026-09-09T18:22:33.424591Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/667 (remote branch feature/bd-ew0vak6b-hub-client-q2-preview). Awaiting CI + review."},"c-9lpme53b":{"id":"c-9lpme53b","author":"Carlos Scheidegger","created_at":"2026-09-09T17:30:20.750231Z","text":"Source study done; plan drafted at claude-notes/plans/2026-09-09-q2-preview-edit-toggle.md. Key finding: preview-renderer already implements read-only mode as editingDisabled (bd-ov4gqk3m) and Q2PreviewIframe already carries it in UPDATE_AST; hub-client never passes it. Awaiting design review (D1-D9) before execution."},"c-cxftsogc":{"id":"c-cxftsogc","author":"Carlos Scheidegger","created_at":"2026-09-09T17:36:35.825472Z","text":"Design review 2026-09-09: D1-D9 accepted as written; D8 confirmed to match the header view-mode toggle pattern (persisted, chrome-only control). Ready to execute on go-ahead."},"c-gk3l87hj":{"id":"c-gk3l87hj","author":"Carlos Scheidegger","created_at":"2026-09-09T17:52:29.704719Z","text":"Implemented on branch braid/bd-ew0vak6b-hub-client-q2-preview: 93b146c (code + tests) and the changelog follow-up. cargo xtask verify --skip-rust-tests, lint, and the Playwright spec are green. Not pushed."},"c-ij4vsi1h":{"id":"c-ij4vsi1h","author":"Carlos Scheidegger","created_at":"2026-09-09T19:10:48.626993Z","text":"CI green on PR #667 after 59cb2e3 (all 10 checks pass). Ready for review/merge."},"c-tsa6uu70":{"id":"c-tsa6uu70","author":"Carlos Scheidegger","created_at":"2026-09-09T18:42:47.347716Z","text":"CI: both 'Run test suite' jobs failed on lint:css (no-physical-box-props, margin-left in ReplayDrawer.css). Fixed in 59cb2e3 (margin-inline-start). Filed the verify/CI gap as a separate chore strand."}}}
 {"id":"bd-ewfppclm","title":"RevealJS panels/tabsets (.panel-tabset) support","description":"::: {.panel-tabset} is not implemented in Q2 for ANY format (no tabset AST transform anywhere; verified 2026-06-16). Greenfield: needs a tabset AST transform + SCSS (+ JS for reveal tab switching). Lowest priority. Deferred from Stage D.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-06-16T18:49:49.240760Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-16T18:49:49.240760Z","labels":["revealjs"],"dependencies":{"bd-yown2ts4:related":{"depends_on_id":"bd-yown2ts4","type":"related","created_at":"2026-06-16T18:49:49.240760Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-ex8qvuwu","title":"aliases: opt-in for case-sensitive deployment targets so case-only alias/page pairs render (issue #944)","description":"GH #944 (split from #939, Posit Connect docs port): an alias whose stub path differs from a rendered page only by case (e.g. r/index.qmd declaring aliases: [/R/]) is a hard Q-5-25 error with no escape hatch. Connect serves from S3 where R/index.html and r/index.html are distinct keys and the old /R/ URLs must keep working; Quarto 1 (exact-string compare) writes both files on Linux. This deliberately revisits decision 3 of claude-notes/plans/2026-08-12-aliases-redirect-stubs.md.\n\nDirection (user, 2026-10-28): keep the error as the default; add an affirmative website.* boolean option (proposed name website.case-sensitive-paths, final name TBD) declaring the deployment target case-sensitive. With it set: the two CaseOnly* conflicts in plan_alias_stubs are not raised and stubs are planned; write_alias_redirects probes whether the *output dir* is case-insensitive and, if so, skips the colliding stub with a warning instead of clobbering the page; Q-5-25 diagnostics gain a hint naming the option; error_catalog.json, docs/errors/project/Q-5-25.qmd and docs/guides/projects/aliases.qmd document the option and the Quarto 1 discrepancy (Q1 is host-dependent: Linux writes both, macOS/Windows replace the page with a self-redirect). Non-website projects warn the key has no effect (mirror website.llms-txt). Tests must branch on the runtime probe, not cfg!(target_os).\n\nTriage record: claude-notes/issue-reports/944/triage.md on branch issue-944 (topic branch in the main checkout, no worktree). Repro fixture alongside it.","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-10-28T12:49:53.520656Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T14:49:26.135347Z","closed_at":"2026-10-28T14:49:26.135347Z","close_reason":"Merged to main as 370e43381 via PR #952; GH #944 closed. website.case-sensitive-paths opts in to case-only alias/page pairs, with Q-5-51 guarding case-insensitive output dirs.","dependencies":{"bd-aliases-redirects-missing-sch7cd1g:related":{"depends_on_id":"bd-aliases-redirects-missing-sch7cd1g","type":"related","created_at":"2026-10-28T12:49:53.520656Z","created_by":"Carlos Scheidegger"}},"comments":{"c-fo08ypyh":{"id":"c-fo08ypyh","author":"Carlos Scheidegger","created_at":"2026-10-28T13:57:23.423240Z","text":"Implemented on branch issue-944 (remote feature/issue-944), commit 7e4ed504e. Verify green locally; both filesystem branches of the new tests exercised locally (default APFS + case-sensitive APFS image). PR opened to watch CI on Linux and macOS/Windows runners."},"c-ne5ynbrw":{"id":"c-ne5ynbrw","author":"Carlos Scheidegger","created_at":"2026-10-28T13:40:03.416176Z","text":"Option name decided by user 2026-10-28: website.case-sensitive-paths: true. Implementation starting on branch issue-944."},"c-oidr1424":{"id":"c-oidr1424","author":"Carlos Scheidegger","created_at":"2026-10-28T14:48:09.361849Z","text":"Merged to main as 370e43381 via PR #952 (CI green on ubuntu + macOS test suites and Windows clippy; GH #944 auto-closed). Branch issue-944 deleted locally and on origin. Strand left open for the user to close."}}}
 {"id":"bd-exhbc6h8","title":"Engine registry init-time diagnostics never reach the user (drain registry.diagnostics; route pass-1 warning through the sink)","description":"Two related gaps in how project-scoped ENGINE diagnostics reach the user. Found while investigating pass-1 engine resolution on feature/ts-engine-extensions (worktree .worktrees/ts-engine-extensions, HEAD 645845ea2). NOT part of the current merge runbook (claude-notes/plans/2026-08-13-ts-engine-extensions-merge-main.md) -- this is unfinished work from plan1c.\n\n(1) registry.diagnostics is written but never drained.\n- Producers: engine_contribution_missing_fields_warning (crates/quarto-core/src/extension/types.rs:283) pushed at crates/quarto-core/src/project/mod.rs:1576, and TsEngine's intermediate-files failure warning (crates/quarto-core/src/engine/ts_engine.rs:977).\n- Consumers in production code: NONE. The only reader is a test (crates/quarto-core/tests/integration/engine_registry_build.rs:626). The leftover marker is still in place: '// Task: drain registry.diagnostics at orchestrator (plan step 10)' at crates/quarto-core/src/project/mod.rs:1645.\n- 'plan step 10' = claude-notes/plans/2026-04-16-plan1c-extension-integration.md, registry-build step 10 ('Drain registry.diagnostics and emit directly to the user-facing diagnostic sink'). It was prose inside the spec, never a checklist item, so it was never reconciled. Both the producer and the marker landed in the same commit, 171aef17f (Engine stop B).\n- User-visible effect: an engine extension missing name/claims/file-extensions/claims-files gets a precise, actionable warning naming exactly which fields are missing -- and it is silently discarded. Verified E2E: rendered a scratch project with the legacy-python fixture (missing claims, file-extensions, claims-files); nothing printed.\n\n(2) The pass-1 fall-through warning bypasses the diagnostic pipeline.\n- pass1_engine_resolution_warning (crates/quarto-core/src/project/orchestrator.rs:278) IS emitted -- but via bare eprintln! at orchestrator.rs:1114, not into any diagnostic collection.\n- Consequence: it is not counted by ProjectRenderSummary::diagnostic_counts and is not promoted by promote_warnings_to_errors, so --strict-style warning promotion (bd-yjs54ptg / GH #220) does not see it.\n- This is the natural landing site for (1): the orchestrator already has a project-grain engine-warning print point; both should go through the same sink.\n\nSuggested shape: give the orchestrator one project-grain diagnostic channel, drain registry.diagnostics into it after build_engine_registry, and route pass1_engine_resolution_warning through it too, so both are counted and promotable.\n\nTDD note: (1) needs a test that a legacy extension's missing-fields warning actually reaches user-visible output (the existing test only reads the vec in-process); (2) needs a test that a fell-through render's warning increments ProjectRenderSummary::diagnostic_counts().warnings and is promoted under strict mode.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-17T03:31:13.123662Z","created_by":"Gordon Woodhull","updated_at":"2026-08-17T23:36:58.788484Z","closed_at":"2026-08-17T23:36:58.788484Z","close_reason":"Drained at end of render into project_diagnostics; Q-16-11 routed off eprintln!. 9 e2e tests, workspace 12,635 passed, full cargo xtask verify green. Commit 8b02c8651.","comments":{"c-3v8cshbg":{"id":"c-3v8cshbg","author":"Gordon Woodhull","created_at":"2026-08-17T03:34:59.073485Z","text":"CORRECTED VERSION of comment c-g8a3nwky (that one lost several backticked\nfragments to shell mangling — read this one instead).\n\nDesign note on message SHAPE — draining as-is is not enough.\n\n`engine_contribution_missing_fields_warning` builds its message with\n`DiagnosticMessage::warning(msg)`, which puts the ENTIRE ~380-char text in the\ndiagnostic's *title* (quarto-error-reporting 0.2.1: `DiagnosticMessage::warning`\n-> `Self::new(kind, title)`, diagnostic.rs:301). `to_text` does not wrap. So\ndraining it verbatim would print one run-on line:\n\n    Warning: Engine extension 'acme/legacy-python' (path: /abs/path/_extensions/acme/legacy-python/dist/legacy-python.js) is missing static declarations: claims, file-extensions, claims-files. An engine missing name/claims/claims-files is treated as a legacy Q1 engine and LoadEngined on every render (slow dynamic path) vs. zero-load when declared. Declare [] or {} to silence a field the engine genuinely has none of.\n\nCompare the pass-1 warning, which uses `DiagnosticMessageBuilder`\n(subject / problem / add_detail / add_hint) and renders as a structured block.\nRebuild the missing-fields message the same way before draining it — roughly:\n\n    Warning: engine extension `acme/legacy-python` is missing static declarations\n    so q2 must load it on every render instead of resolving statically.\n    ✖ missing: claims, file-extensions, claims-files\n    ℹ declare each missing key in _extension.yml; `[]` / `{}` counts as declared\n      (\"this engine genuinely has none\"), e.g. `claims: [python]`.\n\nAlso: the path currently shown is the engine's `.js` bundle (the `path` field of\n`EngineContribution::External`), while the pass-1 warning shows the\n`_extension.yml`. The `_extension.yml` is the file a user would EDIT to fix\nthis, so prefer `extension_yml_path` (already carried on the contribution for\nexactly this reason — see its doc comment at extension/types.rs:105) or show\nboth.\n\nCOALESCING: a single legacy engine will now trigger BOTH warnings in one render\n— missing-fields at registry build (fires regardless of impact) and pass-1\nfall-through at end of Pass-1 (fires only when >=1 doc is affected). In the\nlegacy-python E2E both would print, naming the same engine, with overlapping\nhints. Decide the relationship when wiring the shared sink: suppress the\nmissing-fields warning for engines the pass-1 warning already names, or merge\nthem into one engine-grain diagnostic."},"c-g8a3nwky":{"id":"c-g8a3nwky","author":"Gordon Woodhull","created_at":"2026-08-17T03:34:16.721980Z","text":"Design note on message SHAPE -- draining as-is is not enough.\n\nengine_contribution_missing_fields_warning builds its message with DiagnosticMessage::warning(msg), which puts the ENTIRE ~380-char text in the diagnostic's *title* (quarto-error-reporting 0.2.1: DiagnosticMessage::warning -> Self::new(kind, title), diagnostic.rs:301). to_text does not wrap. So draining it verbatim would print one run-on line:\n\n    Warning: Engine extension 'acme/legacy-python' (path: /abs/path/_extensions/acme/legacy-python/dist/legacy-python.js) is missing static declarations: claims, file-extensions, claims-files. An engine missing name/claims/claims-files is treated as a legacy Q1 engine and LoadEngined on every render (slow dynamic path) vs. zero-load when declared. Declare [] or {} to silence a field the engine genuinely has none of.\n\nCompare the pass-1 warning, which uses DiagnosticMessageBuilder (subject / problem / add_detail / add_hint) and renders as a structured block. Rebuild the missing-fields message the same way before draining it -- roughly:\n\n    Warning: engine extension  is missing static declarations\n    so q2 must load it on every render instead of resolving statically.\n    x missing: claims, file-extensions, claims-files\n    i declare each missing key in _extension.yml;  /  counts as declared\n      ('this engine genuinely has none'), e.g. .\n\nAlso: note the path shown is the engine's .js bundle (the  field of EngineContribution::External), while the pass-1 warning shows the _extension.yml. The _extension.yml is the file a user would EDIT to fix this, so prefer extension_yml_path (already carried on the contribution for exactly this reason, per its doc comment at extension/types.rs:105) or show both.\n\nCOALESCING: a single legacy engine will now trigger BOTH warnings in one render -- missing-fields at registry build (fires regardless of impact) and pass-1 fall-through at end of Pass-1 (fires only when >=1 doc is affected). In the legacy-python E2E both would print, naming the same engine, with overlapping hints. Decide the relationship when wiring the shared sink: suppress the missing-fields warning for engines the pass-1 warning already names, or merge them into one engine-grain diagnostic."}}}
 {"id":"bd-exk3hfxk","title":"hub sessions C6: client alignment (retire One-Tap renewal)","description":"Rely on server sliding re-issue; retire One-Tap renewal dependency (coordinate with bd-qxgoti2b B2); confirm /auth/me exp semantics (sliding); keep periodic /auth/me probe as keep-alive while WS open. Plan phase C6.","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-07-24T12:14:01.965044Z","created_by":"shikokuchuo","updated_at":"2026-07-24T14:04:37.158390Z","closed_at":"2026-07-24T13:22:58.179534Z","close_reason":"Landed in 8527e62d: useSessionKeepAlive (1h /auth/me probe while WS online, slid expiry fed back via applyAuth), One-Tap renewal demoted to documented fallback, sliding exp semantics confirmed + documented. Full One-Tap retirement remains with bd-qxgoti2b B2.","labels":["auth"],"dependencies":{"bd-ey6jg70f:parent-child":{"depends_on_id":"bd-ey6jg70f","type":"parent-child","created_at":"2026-07-24T12:14:01.965044Z","created_by":"shikokuchuo"},"bd-nh5pt1pd:blocks":{"depends_on_id":"bd-nh5pt1pd","type":"blocks","created_at":"2026-07-24T12:15:58.570681Z","created_by":"shikokuchuo"}},"comments":{"c-91ewvxva":{"id":"c-91ewvxva","author":"shikokuchuo","created_at":"2026-07-24T14:04:37.158390Z","text":"History replay (security-review backout): commit a7f47625 is now ee905592."},"c-aeps2rt5":{"id":"c-aeps2rt5","author":"shikokuchuo","created_at":"2026-07-24T13:55:23.017622Z","text":"History replay (plan-file backout, branch rebuilt from C5): commit 8527e62d is now a7f47625."}}}
 {"id":"bd-expy","title":"Pipe table parser commits to caption when ':::' follows a row (issue #206)","description":"Parse error when a fenced-div close `:::` directly follows a pipe table row without an intervening blank line. The first `:` of `:::` is shifted by the parser as the start of a `caption` rule (the caption first-token literal is `:`); once shifted, tree-sitter cannot back out, and the second `:` produces 'unexpected character or token here'.\n\nTriage doc: claude-notes/issue-reports/206/triage.md\nWorktree branch: issue-206\nGitHub issue: https://github.com/quarto-dev/q2/issues/206\nReporter: @rundel\nReal-world example: quarto-dev/quarto-web docs/websites/website-navigation.qmd L155-L159\n\nConfirmed not a fenced-div bug — bare pipe table + `:::` reproduces identically. The fenced div in the reporter's example is incidental.\n\nRecommended fix shape: introduce an external `_caption_start` scanner token that only matches a `:` not immediately followed by another `:` (and followed by inline whitespace). Re-key the `caption` rule on `_caption_start` instead of the literal `:`. Add tree-sitter corpus tests for `:::`-after-table, `: caption`-after-table, bare `:::` after table, and a pampa round-trip test for the repro fixture. Re-test on quarto-web after fix.\n\nOut of scope here, file separately if desired: pipe-table parser also silently absorbs trailing headings/paragraphs as one-cell rows when no blank line intervenes (see triage doc 'Adjacent finding').","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-05-15T18:48:39.090203Z","created_by":"cscheid","updated_at":"2026-05-15T21:35:28.981270Z","closed_at":"2026-05-15T21:35:28.981068Z","close_reason":"Fixed by PR #208 (merged as f0a1fd53)"}
@@ -1014,7 +1021,7 @@
 {"id":"bd-i4wv","title":"L9 follow-up: real version string in feed generator (replace quarto-2)","description":"L9 v1 emits <generator>quarto-2</generator>. Q1 emits quarto-<version> from quartoConfig.version(). Replace with the actual Q2 version string when the version story stabilizes. Site: feed/binding.rs::FEED_GENERATOR.","status":"open","priority":4,"issue_type":"task","created_at":"2026-05-08T17:33:39.786912Z","created_by":"cscheid","updated_at":"2026-05-08T17:33:39.786912Z","dependencies":{"bd-o90m:discovered-from":{"depends_on_id":"bd-o90m","type":"discovered-from","created_at":"2026-05-08T17:33:39.786912Z","created_by":"cscheid"}}}
 {"id":"bd-i4xrcdqx","title":"qmd reader: whitespace-only blank line before ::: in a list item makes the list tight","description":"Pre-existing (reproduces on main at 790eaf89f; found while adding block editorial marks, bd-an9gkxnp).\n\n  printf -- '- item\\n  \\n  ::: {.x}\\n  y\\n  :::\\n' | pampa -t native\n  => BulletList [[Plain [Str \"item\"], Div ...]]   -- expected Para (loose list)\n\nWith an empty blank line ('- item\\n\\n  ::: {.x}') the item is Para, as expected, and a whitespace-only blank line before a *paragraph* ('- item\\n  \\n  para two') is also correctly loose. So only the whitespace-only-blank-line + fenced-div combination loses the blank line for looseness purposes.\n\nWhy it matters: the qmd writer emits exactly this shape (it indents the blank separator lines inside list items: '* item\\n  \\n  ::: {.x}'), so qmd -> qmd round-trips of a list item containing a div flip Para -> Plain. test_qmd_roundtrip_consistency catches it; the editorial_divs.qmd round-trip fixture leaves out its list-item case because of this. Fix could be in the scanner (blank-line detection under list continuation before ':::') and/or the writer (don't emit trailing indentation on blank lines).","status":"open","priority":2,"issue_type":"bug","created_at":"2026-09-24T15:46:55.254809Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-24T15:46:55.254809Z","labels":["parser"],"dependencies":{"bd-an9gkxnp:discovered-from":{"depends_on_id":"bd-an9gkxnp","type":"discovered-from","created_at":"2026-09-24T15:46:55.254809Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-i6jy4","title":"Preview eager-capture driver iterates all files (binaries, configs), causing thread panic on first binary","description":"## Symptom\n\n```\n$ q2 preview   # in a project containing any image/binary\nthread 'tokio-rt-worker' panicked at crates/quarto-parse-errors/src/error_generation.rs:121:35:\nstart byte index 7748 is not a char boundary; it is inside '�' (bytes 7746..7749 of string)\n```\n\nReproduces deterministically with the project at `docs/` (contains `quarto.png`).\n\n## Root cause\n\n`crates/quarto-preview/src/capture_driver.rs:55` (record_eager_captures) calls `ctx.index().get_all_files()` and iterates every entry. The index map covers *all* tracked project files: qmd, config (`_quarto.yml`), binaries (images), extension files, etc.\n\nThe doc comment on the function says: *\"Walk the project's .qmd files and record engine captures for any file that has code cells but no existing sidecar entry yet.\"* The code does not match the comment — it never filters by file kind. Each non-qmd file is shoved through `compute_input_qmd → parse-document`. Two distinct failure modes result:\n\n- `_quarto.yml` → parse-document errors out with \"Indented code blocks are not supported\". `record_one` returns `Err`, caught at line 109, logged as a WARN — visible noise but not crashing.\n- `quarto.png` → tree-sitter parses binary bytes, produces parse errors, then `produce_diagnostic_messages` panics on a non-UTF-8 byte (the lossy-string offset bug tracked in the sibling issue).\n\nThe PNG path also tells us the warning-only soft-fail at L109-115 doesn't catch panics — `compute_input_qmd` runs on a tokio worker, and the panic propagates as a worker-thread panic, not an `Err`.\n\n## Fix\n\nIterate `project_files().qmd_files` instead of `index.get_all_files()`. The currently-unused `_doc_id` from the index map can be looked up if needed, or dropped entirely if the downstream code doesn't need it (check `record_one` — `abs_path` and `rel_path` are the only inputs derived from the iteration item, both reconstructible from a qmd path).\n\nTests:\n- Regression test for the eager driver: project containing both a qmd and a binary, assert no panic and no spurious WARN for the binary.\n- Unit test that walks all `ProjectFiles` field combinations to confirm only `.qmd` files reach `record_one`.\n\nEnd-to-end verification: `q2 preview` on a fixture project with image asset must boot cleanly with no thread panic.\n\n## Out of scope\n\nDefensive hardening of `error_generation.rs` so that even a bug like this one cannot panic the worker. Tracked in sibling issue (link via discovered-from once filed). With the upstream filter in place, the panic still requires bad input to even reach that function.\n\n## Plan file\n\nTo be created at `claude-notes/plans/2026-05-19-bd-XXXX-capture-driver-qmd-filter.md` once issue ID is assigned.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-05-19T23:43:44.009707Z","created_by":"cscheid","updated_at":"2026-05-20T13:05:02.222953Z","closed_at":"2026-05-20T13:05:02.222785Z","close_reason":"Implemented in 408f4025 (merged via c8b4ab0b). record_eager_captures now iterates project_files().qmd_files. Verified by regression test capture_driver::tests::binary_files_are_not_fed_to_parse_pipeline and end-to-end: q2 preview on docs/ boots cleanly (no thread panic, no spurious _quarto.yml WARN). Defensive hardening of error_generation.rs tracked separately as bd-6qbto.","dependencies":{"bd-tnm3k:discovered-from":{"depends_on_id":"bd-tnm3k","type":"discovered-from","created_at":"2026-05-19T23:43:44.009707Z","created_by":"cscheid"}}}
-{"id":"bd-i8k7h6h1","title":"Revealjs deck footer/logo config regions not path-resolved like HTML footer","description":"bd-root-relative-paths-design-fc5pvkcv Phase 3 added rewrite_config_inlines to the html FooterRenderTransform and navbar title, so markdown images/links in page-footer regions resolve page-relative. The revealjs footer path (footer_render_stage's reveal branch, deck-level .footer/.slide-logo) builds its own footer and was not walked. A reveal deck rendered inside a website subdirectory with a site-root-relative footer image would still emit the raw path. Low priority: reveal decks are single-page outputs and deck-in-website-subdir footers are rare. Wire the same walk (or resolve_root_relative_resource_href for the slide logo) into the reveal footer transform when touching it next.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-13T18:50:51.443511Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-13T18:50:51.443511Z","labels":["websites"],"dependencies":{"bd-root-relative-paths-design-fc5pvkcv:discovered-from":{"depends_on_id":"bd-root-relative-paths-design-fc5pvkcv","type":"discovered-from","created_at":"2026-08-13T18:50:51.443511Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-i8k7h6h1","title":"Revealjs deck footer/logo config regions not path-resolved like HTML footer","description":"bd-root-relative-paths-design-fc5pvkcv Phase 3 added rewrite_config_inlines to the html FooterRenderTransform and navbar title, so markdown images/links in page-footer regions resolve page-relative. The revealjs footer path (footer_render_stage's reveal branch, deck-level .footer/.slide-logo) builds its own footer and was not walked. A reveal deck rendered inside a website subdirectory with a site-root-relative footer image would still emit the raw path. Low priority: reveal decks are single-page outputs and deck-in-website-subdir footers are rare. Wire the same walk (or resolve_root_relative_resource_href for the slide logo) into the reveal footer transform when touching it next.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-13T18:50:51.443511Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T23:16:52.507949Z","labels":["websites"],"dependencies":{"bd-root-relative-paths-design-fc5pvkcv:discovered-from":{"depends_on_id":"bd-root-relative-paths-design-fc5pvkcv","type":"discovered-from","created_at":"2026-08-13T18:50:51.443511Z","created_by":"Carlos Scheidegger"}},"comments":{"c-rph91hr9":{"id":"c-rph91hr9","author":"Carlos Scheidegger","created_at":"2026-10-28T23:16:52.507949Z","text":"The logo half is done: PR #958 (bd-7328beha) routes reveal `logo:` through config_paths::resolve_config_path, relativizes per page via ResourceResolverContext::page_url_for, and copies the file. Remaining scope is only the footer markdown walk (footer_slot_html in revealjs/footer_logo.rs builds navigation.footer.center via config_field_to_html without rewrite_config_inlines)."}}}
 {"id":"bd-i992","title":"WASM hub-client recompiles default Bootstrap SCSS on every keystroke","description":"After bd-imiw landed the Q1-parity default theme behavior, hub-client renders re-compile the full Bootstrap + Quarto SCSS layer on every keystroke because the new no-theme code path in CompileThemeCssStage bypasses the runtime cache (unlike the themed path), and the WASM version of compile_default_css has no in-memory cache (unlike the native version, which uses a OnceLock). Users report the editor freezing for tenths of a second after every keystroke on documents with no theme/navbar/footer.\n\nRoot cause evidence:\n- crates/quarto-core/src/stage/stages/compile_theme_css.rs:175-195 — no-theme path calls compile_default() without any cache_get check.\n- crates/quarto-core/src/stage/stages/compile_theme_css.rs:206-279 — themed path DOES use ctx.runtime.cache_get(\"sass\", ...) then cache_set.\n- crates/quarto-sass/src/compile.rs:55,239-276 — native compile_default_css uses DEFAULT_CSS_CACHE OnceLock.\n- crates/quarto-sass/src/compile.rs:357-384 — WASM compile_default_css deliberately has no in-memory cache (comment defers to JS SassCacheManager / IndexedDB).\n\nPlan: claude-notes/plans/2026-04-18-wasm-scss-cache-regression.md","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-04-18T20:30:49.365984Z","created_by":"cscheid","updated_at":"2026-04-18T21:16:33.355871Z","closed_at":"2026-04-18T21:16:33.355017Z","close_reason":"Fixed in commit d89d1514 on feature/navigation. Four-item bundle: (1) OnceLock in WASM compile_default_css, (2) stage routes no-theme path through runtime cache, (3) generational purge on SCSS_RESOURCES_HASH mismatch (ensure_namespace_version helper), (4) per-namespace LRU size cap with 10 MB sass budget. 16 new tests, 7550 workspace tests pass, cargo xtask verify green, manually confirmed on hub-client.","dependencies":{"bd-imiw:discovered-from":{"depends_on_id":"bd-imiw","type":"discovered-from","created_at":"2026-04-18T20:30:49.365984Z","created_by":"cscheid"}}}
 {"id":"bd-i9i5ad2t","title":"Pampa pandoc-oracle tests hard-fail on local pandoc newer than allowlist","description":"has_good_pandoc_version() in crates/pampa/tests/integration/test.rs uses substring matching against a hardcoded list (3.6-3.9) of pandoc versions. 4 tests (test_html_writer, test_json_writer, unit_test_corpus_matches_pandoc_commonmark, unit_test_corpus_matches_pandoc_markdown) assert on it directly and hard-FAIL instead of skipping when local pandoc is outside the list, rather than skip gracefully like the helper functions that call it internally. CI pins pandoc exactly via PANDOC_VERSION=3.8.3 in .github/workflows/test-suite.yml and is unaffected; only local dev machines with a newer or older pandoc are affected. Chris hit this on 2026-07-02 with pandoc 3.10 installed via scoop/winget.","design":"Keep the hard-FAIL signal, it is doing its job: the allowlist in has_good_pandoc_version() is a manual verification ledger (each version in 3.6-3.9 was added by a human confirming the 4 oracle tests still pass against it, see commit 12bca3b5), not a forward-compat assumption. Converting to a graceful skip would hide the signal permanently since nextest shows early-return-plus-eprintln tests as PASS, not SKIP, and stdout for passing tests is not shown by default. That would let the ledger silently rot. CI is unaffected either way since .github/workflows/test-suite.yml pins PANDOC_VERSION=3.8.3 exactly. Fix plan: 1) Replace the fragile version_str.contains(\"3.x\") substring matching in crates/pampa/tests/integration/test.rs with proper major.minor parsing, reusing/sharing logic with pandoc_version_at_least() in crates/xtask/src/dev_setup.rs:239 (currently only a >=3.6 floor check used for a dev-setup warning, no upper bound concept -- the two call sites currently disagree on philosophy and should share one parser). 2) Make the failure message actionable: print the detected version, the calibrated range, and the command to run to verify+bump. 3) Add a cargo xtask command (e.g. cargo xtask pandoc-check) that runs just the 4 oracle tests against the locally installed pandoc bypassing the version gate, and on green, edits the allowlist in test.rs to add the new minor version (or at minimum prints the exact line to add) so bumping is a single command instead of a remembered chore. This turns 'bitten on every local pandoc upgrade' into 'run one xtask command, it self-heals the allowlist when actually compatible, or tells you concretely what broke when not.'","status":"closed","priority":3,"issue_type":"bug","created_at":"2026-07-02T10:44:28.161073Z","created_by":"cderv","updated_at":"2026-07-03T08:01:36.991925Z","closed_at":"2026-07-03T08:01:36.991925Z","close_reason":"Merged in PR #361","comments":{"c-4b9tyrzv":{"id":"c-4b9tyrzv","author":"cderv","created_at":"2026-07-02T13:52:44.701480Z","text":"Rebased onto main (picked up PR #359 fnm/nvm-windows Node discovery and 7 other merged commits). Targeted verify clean post-rebase: clippy + nextest on pampa/xtask/quarto-mcp-launcher, same 8 pre-existing/by-design failures as before (4 pampa oracle tests fail against local pandoc 3.10, outside the calibrated range -- expected; 4 quarto-mcp-launcher cache_tests fail on an unrelated, previously-documented bundle-extraction issue).\n\nRan the new `cargo xtask pandoc-check` end-to-end against local pandoc 3.10:\n\n    Detected: pandoc 3.10\n    Calibrated range: 3.6-3.9\n\n    Nextest run ID ... with nextest profile: default\n    Starting 8 tests across 5 binaries (4037 tests skipped)\n        PASS [   0.056s] test_html_writer_context_include_source_locations\n        PASS [   0.127s] test_html_writer_context_no_source_locations\n        PASS [   0.178s] test_html_writer_context_include_source_locations\n        PASS [   0.194s] test_html_writer_context_no_source_locations\n        PASS [   2.545s] test::test_html_writer\n        PASS [   2.621s] test::test_json_writer\n        PASS [   2.633s] test::unit_test_corpus_matches_pandoc_commonmark\n        PASS [  40.862s] test::unit_test_corpus_matches_pandoc_markdown\n\n    All 4 oracle tests pass against pandoc 3.10.\n    Detected version is past the calibrated ceiling -- bump it:\n      crates/pampa/tests/integration/test.rs:114\n      const PANDOC_ORACLE_MAX_VERSION: (u32, u32) = (3, 10);\n\nCeiling bump to (3, 10) left as a follow-up decision, not applied in this commit (tool is print-only by design)."},"c-kucmfvca":{"id":"c-kucmfvca","author":"cderv","created_at":"2026-07-02T13:54:25.893269Z","text":"For the PR: what bumping the ceiling looks like (crates/pampa/tests/integration/test.rs:113-114). Before:\n\n    const PANDOC_ORACLE_MIN_VERSION: (u32, u32) = (3, 6);\n    const PANDOC_ORACLE_MAX_VERSION: (u32, u32) = (3, 9);\n\nAfter running `cargo xtask pandoc-check` green against pandoc 3.10:\n\n    const PANDOC_ORACLE_MIN_VERSION: (u32, u32) = (3, 6);\n    const PANDOC_ORACLE_MAX_VERSION: (u32, u32) = (3, 10);\n\nOne-line edit, no other code changes -- the range check and the 4 oracle tests pick it up automatically."},"c-mh5a17il":{"id":"c-mh5a17il","author":"cderv","created_at":"2026-07-02T14:26:03.903663Z","text":"Correction to the earlier verification comment: a roborev review caught that the nextest filter in cargo xtask pandoc-check wasn't scoped to the integration binary, so `test(test_html_writer)` substring-matched unrelated `writers::html::tests::test_html_writer_context_*` unit tests too -- the tool was actually running 8 tests while reporting \"4 oracle tests pass\". Fixed by scoping the filter to `binary(integration) & (...)`, with a regression test asserting the scoping. Corrected output:\n\n    Detected: pandoc 3.10\n    Calibrated range: 3.6-3.9\n\n    Nextest run ID ... with nextest profile: default\n    Starting 4 tests across 1 binary (1039 tests and 4 binaries skipped)\n        PASS [   2.783s] (1/4) pampa::integration test::test_html_writer\n        PASS [   2.883s] (2/4) pampa::integration test::unit_test_corpus_matches_pandoc_commonmark\n        PASS [   2.889s] (3/4) pampa::integration test::test_json_writer\n        PASS [  47.250s] (4/4) pampa::integration test::unit_test_corpus_matches_pandoc_markdown\n\n    All 4 oracle tests pass against pandoc 3.10.\n    Detected version is past the calibrated ceiling -- bump it:\n      crates/pampa/tests/integration/test.rs:118\n      const PANDOC_ORACLE_MAX_VERSION: (u32, u32) = (3, 10);\n\nUse this output (not the earlier comment's) for the PR."}}}
 {"id":"bd-ia3kzv4a","title":"q2 preview --static: stale-only startup render and a --render all","description":"FORMAT flag|Q1's project preview renders only inputs newer than their outputs at startup (serve.ts:952-1020) and has --render to force everything. Static mode always renders fully at boot; add a stale-only default once someone hits a slow boot. Deferred from claude-notes/plans/2026-09-22-q2-preview-static.md § Deferred.","status":"open","priority":4,"issue_type":"task","created_at":"2026-09-22T20:25:03.561308Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-22T20:25:03.561308Z","labels":["preview"],"dependencies":{"bd-sl79jjiq:discovered-from":{"depends_on_id":"bd-sl79jjiq","type":"discovered-from","created_at":"2026-09-22T20:25:03.561308Z","created_by":"Carlos Scheidegger"}}}
@@ -1038,12 +1045,13 @@
 {"id":"bd-include-parse-failure-dropped-u4rdjxru","title":"A failed include is dropped but the page is still written, with the included content silently missing","description":"When an {{< include >}} cannot be expanded, IncludeExpansion removes the block\nfrom the AST and the render continues. The page is written to disk missing the\ncontent the include was supposed to supply, and counts toward \"Rendered N of M\".\nQuarto 1 never produces such a page: a broken include aborts the project render\nand no output is written at all.\n\nThe clearest way to see this is a defect rather than a policy is q2's own\nbehaviour on the *same* parse error placed inline instead of in an included\nfile:\n\n  index.qmd   includes _broken.qmd, which has a Q-2-11 unclosed double quote\n              -> HTML written, included content missing\n  direct.qmd  the identical Q-2-11 inline on the page\n              -> no HTML written, document aborted\n  good.qmd    includes a file that parses\n              -> HTML written, content present\n  missing.qmd includes a file that does not exist\n              -> HTML written, included content missing\n\nSame diagnostic, same severity; opposite outcome depending only on which file\nit lives in. q2 summarises that project as \"Rendered 3 of 4 files — 3 errors,\n1 warning\", exit 1, with index.qmd counted among the three successes.\n\nThe missing-file case is the more alarming half. A nonexistent include target\nis only a warning (Q-17-2), so a project containing just that page renders\nexit 0 with no errors at all — a page that quietly lost its main content and\nreports nothing wrong.\n\nROOT CAUSE\n\ncrates/quarto-core/src/stage/stages/include_expansion.rs. All three failure\narms end the same way:\n\n    // See the circular-include arm for why the block is\n    // removed rather than skipped.\n    blocks.remove(i);\n    continue;\n\ncircular include (Q-17-1, warning, line 231), file not found or unreadable\n(Q-17-2, warning, line 254), parse failure (Q-17-3, error, line 324).\n\nRemoving the block is deliberate and correct — it is the fix from bd-qpvoamvu\n/ PR #465, which stopped the leftover shortcode from being misreported\ndownstream as an unrecognized `include`. This is NOT a request to put the block\nback. What is missing is the other half of that story: nothing downstream\ndistinguishes \"expanded every include\" from \"expanded some and deleted the\nrest\", so the writer produces the page as if the document were whole.\n\nThe Q-17-3 arm is otherwise in good shape — it registers the included file in\nboth SourceContexts, remaps the child parse's FileId(0) into the parent so the\ninner diagnostics render real ariadne snippets, and records the include for\ncache invalidation. Only the \"and now abandon the document\" step is absent.\n\nSUGGESTED FIX DIRECTION\n\nHave the failure arms mark the document as failed rather than merely pushing a\ndiagnostic, so an unexpandable include takes the same path an inline Q-2-11\nalready takes: no HTML written, counted against \"Rendered N of M\". Raise Q-17-2\nfrom warning to error at the same time — a page silently missing its content is\nnot a warning-grade outcome, and Quarto 1 treats a missing include as fatal.\n\nWhether a circular include (Q-17-1) should also abort is a separate judgement;\nthe arms share a code path but not obviously a policy.\n\nREAL-WORLD HIT\n\nThe Positron docs page migrate-rstudio-keybindings.qmd includes\n_rstudio-keyboard-shortcuts.qmd. Under an older q2 that included file failed to\nparse, and the page rendered while losing its entire 25-row keybinding table.\nThat specific instance is FIXED at 0.28.0 — the page now renders with 21 <tr>\nelements, matching the Quarto 1 baseline exactly, exit 0. The corpus therefore\nno longer contains a failing include, and the repro below is synthetic by\nnecessity. The swallow-and-continue behaviour it isolates is unchanged.\n\nREPRO\n\n  /Users/gordon/src/q2-positron-docs/llms-info/repros/include-parse-failure-dropped/\n\n  cd <that directory> && sh check.sh\n\ncheck.sh renders the four-page project with q2 and then with Quarto 1, and\nprints per page whether an HTML file was written and whether the included\ncontent reached it. It runs Quarto 1 twice: the first pass aborts the whole\nproject on missing.qmd's absent include and writes no _site at all, which is\nitself the Q1 reference behaviour; the second pass sets that page aside so Q1's\nhandling of the other three is visible.\n\nVERSION\n\nVerified at q2 0.28.0 and at q2 source main 85e98fb02 (no include-related\ncommits since the v0.28.0 tag). Quarto 1 reference: 99.9.9.\n\nOrigin strand: br-include-parse-failure-dropped-w3wlg2zl in the\nq2-positron-docs skein.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-09-03T02:19:29.084643Z","created_by":"Gordon Woodhull","updated_at":"2026-09-03T20:39:09.349645Z","closed_at":"2026-09-03T20:39:09.349645Z","close_reason":"Shipped in q2 0.29.0 (PR #648). Verified downstream by re-running llms-info/repros/include-parse-failure-dropped/check.sh: 'Rendered 1 of 4 files - 4 errors', and no HTML written for either the broken-include page or the missing-include page, where 0.28.0 wrote both with the included content silently absent. The good-include page still renders with its content.","labels":["includes","parity"],"comments":{"c-vnemxfm7":{"id":"c-vnemxfm7","author":"Gordon Woodhull","created_at":"2026-09-03T04:01:24.567989Z","text":"Fixed on braid/bd-include-parse-failure-dropped-u4rdjxru-include-parse-failure-dropped (2bd66e5b3), not pushed.\n\nAll three failure arms now fail the document, at both include positions (block-level and code-fence). Q-17-1/Q-17-2 raised warning -> error. No new diagnostic codes.\n\nDecisions beyond the strand:\n- Q-17-1 (flagged as a separate judgement) also aborts: Q1's standaloneInclude throws on a cycle as well as an unreadable target, and leaving one arm non-fatal restores the inconsistency the fix removes.\n- The code-fence arms move with the block-level ones, because Q-17-1/Q-17-2 are each emitted from two sites and one code cannot carry two severities.\n- Q-17-4 stays a warning (placement mistake, not a failed include).\n\nFound and fixed while verifying: resolve_single_file_deps in quarto-preview discarded the entire watch set when expansion errs, so one broken include would strand q2 preview on the broken version with the file being edited no longer watched. Factored the stage body into a public expand_document_includes() that hands the document back regardless of verdict.\n\nWorkspace: 13595 passed / 199 skipped, vs 13590/199 on main — +5, exactly the tests added."}}}
 {"id":"bd-indented-continuation-parse-error-j7be7kuc","title":"Regression in 0.18.0: indented continuation line is a hard parse error, dropping the whole page","description":"REGRESSION introduced in v0.18.0 by 92737cdd ('Fix digit/dash/plus-leading continuation line terminating the paragraph', bd-w6tod0gh). v0.17.0 rendered all of the affected inputs; v0.18.0 fails them with\n\n  Error: Parse error — unexpected character or token here\n\nand drops the ENTIRE page from the render, not just the offending block. HEAD (7c964559, the v0.18.0 release merge) still has it.\n\n## What happens vs Quarto 1\n\nQuarto 1 accepts all of these; q2 0.18.0 renders none of them.\n\n  1. one\n      1. nested a        <- ordered sub-list indented 4 under a '1. ' item\n\n  This value is in seconds, so to make license leases last\n    30 minutes you would use the following syntax.   <- indented digit-leading wrap\n\n  hello\n    -5 degrees           <- indented dash-leading prose\n    -- em dash\n\n  - one\n          - over-indented bullet\n\n## Root cause\n\nThe two new peek helpers in crates/tree-sitter-qmd/tree-sitter-markdown/src/scanner.c fail on the fall-through path when the line carries leading indentation. An indent sweep isolates it precisely — the error fires exactly when a NEW peek branch runs on an indented line and returns FALSE:\n\n  para + '  -5 degrees'    (peek_dash_plus_opens_block -> false)  ERROR\n  para + '  -- em dash'    (peek_dash_plus_opens_block -> false)  ERROR\n  para + '  - item'        (peek_dash_plus_opens_block -> true)   ok\n  para + '  *emph* here'   (pre-existing '*' peek)                ok\n  para + '  `code` here'   (pre-existing backtick peek)           ok\n  para + '-5 degrees'      (same text, indent 0)                  ok\n\nThe pre-existing '`' and '*' peeks are fine at the same indentation, so this is specific to the two branches added by 92737cdd, not to peeking in general.\n\nThe suspected mechanism is the mark_end bookkeeping at the end of gate 2 (scanner.c ~line 3059). When second_peeked is true, mark_end is deliberately skipped so the peeked run is not swallowed into the SOFT_LINE_ENDING token's range, leaving the range 'where the earlier mark_end put it'. For the '`' and '*' branches that was pre-existing behavior and the earlier mark_end is in the right place; for the two new branches the earlier mark_end appears to sit before the line's consumed indentation, so the parser resumes at a position it cannot reconcile. Indent 0 works because there is no indentation to strand.\n\nSecond, independent defect in the same commit: peek_dash_plus_opens_block (scanner.c ~1357) has no indentation guard, while peek_ordered_marker (~1313) opens with 'if (s->indentation > 3) return result;'. So a '- b' line indented 4 past the enclosing item's content column — indented-code territory, where no list marker can form — still reports that it opens a block. That is what breaks the 'over-indented bullet' case above.\n\n## Suggested fix direction\n\nGive peek_dash_plus_opens_block the same 's->indentation > 3' early-out peek_ordered_marker already has, and make the skipped-mark_end path in gate 2 restore the position explicitly for the new branches rather than relying on an earlier mark_end whose placement only happens to be correct for '`' and '*'. The indent dimension is missing from the test corpus: paragraph.txt cases 15-29 added by 92737cdd all sit at indent 0, which is why this got through 572/572 tree-sitter tests.\n\n## Real-world hit\n\nIn the Posit Connect documentation port, api/index.qmd — the entire API reference, 53k lines — fails to render, taking the project to 351 of 352 files with 3 errors, plus a cascading [Q-13-2] 'Navbar references missing document' warning on all 351 surviving pages. The offending markdown is the endpoint life-cycle nested ordered list in Connect's own OpenAPI spec (api/public/openapi.yaml:8245-8250). That spec text ships to customers as a downloadable artifact, so reformatting it on the docs side is not an option.\n\n## Repro\n\n  /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/indented-continuation-parse-error/\n\nRun 'q2 render' there: 0 of 3 files render, with 5 parse errors. Run 'quarto render --output-dir _site-q1' for the Quarto 1 reference (all 3 render). The README carries the full indent-by-parent-type sweep table.\n\nObserved with q2 0.18.0. Origin strand in the q2-connect-docs skein: br-8wvj5ang.","status":"closed","priority":0,"issue_type":"bug","created_at":"2026-08-11T23:15:04.354553Z","created_by":"Claude (q2-connect-docs)","updated_at":"2026-08-12T14:44:41.894639Z","closed_at":"2026-08-12T00:38:18.130328Z","close_reason":"Div-context regression fixed in 062f4f88 and pushed to PR #510. Full xtask verify green; repro variants byte-identical to v0.18.0; div-wrapped nested list improves on 0.18 (parses where 0.18 errored).","labels":["parser"],"comments":{"c-1rt1h9nm":{"id":"c-1rt1h9nm","author":"Carlos Scheidegger","created_at":"2026-08-12T00:25:58.488597Z","text":"REOPENED: branch regression found by the q2-connect-docs agent (repro: llms-info/repros/attributed-div-ws-blank-list). A nested ordered list inside an ATTRIBUTED fenced div silently renders as paragraph text. Isolated: old scanner + new grammar is CORRECT, so the cause is Part B, not the grammar prec.right the repro README suspected. Two defects: (1) claimable_list_indentation breaks its run at FENCED_DIV, but FENCED_DIV/FENCED_CODE_BLOCK/ANONYMOUS match() without consuming anything — they are transparent and must be skipped, so even '::: {.x} / * a / [4sp]1. b' loses the nested list with no whitespace-only line involved; (2) the over-indent verdict at both gates uses the absorbing mark_end emission (else branch), which makes a bogus paragraph-continuation GLR fork viable across whitespace-only lines (token log: SOFT_LINE_ENDING size 7 swallowing ws-line remainder + newline + next line's indent); the old code's skip-mark_end emission left that fork dead. Fixing both here with tests first."},"c-4oz3w0zf":{"id":"c-4oz3w0zf","author":"Carlos Scheidegger","created_at":"2026-08-11T23:26:40.304966Z","text":"Independent verification + mechanism refinement (session in main q2 checkout, HEAD 3640eae9).\n\nAll sweep rows reproduce exactly as reported. Two additional data points sharpen the mechanism:\n\n1. The suspected cause (skipped mark_end in the peeked soft-break path leaving the token range pre-indent) is CONFIRMED as the trigger, but the reason the pre-existing backtick/star peeks survive is NOT better mark_end placement — their placement is identical. They survive by accident: their continuation lines begin with EXTERNAL-scanner tokens (code-span start, emphasis open) which re-consume the residual whitespace into their own token range on the next scan (CST shows code_span_delimiter spanning cols 0-3 over '  `'). A prose continuation instead hits the INTERNAL lexer, which lexes _whitespace — and the grammar state immediately after _soft_line_ending cannot shift _whitespace (tree-sitter trace: 'detect_error lookahead:_whitespace' at state 2419). So the precise failure condition is: (a) peeked gate branch judges the line prose, (b) after block_continuation consumes any open-block prefix columns there is RESIDUAL leading whitespace, (c) the next token is internal. Evidence: '- a\\n  -5 degrees' PASSES (block_continuation size 2 consumes the indent exactly, no residue), while '1. one\\n    1. nested' fails with exactly 1 residual space after block_continuation size 3.\n\n2. peek_ordered_marker's 'indentation > 3' guard has different semantics at the two call sites: at gate 1, s->indentation is the RAW column count (match_line has not consumed block prefixes yet), so a legitimate nested marker at raw indent 4 inside a list is misjudged as prose — that is what routes the nested-ordered-list case into the broken soft-break path at all. At gate 2 (post-match_line) the same guard reads post-prefix indentation. Any fix copying the guard into peek_dash_plus_opens_block must account for this, and note the guard alone does NOT fix the over-indented-bullet case — it would just reroute it from 'phantom block opener' into the same residual-whitespace parse error, so the mark_end/residual-indent defect must be fixed first (or a grammar-side fix: allow optional _whitespace after _soft_line_ending in paragraph continuation, which also matches pandoc's stripping of continuation-line leading whitespace).\n\n3. Micro-defect: the gate-1 digit branch sets first_peeked=true even when peek_ordered_marker bailed on the indentation guard WITHOUT advancing — in that case the else branch (mark_end at post-indent) would have been taken and the input would parse. first_peeked should mean 'the peek advanced past the delimiter run', not 'a peek function was called'.\n\nTest-corpus gap confirmed: paragraph.txt cases added by 92737cdd are all at indent 0. Any fix should add an indent x context sweep (indent 0/1/3/4+ x top-level/list-item/blockquote x dash/plus/digit/backtick/star) to the corpus first, TDD-style."},"c-59dg54br":{"id":"c-59dg54br","author":"Carlos Scheidegger","created_at":"2026-08-12T00:34:19.191987Z","text":"Div-context regression fixed in 062f4f88. (1) claimable_list_indentation now skips transparent blocks (FENCED_DIV/FENCED_CODE_BLOCK/ANONYMOUS — they match() without consuming) instead of stopping, so div-wrapped nested markers are judged correctly; '::: {.x} / * a / [4sp]1. b' now nests where v0.18.0 hard-errored. (2) The over-indent verdict at both gates uses the peeked skip-mark_end emission again (the absorbing emission kept a bogus paragraph-continuation fork alive across whitespace-only lines). Repro variants now byte-identical to v0.18.0 (verified against a 7c964559 build); canonical-read roundtrip is a fixed point; q2 render of the repro site shows the <ol> restored, and the original indented-continuation repro still renders 3/3. Plain-vs-Para on ws-only separator lines is a pre-existing 0.18 tightness quirk -> bd-25ezfgs8. tree-sitter 582/582, workspace 11728/11728, zero snapshot churn. Full xtask verify running; will close + push to PR #510 on approval."},"c-6clyz2iz":{"id":"c-6clyz2iz","author":"Carlos Scheidegger","created_at":"2026-08-12T00:05:17.144588Z","text":"PR: https://github.com/quarto-dev/q2/pull/510"},"c-c5vlbqc9":{"id":"c-c5vlbqc9","author":"Claude (q2-connect-docs)","created_at":"2026-08-12T00:14:05.832266Z","text":"Verified this branch (bcd02be9) against the full Posit Connect docs port — the original bug is fixed, and the render is the best this project has seen: **352 of 352 files, exit 0, zero errors, 352 warnings**, all of them the single acknowledged Q-14-3 dark-theme gap. api/index.qmd (53k lines) is back, and the cascading Q-13-2 navbar warning is gone. The 77-cell indent sweep from the origin repro is completely clean, and the repro's 3 pages now match Quarto 1's text exactly.\n\n**One regression to flag before this ships.** A nested ordered list renders as literal paragraph text ('1. Exact.' in a <p>) instead of an <ol> when all three hold:\n1. the list is inside a fenced div THAT CARRIES ATTRIBUTES (::: {.definition-list}; a bare ::: does not trigger it);\n2. the separator line between the bullet text and the ordered list is whitespace-only rather than truly empty;\n3. the ordered list is nested inside a bullet item.\n\nControls that render correctly on this same branch: truly-blank separator lines, and the same whitespace-only lines with no div around them. The 0.18.0 release and Quarto 1 both handle all three shapes.\n\nMinimal case:\n\n    ::: {.definition-list}\n\n    * Find:\n    <-- this line holds four spaces\n        1.  Exact.\n    :::\n\nSuspected cause is Part A: _soft_line_break absorbing an optional trailing _whitespace with prec.right, 'so it wins over _attr_ws / _shortcode_sep shapes'. The trigger requires an *attributed* div — exactly the _attr_ws shape that precedence was tuned against.\n\nConnect docs impact: 3 pages (admin/r, admin/quarto, admin/python), all the same .definition-list version-matching pattern, each losing a 5- or 3-step ordered list. These are the only new content differences the branch introduces — the rest of the 352-page text-diff sweep is unchanged.\n\nRepro: /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/attributed-div-ws-blank-list/ (section A broken, B and C controls). WARNING: four lines of index.qmd contain four trailing spaces and are the trigger; an editor that strips trailing whitespace will make it stop reproducing. Tracked in the q2-connect-docs skein as br-vqz9n7si."},"c-etetoyhq":{"id":"c-etetoyhq","author":"Claude (q2-connect-docs)","created_at":"2026-08-12T00:38:55.721729Z","text":"Re-verified 062f4f88 against the Connect docs with a fresh release build. Both the attributed-div regression and the original bug are fixed, and the docs are clean:\n\n- attributed-div-ws-blank-list repro: <ol> in all 3 sections (was 2 of 3 at bcd02be9).\n- indented-continuation-parse-error repro: 3 of 3 files render.\n- full docs render: 352 of 352 files, exit 0, ZERO errors, 352 warnings — all of them the single acknowledged Q-14-3 dark-theme gap, i.e. one distinct diagnostic across the whole site.\n- 352-page text-diff sweep vs the Quarto 1 reference: 47 pages with real net content differences, every one mapped to an existing strand or a documented benign bucket. Compared against the 0.18.0-release baseline the ONLY newly-differing page is api/index.html — which simply did not exist before, since it was the page the parse error killed. Its diff is entirely two already-documented benign buckets (straight-vs-curly apostrophes in the error-code table, and Q1 running table-cell prose together where q2 emits separate blocks). admin/r, admin/quarto and admin/python are back off the list.\n\nI did not re-run your 280-cell sweep, only the 77-cell one from my repro, which is clean.\n\nOn the residual you flagged: I see it too — sections A and C of the repro render the bullet text tight (<li>Find:) where Q1 renders it loose (<li><p>Find:</p>). Agreed it is out of scope here and pre-existing; noted in the repro README as bd-25ezfgs8. It is invisible to our site-text-diff tooling since the extracted text is identical, so it costs the docs nothing today.\n\nNothing else from this end — this looks ready to ship. Local strand br-vqz9n7si closed; br-8wvj5ang stays open until the fix reaches a released binary."},"c-lqbcm5ub":{"id":"c-lqbcm5ub","author":"Carlos Scheidegger","created_at":"2026-08-12T14:44:41.894639Z","text":"Shipped in v0.19.0 (tag 1ba0f2ec): https://github.com/quarto-dev/q2/releases/tag/v0.19.0 — verified against the published darwin_arm64 artifact installed via the real install.sh one-liner: the four regression shapes (indented digit continuation, over-indented 1./- markers in bullet + quote context, legitimately-nested 4-space 1.) all render, exit 0. The same probe still hard-errors at 6:1 under v0.18.0."},"c-y9nxjvv6":{"id":"c-y9nxjvv6","author":"Carlos Scheidegger","created_at":"2026-08-11T23:59:02.045724Z","text":"Fix implemented on branch bugfix/bd-indented-continuation-parse-error-j7be7kuc (commits e63f5831 setup, bcd02be9 fix).\n\nTwo-part fix, per plan claude-notes/plans/2026-08-11-indented-continuation-parse-error.md:\n- Part A (grammar): _soft_line_break absorbs an optional trailing _whitespace under prec.right, so residual continuation-line indentation is shiftable everywhere a soft break occurs. This alone fixed all 96 prose cells of the characterization table.\n- Part B (scanner): marker peeks are shape-only; indentation judgment moved to the call sites — gate 1 compares raw indent against claimable_list_indentation()+3 (new helper: leading LIST_ITEM* run of the stack) and skips the peek on over-indent; gate 2 compares the post-match_line residual against 3 on both paths. Fixes the nested-marker misread, the missing dash/plus guard, and the first_peeked-without-advance bug.\n\nResults: tree-sitter 582/582 (10 new corpus cases add the missing indent dimension); new 138-cell integration table green (fail-first verified: 72+27+3 failing cells matched the sweep prediction exactly); cargo nextest run --workspace 11727/11727 with zero snapshot churn; error table regenerated. E2E: the Connect-docs repro project renders 3/3 files (was 0/3), output inspected — digit wrap joins into one <p>, nested <ol> present. Post-fix 280-cell sweep: only the 28 deliberate *5 unclosed-emphasis cells still error.\n\nDiscovered + filed: bd-mt1ksg9b (star peeks share the over-indent defect, pre-existing), bd-z69hr4o0 (spurious Space after SoftBreak for backtick/star continuations, pre-existing).\n\nProcess note added: claude-notes/instructions/scanner-indentation-contexts.md. Awaiting full cargo xtask verify (WASM leg) before close."}}}
 {"id":"bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe","title":"Inline R: the documented `{r} expr` spelling is never evaluated, and the native `r expr` spelling gets the wrong escaping default","description":"Quarto documents two spellings for inline executable code with deliberately different semantics, and q2 implements neither correctly. The brace spelling `{r} expr` — the cross-engine form quarto-web has recommended since 1.4 — is never evaluated at all. The native knitr spelling `r expr` is evaluated, but with the brace form's escaping default rather than its own.\n\nWHAT HAPPENS VS QUARTO 1\n\nSpelling. `{r} release_version` in prose renders as an ordinary code span, <code>{r} release_version</code>. Inside an attribute value it survives as the literal backticked text. No diagnostic of any kind is emitted and the render exits 0. Q1 evaluates it in prose, in a fenced-div attribute value and in a link title. Verified on q2 0.29.0, 0.28.0, 0.27.0 and 0.24.0 — long-standing, not a regression.\n\nEscaping. quarto-web docs/computations/inline-code.qmd specifies the two defaults. Section 'Markdown Output': the brace form escapes markdown specials in the value by default, and the engine-specific opt-in for markdown is I() for knitr. Section 'Syntax Compatibility > Knitr': the native spelling 'treats all inline output as *containing markdown*', with the equivalence stated outright — `r radius` is equivalent to `{r} I(radius)`. q2 inverts the native default. With release_version <- '**bold**':\n\n  case                  q2 0.29.0       Quarto 1     docs say\n  classic bare          escaped         <strong>     markdown\n  classic I()           <strong>        <strong>     markdown\n  brace   I()           not evaluated   <strong>     markdown\n\nQ1 matches the spec on all three. q2 passes the middle row only because .QuartoInlineRender passes AsIs objects through untouched — i.e. the one case it gets right, it gets right for the brace form's reason. Read the diagonal: q2's classic spelling behaves like Q1's brace spelling. q2 has one inline spelling with one semantics, and it is the brace semantics wearing the classic spelling.\n\nROOT CAUSE\n\ncrates/quarto-core/src/engine/knitr/preprocess.rs:136:\n\n  Regex::new(r\"(^|[^`\\\\])`r[ \\t]([^`]+)`\")\n\nresolve_inline_r_expressions rewrites each match to `r .QuartoInlineRender(expr)`. There is no alternation for {r}, so an unmatched brace form is never rewritten and arrives at the markdown parser as a code span. The pass runs over the serialized document (stage/stages/engine_execution.rs:432,495) before knitr sees it; its only callers are engine/knitr/mod.rs:158-159. Blast radius is one module.\n\nQuarto 1 is not a superset here, it is the mirror image, which is how both halves went missing at once:\n\n- quarto-cli/src/core/execute-inline.ts, executeInlineCodeHandler(language, exec), builds \"(^|[^`])`{\" + language + \"}[ \\t]([^`]+)`\" — the BRACE form and nothing else. src/execute/rmd.ts:576 calls it with 'r' from the knitr execute: handler at rmd.ts:287, rewriting matches to `r .QuartoInlineRender(expr)`.\n- The native `r expr` is knitr's own rmarkdown syntax, predating Quarto. Q1 never matches it. knitr evaluates it downstream, and since Q1 installs no knit_hooks$inline, .QuartoInlineRender never touches it — which is exactly why the native form passes markdown through.\n\nSo Q1's only wrapper insertion is on the brace form, and q2's regex covers exactly the spelling Q1's Quarto layer does not. preprocess.rs's doc comment cites execute-inline.ts as the source of its (^|[^`]) guard: the guard was transplanted off the brace-form matcher onto a classic-form regex, and the spelling went with neither.\n\nNote that q2's engine/knitr/resources/rmd/execute.R is byte-identical to Q1's src/resources/rmd/execute.R, so .QuartoInlineRender itself is unmodified. The whole divergence is in which strings get wrapped.\n\nSUGGESTED FIX\n\nThe documented equivalence gives the contract directly, with no new evaluation machinery:\n\n  `{r} x`  ->  `r .QuartoInlineRender(x)`     (escape by default)\n  `r x`    ->  `r .QuartoInlineRender(I(x))`  (markdown by default), or left\n                                                unwrapped as Q1 leaves it, which\n                                                is the same thing\n\nBoth halves must preserve the existing guards. The (^|[^`\\\\]) prefix guard and the [ \\t] class are load-bearing against display fences — without them a ```r fence anchors a match on its third backtick and the block body is swallowed into .QuartoInlineRender(...), costing the whole page (bd-knitr-inline-r-eats-fence-2ofk91x1). Any brace-form alternation needs to clear the same bar, and the file's regression tests (test_fence_inside_yaml_scalar_not_matched, test_escaped_backtick_fence_not_matched, test_backtick_prefixed_not_matched) are the fence.\n\nWorth deciding explicitly: the escaping half is a behaviour change for existing documents that use `r expr` with values containing markdown specials. It is the documented behaviour and matches Q1, but it is a change from what q2 ships today, and no test currently pins either outcome — all 15 preprocess.rs tests assert regex matching only.\n\nRELATED\n\nbd-u996g8g2 is the jupyter half of the same gap ({python} expr missing from the production text path); it records 'knitr handles inline r expressions natively', which is true of the classic spelling only and is why no R-side counterpart was ever filed. bd-knitr-inline-r-eats-fence-2ofk91x1 owns the same regex and notes in passing that 'Q1's handler covers the braced {r} expr form rather than bare r expr, so it is not a strict 1:1 port'.\n\nNothing in q2's plans, docs or skein states an intent to diverge from Q1 here. Searched: all 1331 claude-notes/*.md (every file mentioning inline R / inline expressions / QuartoInlineRender read), q2 docs/ (inline expressions are not documented; only Q-2-47 and Q-2-50 mention {r}), and the full skein export, 2109 strands by title and description with comments checked on the seven closest. The nearest statements all cut toward parity: claude-notes/plans/2026-08-25-r-display-fence-parse-error.md calls the missing brace form 'a separate parity gap, not this bug'. Going the other way, claude-notes/research/2026-07-04-q1-vs-q2-knitr-jupyter-baseline.md audits the two implementations and records the opposite of what is true — 'Equivalent in both: inline `r expr` wrapping'.\n\nREAL-WORLD HIT\n\nThe Positron website port, download.qmd:100:\n\n  ::: {#download-hero .download-hero data-version=\"`{r} release_version`\"}\n\nThe rendered page carries data-version=\"`{r} release_version`\" where Q1 has data-version=\"2026.08.1-2\". That attribute is read by the page's own JavaScript — download.qmd's detectOS() looks up #download-hero and builds the platform download URLs from the version — so the page's primary function breaks. It is the only inline R on the site, which is why one character of source costs the download page.\n\nThe defect is invisible to every comparative sweep that port runs: text diffing compares visible text and this is an attribute; the asset checker and lychee follow references and data-version is not one; the render log is silent, no diagnostic, exit 0. It was found by reading download.html's markup directly.\n\nREPRO\n\n/Users/gordon/src/q2-positron-docs/llms-info/repros/inline-r-brace-spelling-not-evaluated/\n\n  sh check.sh                                  # q2 and Quarto 1\n  sh check.sh ~/.local/bin/q2.0.28.0-backup    # or any other build\n\nTwelve cases in one render: B1-B3 and C1-C3 are three positions (prose, fenced-div attribute value, link title) x two spellings, where the classic column is the control that proves attributes are not the problem; E1-E3 are the escaping defaults; I1-I3 test the documented I() equivalence directly. The README states expected vs actual and carries the full root-cause and provenance analysis.\n\nTRAP FOR THE FIXER, found by walking into it: the preprocessing pass runs over the whole serialized document, so an inline expression written inside explanatory PROSE is executed like any other. Adding a sentence containing a literal backtick-r-x to the repro fixture killed the render with \"object 'x' not found\". Use the escaping quarto-web prescribes (a doubled brace, or an extra enclosing backtick) when writing about this syntax, in fixtures and in docs alike.\n\nOrigin strand: br-inline-r-brace-spelling-not-evaluated-kucihhme in the Positron docs port's skein (/Users/gordon/src/q2-positron-docs). Observed on q2 0.29.0 (official release build); q2 HEAD checked at 914f20697, nothing has touched crates/quarto-core/src/engine/knitr/ since the 0.29.0 tag. Q1 reference: quarto dev build 99.9.9.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-09-04T14:28:19.538303Z","created_by":"gordon","updated_at":"2026-09-04T19:18:16.045213Z","closed_at":"2026-09-04T19:18:16.045213Z","close_reason":"Merged to main as 3669ab053 via PR #658, stacked on #657 (bd-0gwekaem, merged as 3177c5615). Note the SHAs: gh stack merge --rebase replayed both commits, so the 8c3b591de cited in the earlier comment never reached main.\n\nBoth inline spellings now behave as quarto.org documents them. `{r} expr` evaluates and escapes markdown specials in the value; `r expr` evaluates and inserts the value as live markdown, with `I()` bridging the two. All twelve repro cases match Quarto 1, and the rendered markup is byte-identical to Q1's on every measured attribute -- including the download-hero shape that started this.\n\nThree behaviour changes shipped, not the one the strand anticipated: the escaping change, `r NULL` rendering empty rather than the literal NULL, and a knitr document displaying the {r} syntax now failing if the expression doesn't resolve (all three match Q1). The third is only survivable because #657 landed with it -- the nested-cell mask now covers inline expressions, so a markdown block suppresses execution.\n\nCI, dispatched by hand because stacked PRs match no pull_request trigger (all three workflows filter branches: [main, kyoto]): Test Suite and TS Test Suite green on ubuntu, macOS and WASM. Locally on the stacked tree: workspace 13719/13719, cargo xtask verify, clippy -D warnings, xtask lint.\n\nDocs: guides/authoring/computations.qmd now documents inline expressions for the first time.\n\nOpen follow-ups: bd-tiidc899 (whether to widen the display-class predicate past markdown/empty) and bd-u996g8g2 (the {python} half -- #657's MASK_INLINE_RE already covers any [A-Za-z0-9_]+, so the mask side is ready).","labels":["engine","parity"],"dependencies":{"bd-knitr-inline-r-eats-fence-2ofk91x1:related":{"depends_on_id":"bd-knitr-inline-r-eats-fence-2ofk91x1","type":"related","created_at":"2026-09-04T14:29:46.235351Z","created_by":"gordon"},"bd-u996g8g2:related":{"depends_on_id":"bd-u996g8g2","type":"related","created_at":"2026-09-04T14:29:42.934348Z","created_by":"gordon"}},"comments":{"c-b86cicpz":{"id":"c-b86cicpz","author":"Gordon Woodhull","created_at":"2026-09-04T18:41:38.675503Z","text":"Filed as PR #658 -- https://github.com/quarto-dev/q2/pull/658\n\nSTACKED, and the merge order matters. #658's base is bugfix/bd-0gwekaem-nested-cell-mask-inline-expressions (PR #657), not main. #657 must merge first. If #658 were retargeted at main and merged alone, docs/guides/authoring/computations.qmd would tell authors that wrapping an inline expression in a markdown block displays it rather than running it, which is only true once #657's mask extension is in.\n\nBranch braid/bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe-... rebased onto 8e13ecf8b; single commit 8c3b591de; remote feature/bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe. Clean rebase, no conflicts -- no file overlap with #657.\n\nONE DOCS ERROR CAUGHT BY THE STACK, worth recording because nothing in this branch would have caught it. The section originally said 'a fenced code block is displayed verbatim, so an expression written inside one is shown, not run'. #657 deliberately kept the display-class predicate narrow (info string empty or 'markdown'), so that sentence was wrong for any language-labelled fence. Measured on the stacked build: brace and classic spellings are both displayed verbatim inside a markdown block and inside a bare fence, and both still execute inside a ```r block. The prose now says 'markdown block, or a fence carrying no language at all' and warns about the labelled case. Widening the predicate is bd-tiidc899.\n\nGate on the stacked tree: workspace 13719/13719, cargo xtask verify (--skip-hub-build --skip-hub-tests --skip-rust-tests) all steps passed, clippy -D warnings clean, xtask lint clean, and #657's t31_masked_text_is_not_claimed_by_the_inline_scanner green under this branch's alternation."},"c-c3lnpj4i":{"id":"c-c3lnpj4i","author":"Gordon Woodhull","created_at":"2026-09-04T17:19:48.681700Z","text":"Fix ready on branch braid/bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe-inline-r-brace-spelling-not-evaluated (worktree .worktrees/workspace-3), one commit c95012e99. Not pushed, no PR yet -- see the blocker below.\n\nWHAT LANDED\nPattern gains a spelling-marker group and the replacement branches on it: `{r} expr` -> `r .QuartoInlineRender(expr)`; `r expr` goes to knitr unwrapped with only its separator normalized to a space (knitr's class is [ #], so a tab-separated expression would otherwise stop evaluating). The suggested `.QuartoInlineRender(I(expr))` form was rejected: I(NULL) errors in R, so it would turn `r NULL` into a failed render. Leaving the expression alone reaches the same markdown-passthrough semantics via knitr, which is also exactly what Q1 does.\n\nAll twelve repro cases now match Quarto 1, and the rendered markup is byte-identical to Q1's on every measured attribute including the download-hero shape.\n\nTHREE BEHAVIOUR CHANGES, not one -- the strand anticipated only the escaping change:\n1. `r expr` with markdown specials now renders as markdown (the documented default).\n2. `r NULL` renders empty where it rendered the literal NULL, because knitr's hook replaces the wrapper on this path. Verified byte-identical to Q1.\n3. A knitr document that displays the {r} syntax now fails the render if the expression doesn't resolve. Q1 fails identically.\n\nCORRECTION TO THE STRAND'S 'TRAP FOR THE FIXER'\nIt prescribes 'a doubled brace, or an extra enclosing backtick' for writing about this syntax. Measured: the extra-backtick form does NOT protect anything, in q2 or in Q1 -- `` `{r} x` `` still evaluates. Neither does a display block. Only the doubled brace works, and q2 renders it verbatim ({{r}}) where Q1 collapses it to {r}.\n\nBLOCKED ON bd-0gwekaem\nThat gap is filed as bd-0gwekaem and dispatched. The docs in this commit teach the fenced-block rule, which only becomes true once the nested-cell mask covers inline expressions, so this branch is meant to stack on that fix rather than merge to main first.\n\nTests: 27 -> 46 unit tests in preprocess.rs plus a new 8-test render-level suite. Workspace suite 13703/13703, clippy -D warnings clean, cargo xtask lint clean."}}}
-{"id":"bd-ip1lrgra","title":"Lua API: quarto.project table with .profile (active project profiles)","description":"Q1 exposes quarto.project.profile (list of active project profiles) plus directory/offset/output_directory to Lua filters via init.lua. Q2's pampa Lua runtime has no quarto.project table at all (checked 2026-08-10). When one is added, .profile should read ProjectConfig.active_config_profiles (bd-fu16z22k). Note Q1's type stub declares 'profiles' (plural) which doesn't exist at runtime — implement the singular.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-08-10T17:37:45.313708Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-10T17:37:45.313708Z","dependencies":{"bd-fu16z22k:discovered-from":{"depends_on_id":"bd-fu16z22k","type":"discovered-from","created_at":"2026-08-10T17:37:45.313708Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-ip1lrgra","title":"Lua API: quarto.project table (directory, offset, output_directory, profile) + QUARTO_PROFILE via os.getenv (GH #939)","description":"Quarto 1 exposes quarto.project to Lua filters with directory, offset (relative path from the document's directory back to the project root, forward slashes, '.' when same dir), output_directory, and profile (a pandoc.List of active profile names). q2 defines no quarto.project at all, so any Q1 filter indexing it fails with 'attempt to index a nil value (field project)'. Also, os.getenv('QUARTO_PROFILE') is nil under --profile drafts (only set when it was in the real environment), while Q1 wrote the active list back into the environment. Reproduced 2026-10-27 on main at 3f7d4b0e8 (GH #939, reported by atheriel porting the Connect docs).\n\nFix: add HostContext { input_file, output_file, project: Option<HostProject{directory, output_directory, offset}>, profiles, env: Option<Arc<HashMap>> } to pampa's LuaOptions; quarto-core fills it in a new lmd::lua_options_for_document used by UserFiltersStage and ShortcodeResolveTransform (offset via pathdiff + quarto_util::to_forward_slashes, exactly Q1's projectOffset). register_quarto_project builds the table (nils for single-file renders, profile always a List).\n\nos.getenv (native stdlib os table and WASM synthetic one, shared helper) becomes a pure lookup in an environment SNAPSHOT the host builds on the orchestrator thread in StageContext::new, next to load_project_environment, stored as env_snapshot: Arc<HashMap<String,String>>: (1) real environment via std::env::vars_os, (2) file-defined _environment pairs for keys the real env lacks (same filter as env_for_subprocess; real env wins), (3) QUARTO_PROFILE forced to quarto_profile_env_value(active) when non-empty (the documented deliberate exception). Why a snapshot: filters run on tokio worker threads under block_in_place and libc getenv races with any setenv in the process (glibc UB, not covered by Rust's std::env lock); reading once before the pipeline fans out is the defensive choice, and every script in a render sees one consistent environment (Carlos, 2026-10-27). Layer (2) is a small, agreed step past the issue (Q1 loaded _environment into the Deno process env, so pandoc's Lua saw it). No snapshot (pampa tests, conformance suite) -> fall through to today's behavior. This strand also carries the plumbing #940 depends on.\n\nNote from the original strand: Q1's type stub declares 'profiles' (plural) which does not exist at runtime; implement the singular 'profile'.\n\nPlan: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md","status":"in_progress","priority":2,"issue_type":"feature","created_at":"2026-08-10T17:37:45.313708Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-27T21:54:38.274658Z","external_ref":"https://github.com/quarto-dev/q2/issues/939","labels":["pampa","parity"],"dependencies":{"bd-fu16z22k:discovered-from":{"depends_on_id":"bd-fu16z22k","type":"discovered-from","created_at":"2026-08-10T17:37:45.313708Z","created_by":"Carlos Scheidegger"},"bd-pqj37chw:parent-child":{"depends_on_id":"bd-pqj37chw","type":"parent-child","created_at":"2026-10-27T20:22:30.406822Z","created_by":"claude"}},"comments":{"c-45zpfdec":{"id":"c-45zpfdec","author":"claude","created_at":"2026-10-27T21:54:38.274658Z","text":"PR: https://github.com/quarto-dev/q2/pull/949 (feature/lua-api-q1-parity -> main; closes GH #939-#942 on merge). Full cargo xtask verify green incl. hub-client tests on a rebuilt WASM; Windows cross-check not possible from macOS, flagged in the PR."},"c-68dp9m6f":{"id":"c-68dp9m6f","author":"claude","created_at":"2026-10-27T21:29:19.639863Z","text":"Implemented and merged into feature/lua-api-q1-parity. HostContext on pampa LuaOptions (crates/pampa/src/lua_options.rs), quarto.project + os.getenv shadow in crates/pampa/src/lua/host_context.rs, environment snapshot in quarto-core environment::snapshot_environment taken in StageContext::new and bridged to RenderContext for shortcodes, lmd::lua_options_for_document. CLI e2e: crates/quarto/tests/integration/lua_host_context_cli.rs. GH #939 repro now prints the directory and QUARTO_PROFILE=drafts under --profile drafts. Left open until the epic's PR merges."}}}
 {"id":"bd-iq08mmnh","title":"Title-block remainder from _quarto-rules: links + author/date/doi margins","description":"Port from _quarto-rules.scss L193–201 into title-block.scss: #title-block-header a { text-decoration: none } and .author/.date/.doi { margin-block-end: 0.2rem } (.date is emitted; author renders in the meta grid — adapt selectors to Q2's actual DOM, see kitchen-sink fixture). The >div flex/button rules (L203–223) are code-tools — excluded, see backlog strand. bd-btjkyylx TDD template. Inventory row 6c.","status":"closed","priority":3,"issue_type":"task","created_at":"2026-07-21T17:51:54.038559Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-21T20:01:56.476001Z","closed_at":"2026-07-21T20:01:56.476001Z","close_reason":"Ported #title-block-header a + .author/.date/.doi margins from _quarto-rules.scss:193-201 into title-block.scss. TDD + e2e verified; phase5 hash re-captured. Merged to feature/quarto-rules-scss-parity (de2b9774).","labels":["css","parity"],"dependencies":{"bd-4doe9lvt:parent-child":{"depends_on_id":"bd-4doe9lvt","type":"parent-child","created_at":"2026-07-21T17:51:54.038559Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-iq0hp","title":"Multi-engine preview: browser E2E + composing test engines","description":"A browser end-to-end test of MULTI-engine q2 preview was not possible during bd-5yff4: the preview server uses the default engine registry (real engines), the test-only FixtureEngine isn't registered there, and knitr/jupyter don't compose cleanly (knitr claims python cells). Single-engine preview is unchanged + unit-tested; the capture-sequence transport round-trips via quarto-preview integration tests. Options: (a) wire a flag-gated fixture engine into the preview server for tests, (b) build a real second engine, (c) hand-construct a two-capture sidecar and drive a browser session. Plan: claude-notes/plans/2026-05-27-multi-engine-execution.md","status":"open","priority":2,"issue_type":"task","created_at":"2026-05-27T16:24:40.064471Z","created_by":"cscheid","updated_at":"2026-06-22T19:39:53.089395Z","dependencies":{"bd-5yff4:discovered-from":{"depends_on_id":"bd-5yff4","type":"discovered-from","created_at":"2026-05-27T16:24:40.064471Z","created_by":"cscheid"}},"comments":{"c-c42zz8s8":{"id":"c-c42zz8s8","author":"Gordon Woodhull","created_at":"2026-06-22T19:39:53.089395Z","text":"TS-engine-extensions epic (rebase) resolves the *blocker* behind this strand. The blocker is 'knitr/jupyter don't compose cleanly (knitr claims python cells, jupyter sees no cell).' The epic introduces per-language ownership resolution: engines return kind-tagged claims (Primary/Interop/Fallback), and a tiered resolver assigns each language to one owner; each engine's leave-alone set (handled_languages) is the projection 'everything I don't own.' For 'engine: [knitr, jupyter]' on {r}+{python}, the explicit-Fallback tier routes r->knitr and python->jupyter, and knitr CEDES python via its existing knit_engines passthrough -> jupyter actually runs it. So knitr/jupyter (and any pair of real engines) compose cleanly. The epic also delivers composable engines (the echo engine in Plan 1c, then Julia in Plan 4). The browser E2E still needs writing, but the composition blocker this strand cites is removed. Model: claude-notes/designs/engine-resolution.md (see worked-cases table + §5 enforcement). Tracked as part of the TS-engine rebase."}}}
 {"id":"bd-iqd6niat","title":"lmd: inline cells reject numbers; meta.version (YAML number) needs tostring()","description":"Found while smoke-testing lmd docs examples on main (q2 0.33.0). With 'version: 12.1' in front matter, `{lmd} meta.version` in a paragraph fails: 'an inline lmd cell must evaluate to an Inline, a list of Inlines, a string, or nil; this one evaluated to a number'. The author has to write tostring(meta.version). Numbers (and booleans?) are the one metadata leaf type the inline contract does not accept, and {{< meta version >}} with a YAML number has its own oddity (expanded to nothing in a heading, noted in the exploration plan under spike 5). Proposal: accept a Lua number in inline context by formatting it with Lua's tostring (pampa's inline coercion layer, so pandoc.Strong(3) agrees). Decide whether booleans should format too or stay an error. The docs page will tell readers to use tostring() until this is decided.","status":"in_progress","priority":3,"issue_type":"feature","created_at":"2026-10-15T19:06:20.452839Z","created_by":"claude","updated_at":"2026-10-16T15:49:29.161943Z","external_ref":"https://github.com/quarto-dev/q2/pull/852","labels":["lmd","lua"],"dependencies":{"bd-lmd-docs-41adv2nx:discovered-from":{"depends_on_id":"bd-lmd-docs-41adv2nx","type":"discovered-from","created_at":"2026-10-15T19:06:20.452839Z","created_by":"claude"},"bd-lmd-exploration-jqv2nbl4:related":{"depends_on_id":"bd-lmd-exploration-jqv2nbl4","type":"related","created_at":"2026-10-15T19:06:20.452839Z","created_by":"claude"}},"comments":{"c-ird59ooa":{"id":"c-ird59ooa","author":"Carlos Scheidegger","created_at":"2026-10-16T13:42:00.088316Z","text":"2026-10-16: investigated on topic branch braid/bd-iqd6niat-lmd-inline-numbers (main checkout). Plan skeleton: claude-notes/plans/2026-10-16-lmd-inline-numbers.md. Verdict: ready to design. Key finding: the 'so pandoc.Strong(3) agrees' framing is backwards: pandoc 3.11 rejects pandoc.Strong(3) / pandoc.Inlines(3) / pandoc.Strong(true) (only pandoc.Str(3) coerces), and pandoc never exposes YAML numbers as Lua numbers (version: 12.1 is MetaInlines [Str \"12.1\"]). The need for tostring() is the documented cost of divergence D-num (2026-07-20 review). Options: (a) accept numbers in pampa's fuzzy inline layer (new divergence, 68 call sites), (b) accept them only in the lmd inline-cell contract (recommended), (c) keep the error and add a tostring() hint. Booleans, float formatting, block cells, docs sequencing are open questions. Repro confirmed at main e308835bd: 6 errors. Sibling bug filed: bd-32nrymty ({{< meta version >}} with a YAML real renders an empty Str)."},"c-y8ckjtr3":{"id":"c-y8ckjtr3","author":"Carlos Scheidegger","created_at":"2026-10-16T14:03:55.565865Z","text":"2026-10-16 decisions (Carlos): option (b), lmd inline-cell boundary only; booleans render as true/false; Lua tostring formatting via lua_tolstring; block cells keep rejecting numbers; docs fixed later (reminder left on bd-lmd-docs-41adv2nx). Option (a) filed as bd-q4gnd5os. Implemented on braid/bd-iqd6niat-lmd-inline-numbers: scalar_text + arms in classify_inline_value (crates/pampa/src/lua/lmd.rs), message + hint updated, tests in lmd.rs and lmd_execute.rs. Repro renders with zero errors (repro-output-after.txt)."}}}
 {"id":"bd-iqk0xash","title":"Report automerge-repo 2.6 JS-to-JS sync throughput drop upstream","description":"Measured during bd-ppu6xmvv: syncing 64 x 64 KB documents between two automerge-repo JS peers (BrowserWebSocketClientAdapter client, WebSocketServerAdapter + Repo server) went from ~2 s on 2.5.6 to ~13 s on 2.6.0-alpha.5 on an idle M-series Mac; each 2.6 side slows delivery on its own (2.6 hub + 2.5.6 client ~9 s, 2.5.6 hub + 2.6 client 7-26 s growing with docs the hub pushes to a fresh peer). Not reproducible against the Rust hub (samod): ~5 s on both client versions. saveDebounceRate is not the cause. Candidate area: the DocSynchronizer / CollectionSynchronizer rewrite (semaphores for sync-state loads and share policy). Build a minimal repro outside q2 and file it against automerge/automerge-repo; re-measure when a newer 2.6 alpha lands.","status":"open","priority":3,"issue_type":"task","created_at":"2026-09-17T18:31:49.955925Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T18:31:49.955925Z","labels":["upstream"],"dependencies":{"bd-ppu6xmvv:discovered-from":{"depends_on_id":"bd-ppu6xmvv","type":"discovered-from","created_at":"2026-09-17T18:31:49.955925Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-ir8n","title":"L9 follow-up: inline-code-style syntax-highlight class maps in full feeds","description":"Port Q1's inline-code-style transform: maps highlight classes (token comment, etc.) to inline style=\"color: ...\" so feeds render with colors when the subscriber's stylesheet doesn't include Quarto's CSS. v1 leaves highlight classes verbatim; readers without the CSS render code blocks in default mono. Reader extension lives in feed/reader_ext.rs; gated behind an RssReaderOptions flag.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-05-08T17:33:16.610890Z","created_by":"cscheid","updated_at":"2026-05-08T17:33:16.610890Z","dependencies":{"bd-o90m:discovered-from":{"depends_on_id":"bd-o90m","type":"discovered-from","created_at":"2026-05-08T17:33:16.610890Z","created_by":"cscheid"}}}
+{"id":"bd-irkfpip1","title":"Lua API: env shortcode reads the per-render environment snapshot","description":"Follow-up to the GH #939 work (PR #949): EnvShortcodeHandler still merges QUARTO_PROFILE -> std::env::var -> _environment file values at every lookup, while Lua filters and shortcode scripts read os.getenv from the StageContext::env_snapshot taken once per document. Move the env shortcode onto the same snapshot so the two agree by construction and the real environment is read in one place, on the orchestrator thread. Behavior is unchanged (same three layers); without a snapshot (RenderContext built directly) fall back to the real environment. Stacked PR on feature/lua-api-q1-parity.\n\nPlan: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md (follow-ups)","status":"in_progress","priority":3,"issue_type":"task","created_at":"2026-10-27T21:58:00.132710Z","created_by":"claude","updated_at":"2026-10-28T00:37:31.352026Z","external_ref":"https://github.com/quarto-dev/q2/pull/950","labels":["lua","parity"],"dependencies":{"bd-pqj37chw:discovered-from":{"depends_on_id":"bd-pqj37chw","type":"discovered-from","created_at":"2026-10-27T21:58:00.132710Z","created_by":"claude"}},"comments":{"c-ket6jtqp":{"id":"c-ket6jtqp","author":"claude","created_at":"2026-10-28T00:37:31.352026Z","text":"PR #950 merged into main. After #949 landed the PR retargeted to main, but the main-only workflows (test suite, Clippy, WASM, hub-client E2E) had never run against it, so main was merged into the branch (b3ad0c3b3) to trigger them; all checks then passed."},"c-yqzqm5hp":{"id":"c-yqzqm5hp","author":"claude","created_at":"2026-10-27T22:08:28.204171Z","text":"Implemented on braid/bd-irkfpip1-lua-api-env-shortcode (e4b0c8f30). Stacked PR: https://github.com/quarto-dev/q2/pull/950 (base feature/lua-api-q1-parity). EnvShortcodeHandler holds the StageContext env snapshot; pipeline builders take Option<Arc<EnvSnapshot>> instead of project_env + quarto_profile; RenderContext.env_snapshot bridge removed. clippy clean; nextest quarto-core + quarto: 7254 passed."}}}
 {"id":"bd-is4q72tt","title":"Perf: 1.1 MB api/index.qmd → 1.1 GB peak RSS; serial Connect-docs render peaks at 2.5 GB, parallel at 3.6 GB","description":"Measured with /usr/bin/time -l (release): q2 render api/index.qmd alone = 1.6 s wall, 1.09–1.17 GB max RSS; a small page = 158 MB; the full 352-doc serial render peaks at 2.0–2.5 GB, 16-worker parallel at 3.0–3.6 GB (peak memory footprint 2.9 GB). ~1000× the source size for one document. Not yet attributed (this profile was CPU-only); candidates: Pandoc AST + SourceInfo per node, JSON AST intermediates, grass AST clones (Vec<AstStmt>::clone was 3.5% of self time — re-measure after the sass key fix lands), per-document project artifacts held until end of render (bd-w5qyuzeg). Next step: heap profile (e.g. samply --heap / DHAT / instruments) on the single api/index.qmd render. Research note: claude-notes/research/2026-09-13-connect-docs-render-profile.md","status":"open","priority":2,"issue_type":"task","created_at":"2026-09-13T21:17:11.818026Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-13T21:17:11.818026Z","labels":["perf"],"dependencies":{"bd-fq44dlnm:discovered-from":{"depends_on_id":"bd-fq44dlnm","type":"discovered-from","created_at":"2026-09-13T21:17:11.818026Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-iscjjhx8","title":"Fix TS suite: sync-client Repo mocks missing handles after 5d31edf6b","description":"Commit 5d31edf6b (connection modal) added trackPeers' repo.handles sweep + 2s watch interval to client.ts. Six Repo mocks across applyEditorOperations.test.ts, client.test.ts, client.connect-options.test.ts lacked a handles property, so Object.values(repo.handles) threw TypeError in all 29 tests using those mocks; two connect-options tests also asserted exact timer counts broken by the new interval. Fix: add handles:{} to the six mocks and update the two timer-count assertions. Full suite: 137/137 green.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-08-27T21:59:49.472994Z","created_by":"shikokuchuo","updated_at":"2026-08-27T22:01:40.761379Z","closed_at":"2026-08-27T22:01:40.761379Z","close_reason":"Fixed in 80a4aac03: added handles:{} to the six Repo mocks and updated the two timer-count assertions. Suite green 137/137.","comments":{"c-jjooam5a":{"id":"c-jjooam5a","author":"shikokuchuo","created_at":"2026-08-27T22:00:16.203193Z","text":"Fix verified locally: npm test -w ts-packages/quarto-sync-client -- --retry=2 → 137/137 green (was 29 failures). tsc --noEmit clean. Awaiting commit approval."}}}
 {"id":"bd-itj2mjkr","title":"Engine intermediate FileId slot desync: EngineExecutionStage remaps executed AST onto _quarto.yml's dense slot in project renders","description":"Verified structurally (bd-nv4p0eb1 audit). Main pipeline order is ParseDocument -> MetadataMerge -> ... -> IncludeExpansion -> ... -> EngineExecution (pipeline.rs:279-323). MetadataMergeStage registers _quarto.yml/_metadata.yml into doc.ast_context.source_context via add_file_with_id (grows SourceContext.files, does NOT grow ast_context.filenames). engine_execution.rs:498 then computes the intermediate's slot as FileId(filenames.len()) while add_file actually assigns FileId(files.len()) — larger by the number of registered config files — and :528-530 remaps every executed-AST id by id.0 + new_slot.0. Result: in any project render with an executable engine, engine-produced blocks carry a FileId that resolves (via get_file's dense indexing) to _quarto.yml or an include, not the engine intermediate. The lock-step comment at :333-334 is false after MetadataMerge. Downstream: trace attribution wrong; diagnostics anchored in engine blocks render config-file text; writers/json.rs:1825 positional filenames[idx]/files[idx] pairing compounds it. Needs a repro fixture (project + _quarto.yml + executable cell) asserting engine-block FileIds resolve to the intermediate slot. Also add debug_assert for the lock-step invariant, and consider retiring the parallel filenames table (see research doc §5). Assessment: claude-notes/research/2026-08-09-fileid-span-integrity-audit.md (finding P1).","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-08-09T17:10:54.251287Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-10T00:10:29.711470Z","closed_at":"2026-08-10T00:10:29.711470Z","close_reason":"Merged to main via PR #482 (quarto-dev/q2, merge commit 8518ac79); CI green on all suites (macOS/Ubuntu Rust+TS, WASM, Hub E2E).","dependencies":{"bd-nv4p0eb1:discovered-from":{"depends_on_id":"bd-nv4p0eb1","type":"discovered-from","created_at":"2026-08-09T17:10:54.251287Z","created_by":"Carlos Scheidegger"}},"comments":{"c-pkvn6lb3":{"id":"c-pkvn6lb3","author":"Carlos Scheidegger","created_at":"2026-08-09T17:51:58.428903Z","text":"Committed as 9d1c2581 on braid/bd-itj2mjkr-engine-slot-desync (docs commit 8345e120 precedes it). xtask verify --skip-hub-build green. Strand stays in_progress until PR review/merge."},"c-r14pctoo":{"id":"c-r14pctoo","author":"Carlos Scheidegger","created_at":"2026-08-09T17:45:13.134959Z","text":"TDD red confirmed: new stage-level test engine_blocks_attribute_to_intermediate_after_config_registration (engine_execution.rs) failed on baseline with the exact predicted symptom — engine-produced block attributed to /project/_quarto.yml. Fix on branch braid/bd-itj2mjkr-engine-slot-desync: new_slot now comes from the SourceContext add_file*/add_file_with_info return value (never filenames.len()); remap is conditional (FileId(0)->new_slot, other ids pass through untouched instead of additive shifting); both false lock-step comments corrected. Green: new test + 152 engine tests + full workspace (11147 passed). E2E: cargo run --bin q2 -- render on a project fixture (_quarto.yml + jupyter cell) via real jupyter — rendered clean, 'engine-produced output' present in index.html; attribution assertion carried by the stage-level test, which drives the real stage with the production MetadataMerge registration state. xtask verify (skip-hub-build) running; full verify before push."}}}
@@ -1057,7 +1065,7 @@
 {"id":"bd-ix7psky0","title":"HTML: make the dark/light mode toggle an accessible switch (Quarto 1 #14739 parity)","description":"quarto-cli fa48b15ea (#14739, in 1.11.6) turns the color-scheme toggle into an accessible switch (role/aria-checked, label, keyboard) across its JS, templates and _bootstrap-rules.scss, plus no toggle when no dark theme. q2's toggle (resources/js/quarto-nav, _bootstrap-rules.scss .quarto-color-scheme-toggle) predates it. Related: bd-ld-toggle-into-tools-hpae7m9r.","status":"in_progress","priority":3,"issue_type":"feature","created_at":"2026-10-21T17:28:02.765948Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-21T18:12:44.026469Z","labels":["a11y","html"],"dependencies":{"bd-jsfg1494:discovered-from":{"depends_on_id":"bd-jsfg1494","type":"discovered-from","created_at":"2026-10-21T17:28:02.765948Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-ixdktocp","title":"Docs: revealjs feature page + per-feature example projects + embed-placeholder convention","description":"Document the revealjs authoring features implemented in the revealjs epic (Phase 1-2): creating slides (level-2 slides, level-1 section dividers, HR breaks, auto title slide, slide-level), fragments, incremental lists, columns, speaker notes, asides, per-slide footnotes. \n\nDeliverables:\n1. ~8 minimal self-contained Quarto 2 example projects under examples/presentations/NN-feature/ (each a type: default project: _quarto.yml + one format: revealjs .qmd + README.md + .gitignore), one per feature.\n2. A docs page docs/presentations/revealjs/index.qmd written in Q1-quarto-web style (prose + code block per feature) using the reader-expectations-prose (Gopen) writing skill, each feature section ending with an embed PLACEHOLDER referencing its example project.\n3. A generic embed-placeholder convention: a fenced div with class .q2-website-example-iframe carrying the example path (generic across ALL Q2 example categories, not revealjs-specific, per user). Future work (separate) swaps it for a live <iframe class=q2-website-example-iframe> via a Lua filter on docs/ or a q2 format: html feature.\n4. Navigation: new Presentations navbar entry + sidebar section in docs/_quarto.yml.\n\nVerification: render each example via cargo run --bin q2 -- render examples/presentations/<dir>; render the docs site via cargo run --bin q2 -- render docs/ and inspect the new page. Plan: claude-notes/plans/2026-06-09-revealjs-docs-and-examples.md. The iframe embed MECHANISM itself is explicitly out of scope (tracked separately).","status":"closed","priority":2,"issue_type":"docs","created_at":"2026-06-09T14:32:40.850067Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-09T15:21:50.236632Z","closed_at":"2026-06-09T15:21:50.236632Z","close_reason":"Docs page (docs/presentations/revealjs/index.qmd) + 8 per-feature example projects + generic .q2-website-example-iframe placeholder convention + navbar/sidebar wiring all landed and verified (q2 render docs/ 150/150). The live-iframe embed mechanism is tracked separately as a follow-up.","labels":["docs","revealjs"],"dependencies":{"bd-bea550b0:discovered-from":{"depends_on_id":"bd-bea550b0","type":"discovered-from","created_at":"2026-06-09T14:32:40.850067Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-ixnp4uqj","title":"Lua parity H2: roll Q-11-3/5 error contract across remaining marshaling error sites","description":"Follow-up to bd-9p2686pc (contract defined + core adoption done). Sweep the remaining ~85 bare Error::runtime sites in crates/pampa/src/lua/ (table-part setters 'cannot set field on X', eager property validators like quotetype/mathtype/citation mode, ListAttributes/Meta peekers) onto the granular contract: Q-11-3 invalid argument via type_mismatch_error, Q-11-5 invalid property assignment. Keep hslua's '<expected> expected, got <type>' body shape. Contract + helpers: crates/pampa/src/lua/types.rs (type_mismatch_error); plan claude-notes/plans/2026-07-14-bd-9p2686pc-lua-error-contract.md Phase 3; epic plan Phase 3.3.","status":"closed","priority":3,"issue_type":"task","created_at":"2026-07-14T15:47:40.294655Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-14T18:22:50.052515Z","closed_at":"2026-07-14T18:22:50.052515Z","close_reason":"Rollout complete (commit f79fbf22 on feature/bd-grkrb9nj-lua-api-pandoc-parity): all bare Error::runtime sites in the marshaling layer (types.rs ~30, constructors.rs ~50, list.rs 3) moved onto Q-11-3 (value-conversion/enum-value, hslua expected/got shape) or Q-11-5 (unknown/read-only field, wrong variant, proxy keys); new helpers type_mismatch_error_named/invalid_value_error/read_only_field_error/unknown_field_error in types.rs. Out of scope by design: stdlib/system shims (readwrite/system/io_wasm/quarto_doc/mediabag/utils/...) and walk.rs internal invariant. TDD red->green; conformance ratchet unchanged 184/19; full cargo xtask verify green; e2e via q2 render (both codes observed).","dependencies":{"bd-9p2686pc:discovered-from":{"depends_on_id":"bd-9p2686pc","type":"discovered-from","created_at":"2026-07-14T15:47:40.294655Z","created_by":"Carlos Scheidegger"},"bd-grkrb9nj:parent-child":{"depends_on_id":"bd-grkrb9nj","type":"parent-child","created_at":"2026-07-14T15:47:47.753419Z","created_by":"Carlos Scheidegger"}},"comments":{"c-ehb679cf":{"id":"c-ehb679cf","author":"Carlos Scheidegger","created_at":"2026-07-14T18:14:56.741984Z","text":"Sweep implemented: all bare Error::runtime sites in the marshaling layer (types.rs, constructors.rs, list.rs) now carry Q-11-3 (value-conversion/enum-value errors, wherever invoked) or Q-11-5 (setter-structural refusals: unknown/read-only field, wrong variant, proxy keys). New helpers in types.rs: type_mismatch_error_named, invalid_value_error, read_only_field_error, unknown_field_error. Out of scope (deliberate): stdlib/system shims (readwrite/system/io_wasm/quarto_doc/mediabag/utils/...) — different error families, not the marshaling contract; walk.rs:686 internal invariant. TDD red->green; pampa 4124 pass; workspace 10076 pass; e2e Q-11-3+Q-11-5 observed via q2 render. Full xtask verify running."},"c-wnk34r8b":{"id":"c-wnk34r8b","author":"Carlos Scheidegger","created_at":"2026-07-14T17:53:50.030395Z","text":"Resuming after session crash: no prior work existed (status was in_progress but no commits/comments). Starting the sweep now."}}}
-{"id":"bd-ixz93hjp","title":"Quarto API: settle the 16 Q1-divergence conflicts blocking JSR publish","description":"Gate for publishing @quarto/types + @quarto/api to jsr.io. JSR is immutable (published versions cannot be deleted) and 'deno bundle' freezes the resolved version into each extension author's bundle, so a divergence from Q1 is wrong PERMANENTLY while an absence is merely unfinished.\n\nRelease criterion (not completeness): if Q2 is a strict subtree of Q1, an incomplete SDK is safe to publish -- missing surface is additive later. It is the CONFLICTS that must be settled.\n\nEvidence base: claude-notes/research/2026-08-22-quarto-api-q1-vs-q2-status.md (951 lines, on branch feature/jsr-publish). Section 2 carries the authoritative gate table. Findings were produced partly by DIFFERENTIAL EXECUTION of Q1 vs Q2, not only by reading -- which is how 23 category-D behavioural divergences were found in namespaces a reading-based pass had assessed as clean verbatim ports.\n\nCategories: A absent in Q2 (safe, additive) / B identical (safe) / C same name different signature (blocker) / D same signature different behaviour (WORST -- invisible to any type-level guard) / E Q2-only, no reference to validate against.\n\n16 conflicts survive under the reduced published surface (only 1 of 17 retires). Withholding exports removes NAMES, not CONDUCT: behaviour reaches authors through the injected QuartoAPI, typed by @quarto/types, which is published in full regardless of what @quarto/api exports.\n\nTwo things explicitly NOT in scope (do not spend effort here): the 15 unimplemented jupyter capabilities, the dropped Q1 registry, and everything in the research doc's section 8 -- all category A, safe. Also: two divergences are Q1 BUGS that Q2 fixed (shortcode params containing '=', md-indent garbling) -- keep Q2's behaviour, do not 'settle' those toward Q1.\n\nBlocks: claude-notes/plans/2026-08-22-jsr-publish-quarto-api-types.md Phase 4 (first publish, irreversible). Phases 1-3 and 5-6 of that plan are packaging work and are independent of this epic.","status":"open","priority":1,"issue_type":"epic","created_at":"2026-08-23T02:53:53.872081Z","created_by":"Gordon Woodhull","updated_at":"2026-08-23T02:53:53.872081Z"}
+{"id":"bd-ixz93hjp","title":"Quarto API: settle the 16 Q1-divergence conflicts blocking npm publish","description":"Gate for publishing @quarto/types + @quarto/api to jsr.io. JSR is immutable (published versions cannot be deleted) and 'deno bundle' freezes the resolved version into each extension author's bundle, so a divergence from Q1 is wrong PERMANENTLY while an absence is merely unfinished.\n\nRelease criterion (not completeness): if Q2 is a strict subtree of Q1, an incomplete SDK is safe to publish -- missing surface is additive later. It is the CONFLICTS that must be settled.\n\nEvidence base: claude-notes/research/2026-08-22-quarto-api-q1-vs-q2-status.md (951 lines, on branch feature/jsr-publish). Section 2 carries the authoritative gate table. Findings were produced partly by DIFFERENTIAL EXECUTION of Q1 vs Q2, not only by reading -- which is how 23 category-D behavioural divergences were found in namespaces a reading-based pass had assessed as clean verbatim ports.\n\nCategories: A absent in Q2 (safe, additive) / B identical (safe) / C same name different signature (blocker) / D same signature different behaviour (WORST -- invisible to any type-level guard) / E Q2-only, no reference to validate against.\n\n16 conflicts survive under the reduced published surface (only 1 of 17 retires). Withholding exports removes NAMES, not CONDUCT: behaviour reaches authors through the injected QuartoAPI, typed by @quarto/types, which is published in full regardless of what @quarto/api exports.\n\nTwo things explicitly NOT in scope (do not spend effort here): the 15 unimplemented jupyter capabilities, the dropped Q1 registry, and everything in the research doc's section 8 -- all category A, safe. Also: two divergences are Q1 BUGS that Q2 fixed (shortcode params containing '=', md-indent garbling) -- keep Q2's behaviour, do not 'settle' those toward Q1.\n\nBlocks: claude-notes/plans/2026-08-22-jsr-publish-quarto-api-types.md Phase 4 (first publish, irreversible). Phases 1-3 and 5-6 of that plan are packaging work and are independent of this epic.","status":"open","priority":1,"issue_type":"epic","created_at":"2026-08-23T02:53:53.872081Z","created_by":"Gordon Woodhull","updated_at":"2026-10-28T20:38:34.877112Z","comments":{"c-ilmvnexq":{"id":"c-ilmvnexq","author":"Gordon Woodhull","created_at":"2026-10-28T20:38:34.877112Z","text":"Naturally we mean npm, not JSR: the packages (@quarto/types, @quarto/api) publish to npm. The immutability and version-freezing concerns in the description still apply (npm versions cannot be republished and extension bundles freeze the resolved version), and references to 'JSR' elsewhere on this strand and in the linked plan/research docs should be read as npm. — Gordon (via Claude)"}}}
 {"id":"bd-iy2zbos7","title":"Windows: Jupyter kernel still orphaned after a nextest run (after the KernelScope fix)","description":"Found 2026-09-30 (Windows 11, during the bd-1klbq2zd layer 3 six-crate nextest run: -p quarto -p quarto-system-runtime -p quarto-core -p quarto-hub -p quarto-preview -p quarto-test --build-jobs 6 --test-threads 8, at f450fc93 on bugfix/bd-1klbq2zd-dunce-seam).\n\nObserved: one orphaned kernel survived the run: \"C:\\Users\\chris\\AppData\\Local\\Python\\pythoncore-3.14-64\\python.exe -Xfrozen_modules=off -m ipykernel_launcher -f C:\\Users\\chris\\AppData\\Roaming\\jupyter\\runtime\\kernel-941c5461-....json\". It was created 2 min into the run, its parent PID had already exited, and it stayed alive after nextest finished. It held an inherited stdout/stderr pipe handle, so the PowerShell shell running `cargo nextest ... *> log` never returned. Killing that PID made the shell exit at once (confirmed). Which test spawned it was not recorded; its cwd was lost when it was killed.\n\nContext: bd-hxhnnlzs (PR #491, a217ab5a) fixed kernel leaks with a refcounted KernelScope (shutdown_request, then kill, reap, and remove the connection file) and proved zero orphans on macOS. This is a Windows recurrence, or a path that fix does not cover.\n\nHypothesis (unverified): on Windows the spawned \"python\" can be a launcher stub (WindowsApps\\python.exe or py.exe) that starts the real interpreter as a child. Killing and reaping the stub PID does not kill the real kernel, so the backstop kill misses it whenever the graceful shutdown_request does not complete. The same session saw a live \"WindowsApps\\python.exe -\" (PID 49680) parenting the real python (34820), which shows the stub shape exists on this machine. Check how daemon.rs spawns the kernel (argv[0] from the kernelspec, resolved how?) and whether the kill targets a process tree or a job object.\n\nLiveness check: on Windows, run `cargo nextest run -p quarto-core -E 'binary(integration) & test(jupyter_kernel_cleanup)'` (and the engine_error_policy / engine_output_parity jupyter tests), then look for leftover kernels: `Get-CimInstance Win32_Process | Where-Object CommandLine -match 'ipykernel_launcher' | Select-Object ProcessId,ParentProcessId,CreationDate,CommandLine`. Any kernel created during the run that is still alive afterwards reproduces this. Record its cwd before killing it (e.g. with handle/Process Explorer), to attribute the test.\n","status":"open","priority":2,"issue_type":"bug","created_at":"2026-09-30T12:52:31.636677Z","created_by":"cderv","updated_at":"2026-09-30T14:11:09.482148Z","labels":["windows"],"dependencies":{"bd-1klbq2zd:discovered-from":{"depends_on_id":"bd-1klbq2zd","type":"discovered-from","created_at":"2026-09-30T12:52:31.636677Z","created_by":"cderv"},"bd-hxhnnlzs:related":{"depends_on_id":"bd-hxhnnlzs","type":"related","created_at":"2026-09-30T12:52:43.979788Z","created_by":"cderv"},"bd-t0wpbq59:parent-child":{"depends_on_id":"bd-t0wpbq59","type":"parent-child","created_at":"2026-09-30T12:52:50.021323Z","created_by":"cderv"}},"comments":{"c-cthfsfbe":{"id":"c-cthfsfbe","author":"cderv","created_at":"2026-09-30T13:57:50.463265Z","text":"2026-09-30: reproduced twice more with cargo nextest run -p quarto-test -p quarto --no-fail-fast --build-jobs 6 --test-threads 8 *> log, once on the rebased stack and once on main 9f0f6c66a (so it is not caused by the stack). Both times nextest printed Summary but the PowerShell statement never returned (no EXIT line), until the terminal was killed. Leftover after the main run: ipykernel_launcher PID 18868, created 15:33:32 during the run, parent 52428 already gone. smoke_all runs jupyter documents, so it is a likely spawner there. Workaround to try (untested): cmd /c 'cargo nextest ... > log 2>&1' so the orphan inherits a file handle instead of PowerShell's pipe."},"c-yl4dgxsd":{"id":"c-yl4dgxsd","author":"cderv","created_at":"2026-09-30T14:11:09.482148Z","text":"2026-09-30: the cmd /c 'cargo nextest ... > log 2>&1' workaround works: the shell returned on its own after the stack-tip run (-p quarto-test -p quarto, stack tip ca4bda175). But the following EXIT line did not reach the log (not investigated; the orphan may hold the file open). The run still left an orphan: ipykernel_launcher PID 52036, created 15:59:57 during the run, parent PID 23288. Left alive so its cwd can be recorded to attribute the test."}}}
 {"id":"bd-izay61p8","title":"Phase 1.4: RevealjsFormatOptions + metadata plumbing","description":"Add revealjs: RevealjsFormatOptions to FormatOptions (quarto-core/src/render.rs). Tier-1 fields: theme, transition, transition_speed, slide_number, controls, progress, center, hash, incremental, width, height, slide_level, logo, footer. Surface format.revealjs.* via resolve_format_config(); map to Reveal.initialize keys. Test YAML->initialize JSON round-trip.","status":"closed","priority":1,"issue_type":"task","created_at":"2026-06-08T20:35:34.597914Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-08T21:47:34.229480Z","closed_at":"2026-06-08T21:47:34.229480Z","close_reason":"reveal_config_json() maps flattened format.revealjs.* -> Reveal.initialize keys (read via as_plain_text). Typed FormatOptions struct deferred (see plan 1.4 deviation note). Commit c43c2406.","labels":["revealjs"],"dependencies":{"bd-2m4wanyd:parent-child":{"depends_on_id":"bd-2m4wanyd","type":"parent-child","created_at":"2026-06-08T20:35:34.597914Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-izfv","title":"Phase 9 follow-up: thread user_grammars through RenderToHtmlRenderer","description":"Today render_page_in_project's project-rendering branch drops user_grammars on the floor — the RenderToHtmlRenderer constructs its own per-page RenderContext and there's no path to attach user_grammars there. Single-file projects still wire user_grammars correctly via render_single_doc_to_response.\n\nThreading user grammars through the renderer is straightforward: add a field to RenderToHtmlRenderer holding an Arc<Mutex<Option<JsUserGrammars>>> (or similar) and have its render() method install the provider on the per-page RenderContext.\n\nFiled as discovered-from-bd-ayj6 and tagged P3 since most user-grammar use is for code-block highlighting which still works on the active page's first render — the gap is just for project renders where the hub-client passes a grammars handle but it gets ignored.","status":"closed","priority":3,"issue_type":"task","created_at":"2026-04-29T00:31:56.879941Z","created_by":"cscheid","updated_at":"2026-05-12T13:41:25.597279Z","closed_at":"2026-05-12T13:41:25.597259Z","close_reason":"Threaded user_grammars through RenderToHtmlRenderer (Rc<RefCell<…>>); re-enabled e2e fixture by dropping SKIP_WASM_UNSUPPORTED entry on chore/e2e-ci, and fixed two test-harness gaps (binary fixture forwarding + getPreviewHtml userGrammars context) so the hub-client e2e suite can actually verify the fixture. Rust unit test crates/quarto-core/tests/render_to_html_user_grammars.rs passes; e2e fixture highlighting/03-user-grammar-toml.qmd passes when run in isolation. See claude-notes/plans/2026-05-10-thread-user-grammars-renderer.md for the verification artifact.","dependencies":{"bd-0tr6:parent-child":{"depends_on_id":"bd-0tr6","type":"parent-child","created_at":"2026-04-29T00:31:56.879941Z","created_by":"cscheid"},"bd-ayj6:discovered-from":{"depends_on_id":"bd-ayj6","type":"discovered-from","created_at":"2026-04-29T00:31:56.879941Z","created_by":"cscheid"}}}
@@ -1140,6 +1148,7 @@
 {"id":"bd-k4g6uc2n","title":"FTS Phase 1: wire indexing into sync handlers + FileSidebar search UI","description":"Hook indexing into setSyncHandlers (App.tsx:435-465): bulk index on project open, addOrUpdate/remove/clear on change. Add search box + result list to FileSidebar; select -> onSelectFile. Snippet/highlight rendering from raw-text offsets. Plan items 1.4-1.6. Plan: claude-notes/plans/2026-06-23-hub-client-full-text-search.md","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-06-23T18:23:12.003648Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-23T18:36:29.657858Z","closed_at":"2026-06-23T18:36:29.657858Z","close_reason":"Indexing hook + FileSidebar search UI + snippet highlighting landed; 19 new tests, full suite (651) + typecheck + production build green.","dependencies":{"bd-mdet0yoy:parent-child":{"depends_on_id":"bd-mdet0yoy","type":"parent-child","created_at":"2026-06-23T18:23:12.003648Z","created_by":"Carlos Scheidegger"},"bd-yb7nro8r:blocks":{"depends_on_id":"bd-yb7nro8r","type":"blocks","created_at":"2026-06-23T18:23:40.704232Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-k5rxujiy","title":"q2 preview: logo image (logo: logo.svg) 404s — asset walker misses meta-driven raw-HTML images","description":"AUDIT (2026-06-17, confirmed in browser). With footer/logo DOM now emitted (bd-n2w0sxgd), the logo <img class=slide-logo src=logo.svg> FAILS TO LOAD in q2 preview (naturalWidth 0). Network: GET /logo.svg -> 200 text/html (the SPA shell), i.e. logo.svg is NOT in the preview's served VFS asset set, so the catch-all returns index.html.\n\nROOT CAUSE (confirmed): the parent-side asset walker (ts-packages/preview-renderer/src/q2-preview/assetWalker.ts::collectImagePaths) only collects AST Image nodes (obj.t === 'Image'). The logo comes from metadata (logo: logo.svg), rendered by RevealFooterLogoTransform into rendered.reveal.logo as a RAW HTML <img> string, injected by RevealChrome via dangerouslySetInnerHTML. It is never an Image AST node, so:\n 1. buildAssetManifest never mints a blob URL for logo.svg, and\n 2. the raw <img> bypasses the <Image> React component that rewrites src against the manifest.\nHence src stays 'logo.svg' -> /logo.svg -> not in VFS -> 404-as-200-html.\n\nIn native render this works because logo.svg sits next to slides.html on disk; in VFS-mode preview nothing copies/serves it.\n\nNote: ResourceCollectorTransform is also a no-op in vfs_root mode (resource_collector.rs:96), by design — the asset walker is the producer in preview.\n\nFIX OPTIONS (needs a design decision):\n (A) Walker also collects meta.logo (+ other known meta image keys) and RevealChrome rewrites the injected <img> src via AssetManifestContext before injection. Targeted; couples RevealChrome to the manifest.\n (B) Generalize the walker to scan rendered.reveal.* / rendered.navigation.* HTML strings for src=/href= asset refs, mint blobs, and have the injecting slots rewrite. More general (covers footer images, navbar logos, etc.).\n (C) Represent logo as a real Image AST node somewhere the walker sees it (and render through <Image>). Biggest change.\n\nRecommend (B) as the general solution; (A) as a quick win. Plan: claude-notes/plans/2026-06-17-revealjs-preview-render-parity-audit.md (F5).","status":"open","priority":2,"issue_type":"bug","created_at":"2026-06-17T16:38:41.382596Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-17T17:06:14.418341Z","dependencies":{"bd-67yja58s:parent-child":{"depends_on_id":"bd-67yja58s","type":"parent-child","created_at":"2026-06-17T16:38:41.382596Z","created_by":"Carlos Scheidegger"},"bd-d0f9xoa6:related":{"depends_on_id":"bd-d0f9xoa6","type":"related","created_at":"2026-06-17T16:38:41.382596Z","created_by":"Carlos Scheidegger"},"bd-n2w0sxgd:related":{"depends_on_id":"bd-n2w0sxgd","type":"related","created_at":"2026-06-17T16:38:41.382596Z","created_by":"Carlos Scheidegger"},"bd-zaa4m4bd:discovered-from":{"depends_on_id":"bd-zaa4m4bd","type":"discovered-from","created_at":"2026-06-17T16:38:41.382596Z","created_by":"Carlos Scheidegger"}},"comments":{"c-0vto1r8b":{"id":"c-0vto1r8b","author":"Carlos Scheidegger","created_at":"2026-06-17T17:06:13.523312Z","text":"LAYER 1 DONE (commit bc5b27db, feature/revealjs-preview-render-parity). resolve_single_file_deps now expands document resources: into binary_files (reuses extract_resource_patterns + expand_patterns; globs supported), riding the existing single_file_assets -> with_binary_files VFS sync. Tests: single_file_deps_includes_declared_resources + _glob (8/8 green). E2E: real q2 preview, deck with 'logo: logo.svg' + 'resources: [logo.svg]' (logo NOT a md image) -> 'Discovered project files binary_count=1'; without resources: -> binary_count=0.\n\nREMAINING: Layer 2 (serve bytes to the raw <img>; the logo still 404s without it). Decision pending: (a) per-asset rewrite contained to preview-renderer (walker collects meta.logo/meta.resources -> blob; RevealChrome rewrites src/href via AssetManifestContext), OR defer until (b) the bd-msp0 service-worker path-server (touches hub-client + quarto-hub.com) is on the table. Leaving open for the Layer 2 decision."},"c-pg7y4hlp":{"id":"c-pg7y4hlp","author":"Carlos Scheidegger","created_at":"2026-06-17T16:49:22.570171Z","text":"FEASIBILITY (user proposal: let asset detection honor document-level 'resources:'). Does NOT already work, but it's feasible and well-aligned with existing code. TWO LAYERS:\n\nLAYER 1 — bytes into the VFS (the part the user means; small, clean). The single-file dep resolver crates/quarto-preview/src/config.rs::resolve_single_file_deps (bd-9cyza5vy) already runs the real parse+include stages and collects image deps via collect_referenced_asset_urls(blocks). Adding document-level resources: is a natural extension right there: call quarto_core::project_resources::extract_resource_patterns(&doc.ast.meta, [\"resources\"]) (already used in document_profile.rs:569 + resource_report.rs) then expand_patterns(canonical_root, deck_dir_anchor, patterns, ResourceOrigin::DocumentMetadata, ResourceScope::...) (already used in THIS file for project resources, config.rs:129-150) and fold matches into SingleFileDeps.binary_files. Reuses the exact publish-path machinery (globs included), so 'declare resources: once -> works for publish AND preview'. Not reveal-specific; helps any single-file doc referencing assets the AST walker can't see (raw HTML, shortcodes, css url(), logo:). Bonus: config.rs:122 notes resources: is already the interim trust boundary for what may upload to a sync server, so feeding it into the VFS sync is on-policy.\n\nLAYER 2 — actually SERVE those bytes (necessary; layer 1 alone is insufficient). The logo is a raw <img src=logo.svg> injected by RevealChrome; it fetches /logo.svg -> SPA catch-all (200 text/html), never going through <Image> (which swaps blob URLs). Options: (a) TS asset walker also mints blobs for meta.resources/logo: and RevealChrome rewrites raw src/href against the manifest (targeted, works now); (b) service-worker VFS-by-path server (bd-msp0 epic) serves /logo.svg from the VFS directly — general, future-proof, no rewriting; this is the 'alleviate a lot of future problems' win.\n\nRECOMMEND: do Layer 1 (resources:->VFS) regardless (cheap, publish-parity). Pair with (b) service worker for the general fix, or (a) for a quick interim. Awaiting user direction on scope."}}}
 {"id":"bd-k8ol","title":"Mode B: partial Pass-1 walk to skip non-target profile re-extraction","description":"Phase 8.2 originally planned partial Pass-1 in Mode B: only profile the user-named targets plus their dependency closure. Implementation discovered a chicken-and-egg with sidebar auto: expansion (the membership resolver consults the index, which doesn't yet exist for non-target pages until they've been profiled). v1 ships full Pass-1 in both modes; the cache makes it cheap on the warm path. Decoupling the auto: resolver from index existence lets us land the partial walk and amortize Pass-1 cost on truly cold runs of huge projects. See claude-notes/plans/2026-04-27-websites-phase-8.md §'Sub-phase 8.2 — Deviation from plan'.","status":"closed","priority":3,"issue_type":"feature","created_at":"2026-04-28T00:42:47.359008Z","created_by":"cscheid","updated_at":"2026-07-29T16:45:55.802+00:00","closed_at":"2026-07-29T16:45:55.802+00:00","dependencies":{"bd-0tr6:parent-child":{"depends_on_id":"bd-0tr6","type":"parent-child","created_at":"2026-04-28T00:42:47.359008Z","created_by":"cscheid"},"bd-fegm:discovered-from":{"depends_on_id":"bd-fegm","type":"discovered-from","created_at":"2026-04-28T00:42:47.359008Z","created_by":"cscheid"}}}
+{"id":"bd-k8ptcrd9","title":"Phase 0: main-thread render-path measurement harness + baselines","description":"Build the TS-side profiling driver, add perf gauges, record baseline numbers, and write characterization tests for the paths Phases 1-3 will touch. Plan: claude-notes/plans/2026-10-24-hub-client-main-thread-render-path.md (Phase 0).","status":"closed","priority":1,"issue_type":"task","created_at":"2026-10-29T18:49:15.953703Z","created_by":"shikokuchuo","updated_at":"2026-10-29T19:45:06.111739Z","closed_at":"2026-10-29T19:45:06.111739Z","close_reason":"Phase 0 done: harness, gauges, baselines, characterization tests. Commits d3b294042 + cf3e5290c on feature/bd-k8ptcrd9-render-path-phase0.","comments":{"c-8g5ztba6":{"id":"c-8g5ztba6","author":"shikokuchuo","created_at":"2026-10-29T19:41:32.877537Z","text":"Phase 0 complete (2026-10-29). Deliverables: (1) ~99KB/800KB fixture series + throwaway tsx/Node driver (real release WASM, fake-indexeddb for the SASS cache bridge, requestIdleCallback override, --cpu-prof). (2) Gauges: render.started/completed/wasm-call/discarded-stale, wasm-envelope.parse (segment+bytes), vfs.write (segment+bytes), intelligence.semantic-tokens(.tokenize), parse.qmd-to-ast, router.format-check(.*). perfCounters now exists as two copies (hub-client + preview-runtime) sharing one globalThis sink — drift-guarded by perfCounters.sync.test.ts. (3) Characterization: renderQueue.test.ts (serial, no-coalescing pin for Phase 1 to flip, reject-no-jam), 3 editor-edge classification pins (unknown scalar key → class neither; unterminated YAML resolves OK as default html; broken body still classifies), exact 13-key UPDATE_AST payload pin. (4) Baseline (1x/8x, warm cache): render 302/2735ms, discarded parse 79/708ms (18% of per-keystroke at 1x), tokens 44/386ms, envelope parse 14.4/132ms + 16.3/134.5MB, vfs 2 writes/keystroke; burst of 10 keystrokes → 10 renders (no coalescing). Full table in the plan file. Gates: unit 1743+82 green, wasm leg green except PRE-EXISTING smokeAll version-shortcode failure on main (filed bd-ucxaqjn2). build:all green."}}}
 {"id":"bd-k8y0","title":"[websites] Sidebar vertical border missing (Q1 parity) — needs doc-derived SCSS variables seam","description":"Quarto 1 renders a faint vertical line between a docked sidebar and main content; Quarto 2 does not. Verified visually in examples/websites/03-nested-sidebar (q1-site has it, _site does not). \n\nRoot cause: two missing pieces in Q2.\n\n1. The SCSS rule '.sidebar.sidebar-navigation:not(.rollup) { border-right: 1px solid $table-border-color !important; }' (gated by '@if $sidebar-border') exists in Q1 (quarto-cli/src/resources/projects/website/navigation/quarto-nav.scss:552-556) but was never ported to resources/scss/bootstrap/_bootstrap-rules.scss in Q2.\n\n2. Q2 has no mechanism to inject per-document SCSS variables before the framework defaults. CompileThemeCssStage only reads theme config from doc.ast.meta; it never threads website.sidebar info into the SCSS bundle. Q1's format-html-scss.ts:631-642 synthesizes '$sidebar-border: <bool>;' per-document, defaulting to (style == 'docked').\n\nCustomization: end users override via 'sidebar.border: true|false' in YAML (Q1). Color follows the active Bootstrap/Bootswatch theme via $table-border-color (no special handling needed — themes already override it).\n\nPhased plan in claude-notes/plans/2026-04-30-sidebar-vertical-border.md:\n- Phase 1: port the @if $sidebar-border SCSS rule.\n- Phase 2: introduce derive_doc_scss_layer hook in CompileThemeCssStage; emit $sidebar-border per-document. Generic enough to reuse for sidebar/navbar/footer bg+fg later.\n- Phase 3: add 'border: Option<bool>' field to Sidebar struct for full Q1 parity.\n- Phase 4: end-to-end verification + docs.\n\nPhase 2 also tightens the theme CSS cache key to include the doc-vars layer so two docs with different sidebar configs don't collide on the generic 'default_minified' key.\n\nPlan file: claude-notes/plans/2026-04-30-sidebar-vertical-border.md","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-04-30T20:16:52.060454Z","created_by":"cscheid","updated_at":"2026-04-30T20:36:56.920069Z","closed_at":"2026-04-30T20:36:56.919907Z","close_reason":"Implemented across 4 phases: Phase 1 ported the @if $sidebar-border SCSS rule from quarto-cli quarto-nav.scss:552-556 into resources/scss/bootstrap/_bootstrap-rules.scss. Phase 2 introduced derive_doc_scss_layer() in CompileThemeCssStage and quarto_sass::compile_with_doc_vars() to thread per-document SCSS variables through the bundle ahead of framework defaults. Phase 3 added Sidebar.border (Option<bool>) for full Q1 parity on the sidebar.border knob. Phase 4 verified end-to-end via cargo run --bin q2 -- render on examples/websites/03-nested-sidebar — the rule now appears in the emitted CSS — and documented the knob in docs/navigation.qmd. Follow-up bd-8oqw filed for refactoring cache_key into a structured CompileInputs value type.","labels":["scss","sidebar","websites"],"dependencies":{"bd-0tr6:parent-child":{"depends_on_id":"bd-0tr6","type":"parent-child","created_at":"2026-04-30T20:16:52.060454Z","created_by":"cscheid"}}}
 {"id":"bd-k9i1","title":"project.resources support for non-renderable site resources","description":"Q1 supports project.resources: [patterns...] in _quarto.yml, which are globs for files to copy to the output dir alongside rendered HTML but not passed through the render pipeline (e.g. CNAME, robots.txt, images not referenced from qmds). Phase 1 of the website epic does not support this. Implement in ProjectPipeline (probably a third phase after Pass 2) or WebsiteProjectType::post_render.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-04-24T01:05:17.897545Z","created_by":"cscheid","updated_at":"2026-04-24T01:05:17.897545Z","dependencies":{"bd-w5os:discovered-from":{"depends_on_id":"bd-w5os","type":"discovered-from","created_at":"2026-04-24T01:05:17.897545Z","created_by":"cscheid"}}}
 {"id":"bd-k9tfhsst","title":"Phase 2: service-worker asset proxying done properly","status":"closed","priority":1,"issue_type":"task","created_at":"2026-09-01T20:00:05.726546Z","created_by":"elliot","updated_at":"2026-09-01T22:02:21.655393Z","closed_at":"2026-09-01T22:02:21.655393Z","close_reason":"Namespace-based SW asset proxy landed (e7f9fcc2a): shared assetPolicy, proxy-URL manifest, id-correlated round trips, theme-font url() rewriting; 19 new tests + end-to-end chromium smoke green","dependencies":{"bd-r9yr0hbe:parent-child":{"depends_on_id":"bd-r9yr0hbe","type":"parent-child","created_at":"2026-09-01T20:00:05.726546Z","created_by":"elliot"}}}
@@ -1250,6 +1259,7 @@
 {"id":"bd-lu2w2fk6","title":"Perf: BinaryDependencies::discover runs which() for 5 binaries twice per document","description":"BinaryDependencies::discover (sass, esbuild, pandoc, typst, git via NativeRuntime::find_binary -> which::which) is called per document from AstTransformsStage (ast_transforms.rs:174) and again from render_to_file.rs:306 (plus attribution_generate.rs:118). Each which() does current_dir + a stat per PATH entry: 194 samples (1.6% of CPU, ~10% of all open/stat syscalls) in the warm Connect docs render. Discover once per process (OnceLock) or once in ProjectContext and pass it down; the env-var override can still be honoured at that single point. Profile: claude-notes/research/2026-09-17-connect-docs-render-profile.md","status":"open","priority":3,"issue_type":"task","created_at":"2026-09-17T20:07:36.458960Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T20:07:36.458960Z","labels":["perf"],"dependencies":{"bd-5yektmwt:discovered-from":{"depends_on_id":"bd-5yektmwt","type":"discovered-from","created_at":"2026-09-17T20:07:36.458960Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-lua-filter-table-form-ignored-ph23becz","title":"Lua filters in the returned-table form are silently ignored","description":"WHAT HAPPENS. A Lua filter written in the standard Pandoc returned-table form is read, runs none of its handlers, and produces no error, warning, or diagnostic of any kind:\n\n    -- silently does nothing\n    return { Str = function(el) … end }\n\n    -- works\n    function Str(el) … end\n\nBoth forms are valid Pandoc, and Quarto 1 runs both. The returned-table form is the more common of the two in real filters: it keeps handlers out of the global namespace, and it is the only way to express an ordered list of passes (`return { {Str=…}, {Para=…} }`), which is likewise ignored today.\n\nVerified with identical handler bodies in both forms, for `filters:` declared in project metadata (`_quarto.yml`), in document frontmatter, and on a standalone single-file render — all three behave the same.\n\nThe failure mode is the worst available one for a project migrating extensions: a declared filter quietly does nothing, and nothing in the output says so.\n\nWHERE IT LIVES. crates/pampa/src/lua/filter.rs. `apply_lua_filter` (~line 221) loads the filter chunk with\n\n    lua.load(&filter_source).set_name(…).exec_async().await?;\n\n`exec_async` DISCARDS the chunk's return value, so a top-level `return { … }` goes nowhere. The handler table is then rebuilt from scratch by `get_filter_table` (~line 323), which only ever reads globals: it walks a hardcoded list of ~50 element names and copies any same-named global function into a fresh table.\n\n`get_filter_table`'s own doc comment states the intended contract and describes behavior the code does not implement:\n\n    // Pandoc filters can either:\n    // 1. Return a table with filter functions\n    // 2. Define filter functions as globals\n    // We'll support both by creating a table that checks globals\n\nCase 1 was never wired up.\n\nNothing downstream needs to change. `apply_full_filter` already takes a handler table as an argument, and the repro proves the table-consuming path works end to end: a filter defining a global `Pandoc` handler that calls `doc:walk{Str = …}` fires correctly and rewrites the marker. The bug is confined to the entry point that throws the returned table away.\n\nSUGGESTED FIX DIRECTION. Load the chunk with `eval_async::<Value>()` instead of `exec_async()` and inspect the result:\n\n  - a table of handlers → use it directly, with no name whitelist (matching Pandoc, so filters may define handlers q2's hardcoded list omits);\n  - a SEQUENCE of handler tables → apply them as successive passes, in order. This shape is why authors reach for the returned-table form in the first place, so a fix handling only the single-table case is incomplete;\n  - nil / anything else → fall back to the existing globals scan, so the currently-working form keeps working.\n\n`get_walking_order` (~line 88) already reads `traverse` off a filter table, so per-table traversal mode comes along for free.\n\nIf the returned-table form is instead meant to stay unsupported — q2's own docs at docs/guides/authoring/lua-filters.qmd show only the top-level function form, so this may be a deliberate subset — then it needs a load-time diagnostic. A chunk that returns a table and defines no recognised globals is unambiguously this mistake, and it is silent today.\n\nREAL-WORLD HIT (Posit Connect docs). The mermaid-zoom extension (_extensions/posit-dev/mermaid-zoom/, declared as `filters: [mermaid-zoom]`) is written in the returned-table form, so it never runs: the pan/zoom overlay for large diagrams is silently absent from all 14 diagram pages, with no hint that a declared filter did nothing. Diagnosing that took a bisect against a hand-written control filter, because there is no signal anywhere in the render output.\n\nADJACENT FINDING, same investigation, likely worth its own decision. User filters run BEFORE the mermaid transform under q2: a mermaid block is still a CodeBlock with class \"mermaid\" when user filters see it, and crates/quarto-core/src/transforms/mermaid.rs converts it to a RawBlock later, at TransformPhase::Finalization. Quarto 1 is the other way round — mermaid-zoom.lua's own comment says \"By the time user filters run, Quarto has already turned a {mermaid} cell into a RawBlock of HTML\", which is why it matches RawBlock. So a ported filter that reacts to diagrams must match CodeBlock under q2 and RawBlock under Q1. Whether that ordering is intended is a separate question from this strand, but it bites the same extensions.\n\nAlso noted in passing: quarto.doc.add_html_dependency works (assets land under _files/<name>/), but warns Q-11-1 that field 'version' is not yet supported and will be ignored.\n\nREPRO. /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/lua-filter-table-form-ignored/ — run `q2 render` there. Four documents, identical handler bodies, one per form:\n\n  index.qmd      + table-form.lua     `return { Str = … }`                  → MARKER unchanged, filter did not run\n  list-form.qmd  + list-form.lua      `return { {Str=…}, {Str=…} }`         → MARKER unchanged, filter did not run\n  control.qmd    + function-form.lua  `function Str(el)`                    → FUNCTION-FORM-RAN, works\n  walk-form.qmd  + walk-form.lua      global Pandoc + `doc:walk{Str = …}`   → WALK-TABLE-RAN, works\n\nREADME.md carries expected vs actual, the root-cause analysis above, and the adjacent findings. That repo is local-only, hence the absolute path.\n\nOrigin strand in the connect-docs porting skein: br-u35hvil9. Observed with q2 0.16.0 and re-verified failing on 0.17.0 / origin/main @ 001cb6a5.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-08-11T18:48:05.131353Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-12T17:20:40.996236Z","closed_at":"2026-08-12T17:20:40.996236Z","close_reason":"Fix merged in PR #508 (24c6a842), shipped in v0.18.0. Verified end-to-end in installed q2 0.19.0 using the q2-connect-docs repro (llms-info/repros/lua-filter-table-form-ignored): returned-table, ordered-list, walk-table, and top-level-function forms all execute. Plan closed out in e40559c3.","labels":["pampa","parity"],"comments":{"c-9rdyl3a7":{"id":"c-9rdyl3a7","author":"Carlos Scheidegger","created_at":"2026-08-11T20:44:27.253272Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/508 (branch bugfix/bd-lua-filter-table-form-ignored-ph23becz, 5 commits). Local gate is green: cargo xtask verify passed all 14 steps — Rust 11708/11708 with -D warnings, hub-client 861+109, WASM 131/131 against a freshly built artifact. CI running. Note main was left untouched at origin/main; the work never landed there despite being developed in the main checkout."},"c-fa6qf5xz":{"id":"c-fa6qf5xz","author":"Carlos Scheidegger","created_at":"2026-08-11T19:14:46.001723Z","text":"INVESTIGATED 2026-08-11 (main @ 808215fc, commit b457433c). Plan skeleton:\nclaude-notes/plans/2026-08-11-lua-filter-returned-table-form.md\nArtifacts: claude-notes/plans/2026-08-11-lua-filter-returned-table-form-investigation/\n(repro copied in-tree so it survives the local-only q2-connect-docs checkout,\nplus 11 pandoc probe scripts and run-parity-matrix.sh).\n\nVERDICT: ready to design. Root cause confirmed exactly as filed. Also confirmed\nthis is not a regression - get_filter_table arrived globals-only with the\noriginal Lua landing (0f9fc8da, 2025-12-02); the 2026-04 async conversion\n(e537fb80) only changed exec() to exec_async().\n\nConfirmed the fix needs no whitelist: the walker is fully name-agnostic\n(inline_fn/block_fn look handlers up by tag; pass_is_active iterates the\ntable's own keys). The whitelist is an artifact of the globals path only.\nThe fix pattern already exists in-tree - the shortcode loader (shortcode.rs:190)\ndoes eval_async + returned-table registration already.\n\nNEW: pinned pandoc's semantics by probe against 3.9.0.2 (the version the\ndifferential oracle is pinned to, so these convert straight into cases). Rule:\nno returned value -> globals; ANY returned value wins outright and globals are\nnever consulted (rawlen 0 -> single filter, rawlen > 0 -> list of passes,\nnon-table -> error).\n\nThat last part makes strict parity a behavior CHANGE, not a pure addition.\nFive filter shapes that work today would flip:\n  mixed     global Str + return {Str=g}   pandoc TABLE-RAN   q2 GLOBAL-RAN\n  empty     global Str + return {}        pandoc nothing     q2 GLOBAL-RAN\n  emptylist global Str + return { {} }    pandoc nothing     q2 GLOBAL-RAN\n  nilret    global Str + return nil       pandoc ERROR       q2 GLOBAL-RAN\n  num       global Str + return 5         pandoc ERROR       q2 GLOBAL-RAN\nThis is design question 1 in the plan and needs a decision before implementation.\n\nADJACENT FINDING (ordering) - resolved down to a defaulting question. Verified\nby running it: the Q1-equivalent RawBlock view IS reachable today with no code\nchange. With a plain ```mermaid fence, a filter at the default position sees\nCodeBlock[mermaid]; listed after the `quarto` sentinel, or with\n`at: post-quarto`, it sees RawBlock html <pre class=\"mermaid\">. So the real\nquestion is only whether pre-quarto is the right default and whether the choice\nis discoverable - not a port blocker.\n\nCORRECTION to this strand's write-up: the class is `mermaid` only for the plain\nfence. Q1's ```{mermaid} executable-cell spelling keeps the literal class\n`{mermaid}` and renders no diagram - but that is NOT a bug: it is a deliberate,\ndocumented syntax decision (claude-notes/plans/2026-07-20-mermaid-regular-rendering.md\nsection \"Syntax decision\", guarded by a brace_form_mermaid_cell_untouched test).\n\nPre-flight: cargo xtask verify --skip-hub-build - all 11664 Rust tests pass.\nThe one failure (hub-client test:wasm, mermaid/basic.qmd expecting\nmermaid.min.js) was a stale WASM artifact from --skip-hub-build; after\nnpm run build:wasm, test:wasm is 131/131 green. HEAD is green."},"c-hk482oso":{"id":"c-hk482oso","author":"Carlos Scheidegger","created_at":"2026-08-11T20:18:35.055478Z","text":"IMPLEMENTED 2026-08-11, commit ce813478 on main (NOT pushed; awaiting review).\n\nFilter scripts now follow pandoc's rule end to end:\n  - returns nothing      -> handlers from the globals\n  - returns any value    -> that value IS the filter, globals ignored\n      table rawlen == 0  -> one filter\n      table rawlen > 0   -> list of filters applied as successive passes,\n                            each honoring its own `traverse`\n      anything else      -> Q-11-6 error\n\nTwo implementation notes worth carrying forward, both non-obvious and both\nplaces where the obvious reading of the plan is wrong:\n\n  - call_async, NOT eval_async. mlua's eval first tries to compile the source\n    as an EXPRESSION and only falls back to a block, which would make q2 accept\n    a filter file containing a bare `{ Str = f }` that pandoc rejects as a\n    syntax error. call_async compiles strictly as a block, like pandoc's\n    dofile. (Note in passing: shortcode.rs uses eval_async and has the same\n    latent widening. Not filed - different contract, no pandoc equivalent.)\n\n  - MultiValue, NOT Value. Value collapses \"returned nothing\" and \"returned\n    nil\" into the same Value::Nil, but those must diverge (globals vs error).\n    The classifier switches on a COUNT of returned values.\n\nDIAGNOSTICS. Q-11-6 (invalid script return) and Q-11-7 (unrecognized handler\nname, with an edit-distance suggestion), each with catalog entry + docs page in\nthe same commit. Live output:\n\n  Warning [Q-11-7]: Lua filter 'typo.lua' defines a handler 'Strs', which is\n  not an element type Quarto knows about, so it will never run. Did you mean\n  'Str'?\n\nQ-11-6 got its own LuaFilterError variant so it renders as\n\"Q-11-6: Lua filter 'num.lua' returned number, ...\" rather than the nested\n\"Filter error: Lua filter error: runtime error: Q-11-6: ...\".\n\nLIMITATION worth knowing: the Q-11-7 warning covers the RETURNED-TABLE form\nonly. A typo in a GLOBAL handler name (`function Strs(el)` at top level) is\nindistinguishable from an ordinary helper function, so it cannot be warned\nabout. Mitigated in docs by steering authors toward `local` helpers.\n\nbd-18a2r2lp FIXED as part of this. Rather than hand-writing the name list a\nsecond time with a test to keep it in step - which is the arrangement that\ncaused the bug, just with an alarm on it - tag_name is now generated by an\nelement_tag_names! macro emitting both the match arms and the list of names it\ncan return. The match stays exhaustive, so a new AST variant fails to compile\nuntil listed, and the constant cannot disagree with the dispatch. Drift is\nunrepresentable rather than merely detected.\n\nTESTS. 13 return-value unit tests + 2 bd-18a2r2lp regressions + 7\nwarning/edit-distance tests, all observed FAILING first (messages recorded in\nthe plan's phase log). 8 new differential cases with real pandoc 3.9.0.2\noracles; 0 existing oracles changed. The three error-shaped probes cannot be\noracle cases (the harness compares JSON ASTs, pandoc produces none) so they are\nunit tests.\n\nVERIFIED. pampa 4378/4378; workspace 11708/11708; cargo xtask lint clean. Full\n`cargo xtask verify` (with the WASM leg, since pampa is in that closure) in\nprogress. End-to-end: the four-document repro renders TABLE-FORM-RAN and\nLIST-FORM-RAN where it rendered an untouched MARKER, and run-parity-matrix.sh\nmatches pandoc on every AST-shaped row.\n\nPlan + phase log: claude-notes/plans/2026-08-11-lua-filter-returned-table-form.md\nDownstream verification strand in the connect-docs skein: br-vuf8asgk."},"c-kyop8war":{"id":"c-kyop8war","author":"Carlos Scheidegger","created_at":"2026-08-11T19:30:05.366269Z","text":"DESIGN SETTLED 2026-08-11 (commit 54a1114d). All four questions answered:\n\n1. Strict pandoc parity, wholesale - no softening for the shapes pandoc treats\n   as errors. Every divergent shape is a filter doing two conflicting things at\n   once (returning a value AND defining globals), so there is no defensible\n   \"right\" answer to pick between them; matching pandoc is the least harm. All\n   five flipping shapes flip deliberately.\n\n2. Invalid script return -> hard error with an actionable diagnostic. Behavior\n   matches pandoc; the message does not (ours names the filter path and the\n   offending Lua type). Explicit `return nil` errors too. Rationale:\n   `return { ... }` is an affirmative choice, whereas a global `Str = function`\n   can plausibly be an accidental collision - a script that returns something\n   uninterpretable has stated an intent we cannot honor.\n\n3. Warn on unrecognized handler names - adopted into Phase 3, not deferred. The\n   argument is regression surfacing more than typo catching: if q2 ever fails\n   to recognize a handler name it should support, that reaches the user as a\n   visible diagnostic instead of silence.\n\n4. New code Q-11-6 \"Invalid Lua Filter Script Return Value\", not a widened\n   Q-11-4. Docs page ships in the same commit (error-docs-page-missing lint).\n   The decision-3 warning needs its own code too, likely Q-11-7.\n\nDISCOVERED while scoping the decision-3 warning: bd-18a2r2lp. The globals\nwhitelist in get_filter_table has drifted from the walker's actual dispatch\nvocabulary (tag_name in types.rs) and silently drops five dispatchable tags:\nAttr, BlockMetadata, CaptionBlock, NoteDefinitionFencedBlock,\nNoteDefinitionPara. Proven with doc:walk, which hands the walker a table\ndirectly and bypasses the whitelist: the same handler name fires as a table\nentry and does not fire as a global. It is a live instance of exactly the class\nthe warning is meant to surface, and it settles how the warning must be built -\nderive the name set from tag_name plus the catch-alls, not a third hand-written\nlist.\n\nPlan phases are now written (Phase 0 tests-first through Phase 4 docs).\nImplementation has NOT started and awaits go-ahead."}}}
 {"id":"bd-lucp","title":"ReplayEngine miss in q2 preview: server's recorded input_qmd != WASM-produced input_qmd (mtime-derived listing-item.date-modified)","description":"After bd-m0mu (project registry plumbing) and bd-4uvv (samod automerge: URL prefix) are fixed, the SPA delivers the recorded capture bytes to the WASM. EngineExecutionStage now constructs ReplayEngine successfully — but ReplayEngine's strict byte-equality miss policy fires: 'replay miss: input QMD does not match recorded input (recorded 579 bytes, got 551 bytes).' 28-byte delta. The recorded input_qmd includes 'listing-item.date-modified: 2026-05-18' (server reads file mtime via NativeRuntime). The WASM-side q2-preview pipeline, running over the same source under WasmRuntime, doesn't have the same mtime in VFS so the listing-item.date-modified line is absent (or different). Net: every q2 preview render of a project with a website-type _quarto.yml hits a replay miss and shows the error overlay instead of rendered output. Repro fixture: ~/Desktop/daily-log/2026/05/15/q2-preview-test-website. Diagnosed end-to-end on 2026-05-18. Design space (pick one, needs user judgment): (a) compute_input_qmd strips mtime-derived metadata before serialization on server; (b) WasmRuntime populates mtimes in VFS so listing-item.date-modified surfaces in WASM too; (c) ReplayEngine loosens miss policy under preview mode (probably wrong — replay is a regression-testing tool); (d) cell-only diff for input_qmd canonicalization (the Phase-C polish path the C.3 plan flagged as Q-C3 v2). Plan §Q-C3 originally accepted whole-QMD byte-equality as a known v1 limitation; this surfaces it as a hard user-facing blocker.","status":"closed","priority":0,"issue_type":"bug","created_at":"2026-05-18T15:07:11.786286Z","created_by":"cscheid","updated_at":"2026-07-27T22:47:02.615+00:00","closed_at":"2026-07-27T22:47:02.615+00:00","dependencies":{"bd-kw93:parent-child":{"depends_on_id":"bd-kw93","type":"parent-child","created_at":"2026-05-18T15:07:11.786286Z","created_by":"cscheid"},"bd-m0mu:discovered-from":{"depends_on_id":"bd-m0mu","type":"discovered-from","created_at":"2026-05-18T15:07:11.786286Z","created_by":"cscheid"}},"comments":{"c-9wgq5m1e":{"id":"c-9wgq5m1e","author":"cscheid","created_at":"2026-05-18T16:00:50Z","text":"Implementation complete 2026-05-18 on branch beads/m0mu-preview-project-replay-engine. The splice path landed exactly as designed: `quarto_core::engine::capture_splice` (9 unit tests green) + `stage::stages::capture_splice::CaptureSpliceStage` (inserted before EngineExecutionStage in `build_q2_preview_pipeline_stages`) + WASM `render_page_for_preview` deserializing capture_gz_json into a typed EngineCapture and threading it through. E2E verified against ~/Desktop/daily-log/2026/05/15/q2-preview-test-website: iframe DOM now contains `<div class=\"cell\">` with `<div class=\"cell-output cell-output-stdout\"><pre><code>Hello, world</code></pre></div>` for the R cell. Live-edit case (appending a new prose paragraph) verified: the captured engine output survives the edit (which byte-equality replay never could). Awaiting user review for merge."},"c-huwep91o":{"id":"c-huwep91o","author":"cscheid","created_at":"2026-05-18T15:41:13Z","text":"Design pivoted 2026-05-18 after user clarification: preview should NOT consume captures through ReplayEngine (which is a deterministic regression-testing tool with intentionally strict miss policy). Instead, derive an AST-level transformation from (parse(input_qmd), parse(result.markdown)) and splice it onto the live-edited pre-engine AST. Full design at claude-notes/plans/2026-05-18-q2-preview-project-replay-engine.md. Supersedes bd-m0mu (closed). bd-4uvv stays valid. Key shape per user note: (structural_hash(cell_content), occurrence_index_in_doc_order) to disambiguate repeated identical cells."}}}
+{"id":"bd-lwvumpfk","title":"Config paths Phase 1: migrate the remaining doc-dir-joining consumers to resolve_config_path","description":"Follow-on to bd-nqq41zgx (plan claude-notes/plans/2026-10-28-config-path-late-binding.md, Phase 1). Each site in the VIOLATIONS table of claude-notes/designs/path-resolution-model.md marked 'Phase 1 of bd-nqq41zgx' swaps its doc_dir.join(as_plain_text()) for crate::config_paths::resolve_config_path with a failing-first test in the declaring-dir shape: template (missing-file case) + template-partials (HTML and Typst legs, closes the rest of bd-rdcvjy2s), shortcodes, highlight-style/syntax-highlighting theme files, language: file, font-paths, listing.template (bd-o1meelim), revealjs logo, brand: read from merged metadata (four call sites), book.chapters leading /, front-matter image / listing-item.image (bd-epk8cmdl), title-block-banner. Also route every Path-kind consumer listed in plan §4.6 through the resolver with the stored form unchanged (behaviour-preserving), so Phase 1b can flip the stored form in one commit. resolve_metadata_path becomes a wrapper over the resolver. Full sweep: claude-notes/issue-reports/945/consumer-sweep-2026-10-28.md.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-10-28T14:41:40.484200Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T16:58:52.817546Z","closed_at":"2026-10-28T16:58:52.817546Z","close_reason":"Phase 1 merged in PR #955: every remaining filesystem-space consumer resolves config paths against the declaring file via resolve_config_path; scope-outs bd-7328beha (revealjs logo) and bd-epk8cmdl (listing images).","dependencies":{"bd-epk8cmdl:related":{"depends_on_id":"bd-epk8cmdl","type":"related","created_at":"2026-10-28T14:41:40.484200Z","created_by":"Carlos Scheidegger"},"bd-hjv5o:related":{"depends_on_id":"bd-hjv5o","type":"related","created_at":"2026-10-28T14:41:40.484200Z","created_by":"Carlos Scheidegger"},"bd-nqq41zgx:related":{"depends_on_id":"bd-nqq41zgx","type":"related","created_at":"2026-10-28T14:41:40.484200Z","created_by":"Carlos Scheidegger"},"bd-o1meelim:related":{"depends_on_id":"bd-o1meelim","type":"related","created_at":"2026-10-28T14:41:40.484200Z","created_by":"Carlos Scheidegger"},"bd-rdcvjy2s:related":{"depends_on_id":"bd-rdcvjy2s","type":"related","created_at":"2026-10-28T14:41:40.484200Z","created_by":"Carlos Scheidegger"}},"comments":{"c-255km0mb":{"id":"c-255km0mb","author":"Carlos Scheidegger","created_at":"2026-10-28T15:53:27.107182Z","text":"Phase 1 implemented on braid/bd-lwvumpfk-config-paths-phase-1 (stacked on issue-945 / PR #954), four commits: (1) PathCtx::for_document + ConfigPath, template-partials/shortcodes/language file; (2) title-block-banner + listing.template (bd-o1meelim) with the closure-based existence policy; (3) the pandoc legs (epub-*, Typst template/partials/bib/csl, reference-doc, highlight-style, font-paths); (4) brand: via a quarto-sass BrandPathResolver closure, include-*/css/components routed through the resolver, book.chapters leading /. Nine failing-first tests in tests/integration/config_path_consumers.rs. Scope-outs: revealjs logo -> bd-7328beha, listing image -> bd-epk8cmdl (URL space). Behaviour change to review: a brand: in a _metadata.yml now resolves against that directory (one fixture respelled ../_brand.yml); plan notes the open question of a Q1-style project-root fallback for existence-driven keys. Workspace check + 7,712 tests green across quarto-core/quarto/quarto-test/quarto-sass; clippy clean; full verify with WASM rebuild in progress, then PR."},"c-cr8gu9t3":{"id":"c-cr8gu9t3","author":"Carlos Scheidegger","created_at":"2026-10-28T16:01:26.578572Z","text":"PR https://github.com/quarto-dev/q2/pull/955 opened against main (PR #954 merged first, so the diff is Phase 1 only). Full verify green with the WASM artifact rebuilt; clippy clean for quarto-core and quarto-sass."},"c-ildq7hho":{"id":"c-ildq7hho","author":"Carlos Scheidegger","created_at":"2026-10-28T16:58:49.183799Z","text":"PR #955 merged to main 2026-10-28 after review + clean CI."}}}
 {"id":"bd-ly83qewg","title":"External scanner lexes '< text >' with inner whitespace as html_element, swallowing inline closers (unclosed-emphasis errors)","description":"Repro: '*a < b this text is interpreted as an HTML element.* a > b' fails with Q-2-12 Unclosed Star Emphasis. parse_open_angle_brace in crates/tree-sitter-qmd/tree-sitter-markdown/src/scanner.c scans from '<' to the next '>' and emits a best-effort HTML_ELEMENT token (54 bytes here), swallowing the closing '*' of the emphasis. Deleting the '>' makes the error vanish because the bd-j9cf EOF fallback then emits LT_STR_LITERAL instead. Fix: whitespace immediately after '<' disqualifies HTML_ELEMENT (per CommonMark/HTML an open tag's name must immediately follow '<'), so bare '< ' becomes a literal Str, matching Pandoc. Plan: claude-notes/plans/2026-08-07-angle-bracket-inner-whitespace.md","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-07T23:56:14.638024Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-08T00:11:32.629046Z","closed_at":"2026-08-08T00:11:32.629046Z","close_reason":"Fixed and verified; commit 59d1edf9 on braid/bd-ly83qewg-angle-inner-whitespace (not yet pushed)","comments":{"c-jwe8sptl":{"id":"c-jwe8sptl","author":"Carlos Scheidegger","created_at":"2026-08-07T23:57:07.232636Z","text":"Reproduced and diagnosed. Plan (awaiting review, not yet implemented): claude-notes/plans/2026-08-07-angle-bracket-inner-whitespace.md. Key evidence: 'pampa -v' shows 'lexed_lookahead sym:html_element, size:54' — parse_open_angle_brace's HTML_ELEMENT arm has no whitespace guard (unlike AUTOLINK's could_be_autolink), so '< … >' spanning an emphasis closer swallows the '*'. Pandoc parses the same input as literal Str '<' / '>' with the emphasis closed. Proposed fix: whitespace immediately after '<' disqualifies HTML_ELEMENT; RAW_SPECIFIER path untouched; bd-j9cf LT_STR_LITERAL fallback covers the rest."},"c-s9h246v2":{"id":"c-s9h246v2","author":"Carlos Scheidegger","created_at":"2026-08-08T00:11:31.147341Z","text":"Fixed on branch braid/bd-ly83qewg-angle-inner-whitespace (commit 59d1edf9). TDD: 3 corpus cases + 3 pampa tests written and verified failing first (repro died with Q-2-12). Fix: html_possible guard + LT_STR_LITERAL fast path in parse_open_angle_brace; RAW_SPECIFIER path unchanged; whitespace-before-'>' deliberately not disqualifying ('<div >' is a valid open tag). Verification: tree-sitter corpus 545/545; full 'cargo xtask verify' green (workspace build+tests, WASM, hub-client); no .snap churn. E2E (output inspected): 'target/debug/pampa repro.qmd' on '*a < b this text is interpreted as an HTML element.* a > b' → [ Para [Emph [Str \"a\", Space, Str \"<\", …, Str \"element.\"], Space, Str \"a\", Space, Str \">\", Space, Str \"b\"] ] — structurally identical to pandoc -t native. Plan: claude-notes/plans/2026-08-07-angle-bracket-inner-whitespace.md"}}}
 {"id":"bd-lzugm2f4","title":"Update ejs-listing-port skill to recommend lmd (Lua markdown) listings over doctemplates","description":"The .claude/skills/ejs-listing-port skill (and its q2-port-analysis:ejs-listing-port twin) tells agents to port Q1 EJS listing templates to Q2 doctemplates. Main now has lmd-driven listings (bd-2oqen84t, PR #826): listings are published as data at meta.listings.<id>, 'type: metadata' resolves items and renders nothing, and an lmd (Lua markdown) cell renders them. See docs/guides/projects/listings.qmd (~l.353-400, 'listings as data'), docs/guides/authoring/components.qmd, claude-notes/plans/2026-10-13-lmd-listings.md, and the example in 7e9bfb329 (interactive listing: metadata listing + lmd cell).\n\nScope: rewrite SKILL.md and references/worked-examples.md so the recommended EJS replacement is type: metadata + an lmd cell reading doc.meta.listings.<id>.items, keeping what still applies (markdown links/images so .qmd rewriting and resource copying work; description-placeholder behavior if it carries over — verify). Decide whether doctemplates stay as a documented secondary path or are dropped from the skill, and update the skill name/description triggers (Q-12-7/9/24/10 warnings, EJS port phrasing) accordingly. Also reconcile with the q2-port-analysis plugin copy and docs/guides/projects/listing-templates.qmd if they should agree. Verify claims against current main by rendering a ported listing.","status":"open","priority":2,"issue_type":"task","created_at":"2026-10-15T15:35:17.343134Z","created_by":"Gordon Woodhull","updated_at":"2026-10-15T15:35:17.343134Z","labels":["listings","skills"]}
 {"id":"bd-m01zzjf3","title":"CSP for the SPA document + hub CSP additions","description":"Add a Content-Security-Policy to the SPA HTML document (served by nginx / the local-prod Node proxy, which the hub never sees, so hub build_csp cannot cover it), and consider the hub-side additions frame-ancestors/base-uri/object-src/form-action.\n\nSTATUS 2026-07-28: implemented once, then REMOVED from the branch entirely (history replayed so it was never implemented) because the SPA policy BROKE THE PREVIEW PANES in hub-client. No longer part of the bd-uv8xynxk plan; re-land as standalone work. See the comment below for the diagnosis — read it before re-attempting.","status":"open","priority":2,"issue_type":"task","created_at":"2026-07-28T10:24:50.159615Z","created_by":"shikokuchuo","updated_at":"2026-07-28T12:35:20.811168Z","closed_at":"2026-07-28T11:11:40.803176Z","close_reason":"Implemented on feature/auth-hardening-current-flow (6131a283 + changelog d52d453d): hub build_csp gains frame-ancestors/base-uri/object-src/form-action; SPA document gets a CSP for the first time from a single source of truth (scripts/spa-csp.mjs) shared by the nginx template and the Node proxy. 15 tests. Plan records the three directives the plan's list was missing (wasm-unsafe-eval, Google Fonts, sandbox frame-src origin) and the two E2E gaps (nginx not installed; no real GIS browser login).","comments":{"c-3jbt1hti":{"id":"c-3jbt1hti","author":"shikokuchuo","created_at":"2026-07-28T12:35:20.811168Z","text":"Why the first attempt broke the preview (diagnosis, 2026-07-28)\n\nThe SPA policy was applied to EVERY static response, and the preview host\ndocuments are static responses too — not just index.html. Two conflicts,\nfound by code inspection:\n\n1. style-src lacked blob:. Q2PreviewIframe ships the compiled theme CSS to\n   the iframe as a blob URL (URL.createObjectURL,\n   ts-packages/preview-renderer/src/iframe/Q2PreviewIframe.tsx:405-433) and\n   q2-preview.html consumes it as <link rel=stylesheet href=blob:...>\n   (preview-renderer/src/q2-preview/entry.tsx applyTheme). An explicit\n   style-src means no default-src fallback rescues it, so the React preview\n   loses its theme.\n2. srcdoc iframes INHERIT the embedder policy. The HTML preview path assigns\n   iframe.srcdoc = html (MorphIframe.tsx:269), so the rendered document runs\n   under the SPA script-src, which had no 'unsafe-inline'. Any inline\n   <script> in rendered Quarto output is blocked. NOT confirmed against real\n   rendered output — verify before re-landing.\n\nRecover the removed code from the pre-rewrite commit 6131a283 (kept locally\non branch backup/auth-hardening-pre-h4-removal):\n  git show 6131a283 -- scripts/spa-csp.mjs hub-client/src/utils/spaCsp.test.ts\nThe H2 follow-up bf394fcc added only a comment to spa-csp.mjs.\n\nA re-land needs: blob: in style-src; a decision on the srcdoc-inherited\nscript-src; and an audit of EVERY document the static layer serves\n(index.html, q2-preview.html, q2-debug.html, q2-raw.html) rather than\nindex.html alone — a per-document policy is probably the right shape, since\nthe preview hosts and the SPA shell have genuinely different needs.\n\nProcess lesson: the plan called a real browser E2E mandatory for exactly\nthis reason ('CSP breakage is runtime-only') and it was skipped — no browser\ntooling. The curl check confirmed the header was PRESENT, which is not the\nsame property. Static assertions over index.html could not have caught it,\nbecause the preview hosts load their styles at runtime."}}}
@@ -1281,6 +1291,7 @@
 {"id":"bd-mcy2316m","title":"add_html_dependency: 'name' on a script/stylesheet entry is silently ignored (Q1 uses it as the output filename)","description":"quarto.doc.add_html_dependency accepts each scripts/stylesheets entry either as a bare string or as a table. Quarto 1 treats the table form as { name, path }, where 'name' is the filename the file is COPIED AS (init.lua:511-521 fills name from pandoc.path.filename(v.path) when absent; pandoc-dependencies-html.ts:374 does join(targetDir, file.name)). So Q1 supports copying src/zoom.js out as zoom.min.js by passing { name = 'zoom.min.js', path = 'src/zoom.js' }.\n\nq2 drops the sub-field: normalize_dep_file (crates/pampa/src/lua/quarto_doc.rs) reads only 'path' and returns a PathBuf, so read_dep_files yields Vec<PathBuf> and the sub-field never reaches Rust. store_html_dependencies (crates/quarto-core/src/dependency.rs) then derives the filename from path.file_name(). Net effect: the table form works, but 'name' is accepted and silently ignored — the output file always takes the base name of 'path'.\n\nSilent is the problem. A Q1 filter that relies on renaming produces a differently-named asset in q2 with no diagnostic, and the emitted <script src> follows the q2 name, so the page still works and the divergence is invisible until someone compares outputs.\n\nImpact is low: renaming on copy is rare, and the common bare-string form is unaffected. Options are (a) honor 'name' for Q1 parity — needs a struct richer than PathBuf threaded from normalize_dep_file through HtmlDependency into dependency.rs; or (b) keep ignoring it but warn, which is cheap and removes the silence.\n\nFound while documenting add_html_dependency for bd-add-html-dependency-version-5tnub5ds (docs/guides/authoring/lua-filters.qmd); the doc text there deliberately describes only the { path = ... } form so it does not promise renaming.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-08-14T17:38:35.872030Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-14T17:38:35.872030Z","labels":["lua"],"dependencies":{"bd-add-html-dependency-version-5tnub5ds:discovered-from":{"depends_on_id":"bd-add-html-dependency-version-5tnub5ds","type":"discovered-from","created_at":"2026-08-14T17:38:35.872030Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-mdcqnl84","title":"Preview→editor scroll sync uses crude ratio matching; should use data-loc like the click path","description":"Scrolling the preview drags the Monaco editor along via SCROLL-RATIO MATCHING (`syncPreviewToEditor` in `hub-client/src/hooks/useScrollSync.ts`): it maps the preview's scroll percentage onto the editor's scroll percentage, with a 50ms debounce and a 300ms smooth scroll. Pre-existing on both preview paths (wired at `ReactPreview.tsx:874` -> `ReactRenderer.tsx:336` -> `Q2PreviewIframe.tsx:228`, and for the HTML preview via `Preview.tsx`).\n\nProblem: the ratio tracks pixel position, not content, so on a document with uneven block heights (code blocks, display math, tables, tall divs) the editor drifts away from whatever is actually on screen in the preview. Measured while working on the click-to-editor-scroll plan: a block at source line 73 of 78 sits at preview scroll ratio 0.17, and an editor placed at ratio 0.17 lands near line 13. Gordon noticed it as 'the editor scrolling in tandem with the preview, with a lag, without clicking anything' (2026-08-21).\n\nProposed fix, now cheap because the primitive exists: on preview scroll, resolve the topmost visible `[data-loc]` block in the preview and reveal ITS line, the same way the new click path does — `lineForClickTarget` / `revealEditorLine` in `ts-packages/preview-renderer/src/iframe/scrollSyncDom.ts` and `useScrollSync.ts`. `findElementForLine` already does the line->element direction.\n\nConstraints a fix must respect:\n- `syncPreviewToEditor` is SHARED with the HTML preview, which the click-to-editor-scroll plan deliberately left untouched (decision D3, because it was the path reported as working). Changing it affects both previews — decide deliberately whether to change one or both.\n- U4 in `hub-client/src/hooks/useScrollSync.test.ts` is the regression cover for the current ratio behaviour (asserts `setScrollTop(300, 1)` for ratio 0.5 over a 1000/400 editor, plus the focus gate). Any replacement needs to update or supersede U4 deliberately, not incidentally.\n- `revealEditorLine` brackets itself with `isSyncingRef` for 300ms so a post-click reflow can't let ratio sync overwrite a loc-based reveal (see `claude-notes/plans/2026-08-21-preview-click-to-editor-scroll.md`, 'the reveal-then-overwrite race'). A loc-based scroll path should reuse that idiom rather than inventing a second one.\n\nFound while executing claude-notes/plans/2026-08-21-preview-click-to-editor-scroll.md. Out of scope for that plan, which only added U4 as regression cover for the existing ratio path.","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-08-22T14:08:03.800352Z","created_by":"Gordon Woodhull","updated_at":"2026-08-22T15:18:48.796577Z","closed_at":"2026-08-22T15:18:48.796577Z","close_reason":"Not a defect to fix: ratio scrolling is governed by the existing scroll-sync UI toggle, so a user who does not want the editor dragged along by preview scrolling simply turns it off — and click-to-align still works with the toggle off, because revealEditorLine has no enabledRef gate (unlike syncPreviewToEditor, flushPendingScroll and scrollToLineDeferred, which all check it). Gordon's call, 2026-08-22.\n\nPreserving the finding this strand turned up, since it outlives the strand: ratio matching for the preview->editor leg was NOT a deliberate design choice. The 2025-12-29 matched-scrolling plan (claude-notes/plans/2025-12-29-matched-scrolling-hub-client.md, issue k-suww) specified that leg as 'find topmost visible element with data-loc / parse it to get a line / calculate editor scroll position to show that line', its architecture diagram says 'scroll position -> element -> line', and steps 3-5 were never implemented. The originating commit 1716275d8 (Carlos Scheidegger, 2025-12-30) shipped findElementForLine loc-based for the OTHER leg exactly as designed while substituting 'scrollRatio = previewScrollY / previewMaxScroll' for this one; git log -S'topmost' across hub-client and ts-packages finds no scroll-sync implementation of it at any point in history. That plan is nonetheless marked Status: Completed. Recorded here rather than acted on.\n\nSuperseded in practice by claude-notes/plans/2026-08-22-click-align-editor-y.md, which delivers precise loc-based preview->editor positioning on the CLICK gesture instead of the scroll gesture.","comments":{"c-vdmv73di":{"id":"c-vdmv73di","author":"Gordon Woodhull","created_at":"2026-08-22T14:11:13.523852Z","text":"ORIGIN TRACED (2026-08-22). This is not a workaround for click-to-editor-scroll — it is an UNIMPLEMENTED HALF of the original matched-scrolling design, shipped as an approximation and then marked complete.\n\nIntroduced by 1716275d8 'source location in HTML preview, scroll matching' (Carlos Scheidegger, 2025-12-30), the commit that created useScrollSync.ts. Its design doc is claude-notes/plans/2025-12-29-matched-scrolling-hub-client.md (issue k-suww), **Status: Completed**.\n\nThat doc's architecture diagram says the preview->editor leg is 'scroll position -> element -> line', and its algorithm section specifies it explicitly:\n\n    Preview -> Editor:\n    1. Listen to iframe scroll event\n    2. Debounce 50ms\n    3. Find topmost visible element with data-loc attribute\n    4. Parse data-loc to get line number\n    5. Calculate editor scroll position to show that line\n    6. editor.setScrollTop() without moving cursor\n\nSteps 3-5 were never implemented. Verified: the originating commit's useScrollSync.ts already computed 'scrollRatio = previewScrollY / previewMaxScroll' (line 166) for this leg, while implementing findElementForLine (line 47) loc-based for the OTHER leg exactly as designed. A git log -S'topmost' across hub-client and ts-packages finds no scroll-sync implementation of it at any point in history.\n\nSo the editor->preview leg shipped loc-based per the design; the preview->editor leg silently substituted ratio matching; and the plan was marked Completed.\n\nThis also explains a confusion recorded in claude-notes/plans/2026-08-21-preview-click-to-editor-scroll.md: that plan initially assumed the HTML preview's preview->editor precision came from the click/ratio path, then measured that ratio 'contributes nothing precise' and found the real mechanism was useSelectionSync (loc-based, off selectionchange). The ratio substitution is the root of that whole misreading.\n\nScope is now much smaller than when this strand was filed: every primitive exists.\n- parseDataLoc — exists (the design doc even contains the same function verbatim)\n- element -> line — lineForClickTarget in ts-packages/preview-renderer/src/iframe/scrollSyncDom.ts (with the fileId / active-region / section guards already worked out)\n- reveal a line without moving the cursor — revealEditorLine in hub-client/src/hooks/useScrollSync.ts (already brackets isSyncingRef so it cannot be overwritten)\nThe only missing piece is 'find the topmost visible [data-loc]' plus wiring it to the existing scroll handler in place of the ratio computation.\n\nConsider also correcting the Status of the 2025-12-29 plan, since it is marked Completed while step 3-5 of its own algorithm was not delivered."}}}
 {"id":"bd-mdet0yoy","title":"Hub-client full-text search","description":"Add full-text search to the Quarto Hub web client. Phase 1: open-project, client-side in-memory index behind a SearchProvider interface (no server changes), hooked into existing sync handlers. Phase B: relevance enrichment via WASM plaintext + DocumentProfile field boosting. Phase 2: cross-project search, riding on the upcoming server-side project-sets work (logged-in-user -> Automerge project-set doc). Design doc: claude-notes/plans/2026-06-23-hub-client-full-text-search.md","status":"open","priority":2,"issue_type":"epic","created_at":"2026-06-23T18:22:53.639952Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-23T18:45:56.410389Z","comments":{"c-qvcb93by":{"id":"c-qvcb93by","author":"Carlos Scheidegger","created_at":"2026-06-23T18:45:56.410389Z","text":"Phase 1 shipped on epic/hub-client-full-text-search; PR #334 (https://github.com/quarto-dev/q2/pull/334) opened against main for CI. Phase 1 strands all closed; Phase B (bd-b5ub29xd) and Phase 2 (bd-ghao01p6) remain open."}}}
+{"id":"bd-mejgdrlh","title":"Body links that climb above the project root are silently retargeted inside the site","description":"resolve_to_project_root (crates/quarto-core/src/transforms/navigation_href.rs) pops '..' segments on an empty stack without signalling. From guide/index.qmd, [x](../../outside.csv) renders as href=\"../outside.csv\" — i.e. <site>/outside.csv, a different (nonexistent) file — with no diagnostic. Same clamp applies to images via resolve_static_resource_href. Expected: leave the href verbatim (or warn) when it escapes the project root. Found while investigating bd-twfep08m.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-29T18:22:49.278500Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-29T18:22:49.278500Z","labels":["website"],"dependencies":{"bd-twfep08m:discovered-from":{"depends_on_id":"bd-twfep08m","type":"discovered-from","created_at":"2026-10-29T18:22:49.278500Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-mermaid-cell-options-9wo3crl0","title":"Mermaid %%| cell options are not processed (fig-cap/fig-alt lost in the GFM fence form)","description":"DIRECTION FIRST (Carlos, 2026-08-10): we are standardizing on the GFM diagram spelling — ```mermaid, the way GitHub-flavored Markdown and other processors do it — rather than Q1's ```{mermaid} executable-cell form. The React preview layer already implements it. The sequencing for this strand is: get diagrams working in `q2 render` for HTML formats first, then add the processing that pulls the cell options into the diagram initializer code. Phase one turns out to be largely done already (see below), so the live work here is the second half.\n\nSTATE OF PLAY, verified against q2 0.16.0 with a three-section probe:\n\n  A. ```mermaid, no options        -> <pre class=\"mermaid\"> plus the once-per-document\n                                      pinned-CDN module script. Diagram draws. WORKS.\n  B. ```mermaid, with %%| options  -> diagram draws, but the %%| lines are left verbatim\n                                      in the <pre class=\"mermaid\"> body. No <figcaption>,\n                                      no accessible description. THIS IS THE GAP.\n  C. ```{mermaid} (Q1 form)        -> <pre class=\"{mermaid} code-with-copy\">, braces carried\n                                      into the CSS class, rendered as a plain code listing.\n                                      Deliberate, per the decision recorded in the source.\n\nSo `q2 render` already has HTML diagram support: crates/quarto-core/src/transforms/mermaid.rs (bd-5m4ga0s1, plan claude-notes/plans/2026-07-20-mermaid-regular-rendering.md). It runs at TransformPhase::Finalization ahead of CodeBlockRenderTransform so diagram blocks never grow copy-button chrome, rewrites the CodeBlock to RawBlock(\"html\", ...), and appends the mermaid module to rendered.includes.after-body. It is excluded from the q2-preview pipeline (Q2_PREVIEW_TRANSFORM_EXCLUDED) because there the raw CodeBlock must survive to the React component at ts-packages/preview-renderer/src/q2-preview/blocks/MermaidCodeBlock.tsx, which pins the same mermaid version.\n\nWHAT Q1 DOES, for the part that is missing. Q1 treats %%| lines as cell options: it strips them from the diagram source and applies them, so fig-cap becomes a <figcaption> under the diagram and fig-alt becomes the accessible name. q2 currently passes them straight through. Because %% opens a mermaid comment the diagram still draws, so the failure is quiet: no warning, and the only visible symptom is the option text sitting in the diagram source plus missing captions and alt text.\n\nSUGGESTED FIX DIRECTION (phase two). In transforms/mermaid.rs, before emitting the RawBlock: split leading %%| lines off the source, parse them as the YAML cell-option block they are, and (a) keep them out of the emitted diagram text, (b) route the recognized options into the output — fig-cap wrapping the <pre> in a figure with a caption, fig-alt onto the accessible name. Worth settling while there: which option subset is honored beyond fig-cap/fig-alt (label/fig-align?), whether unrecognized options warn or are dropped silently, and whether the same parsing should feed the React component so `q2 preview` and `q2 render` agree. Note that only *leading* %%| lines are cell options; a %%| further down is an ordinary mermaid comment.\n\nMIGRATION CONSEQUENCE, possibly a separate strand. Every Q1 document in the wild spells this ```{mermaid}. Given the GFM direction, those sources need rewriting rather than q2 growing brace support — a natural qmd-syntax-helper rule (crates/qmd-syntax-helper/, Rule trait in src/rule.rs, conversions in src/conversions/), mechanically rewriting ```{mermaid} to ```mermaid and leaving the body alone. Whether that is tracked here or on its own strand is undecided.\n\nREAL-WORLD HIT (Posit Connect docs). 33 diagrams across 14 pages, all in the Q1 brace form, so today they all render as source text: the 8 admin/reference-architectures/* pages (aws and azure, each single-server / single-server-robust / load-balanced / k8s), user/orientation (4 sequence diagrams and their 4 captions), admin/appendix/off-host/arch-overview (4), admin/public-access, admin/appendix/airgapped. On the reference-architecture pages the diagram is the entire content of the page. Rewriting those files to the GFM spelling gets all 33 drawing immediately; the captions and the fig-alt accessible descriptions — which is what the accessibility of those architecture pages rests on — stay lost until the cell-option work lands.\n\nREPRO. /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/mermaid-cell-options-ignored/ — run 'q2 render repro.qmd' there. repro.qmd is the three-section probe above (A works, B drops the options, C is unrecognized) so all three outcomes are visible in one render; README.md has the table, the source-level explanation, and the docs impact. That repo is local-only, hence the absolute path.\n\nOrigin strand in the connect-docs porting skein: br-mermaid-cell-options-r52f8p3v. Observed with q2 0.16.0; re-verified against origin/main @ b2b6100c (v0.16.0 plus docs commits), where nothing has touched mermaid since v0.15.0.","status":"closed","priority":1,"issue_type":"feature","created_at":"2026-08-10T21:30:54.875001Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-10T23:15:21.048321Z","closed_at":"2026-08-10T23:15:21.048321Z","close_reason":"Implemented and merged in PR #499 (main @ a429becc). Mermaid %%| cell options are processed: fig-cap/fig-scap/fig-alt and label all route into the output, with source-mapped Q-2-47/Q-2-48 diagnostics. Follow-ups tracked separately: bd-vg8p5yoj (qmd-syntax-helper migration rule), bd-e3m3rkik (preview copy-button chrome).","labels":["parity"],"comments":{"c-3g4gaqhi":{"id":"c-3g4gaqhi","author":"Carlos Scheidegger","created_at":"2026-08-10T22:47:42.860598Z","text":"Implemented on branch feature/bd-mermaid-cell-options-9wo3crl0 (9 commits, f44b3a81..cadd7ba3). Not pushed — awaiting review. Plan + verification record: claude-notes/plans/2026-08-10-mermaid-cell-options.md\n\nAll six design questions settled with Carlos and recorded in the plan. What landed:\n\n- cell_options::comment_syntax_for learns mermaid's `%%` (Q1 keeps it on the language handler and in constants.lua, not in kLangCommentChars, which is why the port missed it).\n- crossref/codeblock_shorthand migrated off its private `#|`-only, split_once(':') matcher onto the shared cell_options facility: the marker now follows the cell language, and options are parsed as real YAML.\n- Unlabelled fig-cap emits Block::Figure — the AST-level answer Q1 could not reach, since its cell handling was textual.\n- fig-alt is injected as mermaid's native accDescr:, which survives mermaid.js swapping the <pre> for an inline <svg>.\n- Q-2-42 / Q-2-43 diagnostics, source-mapped to the offending key, scoped to diagram cells so engine options keep passing through silently.\n- Docs section + worked example examples/diagrams/03-mermaid-captions.\n\nDecision 5 is a deliberate behavior change: `#|` in a ```mermaid fence is no longer a cell-option marker. Q-2-43 explains it at the exact source line rather than letting mermaid fail on the leftovers.\n\n`cargo xtask verify` (full, including the hub/WASM leg) passes. No snapshot files changed anywhere in the workspace.\n\nAccessibility verified against real mermaid 11.12.0 under jsdom (both browser routes were unavailable this session): the injected line becomes <desc> inside the SVG with aria-describedby pointing at it; a negative control confirms mermaid rejects accDescr placed before the diagram-type line. Not verified: an actual screen reader, and the q2 preview UI in a live browser.\n\nFixes the three discovered-from strands: bd-5jcmmj1f (quoted captions), bd-sdpp9rw4 (markdown captions), bd-il6pxq4f (fig-alt dropped). Leaving all four open until the PR merges.\n\nAlso filed: bd-vg8p5yoj (qmd-syntax-helper rule for the ```{mermaid} -> ```mermaid migration — the other half of the Connect-docs story) and bd-e3m3rkik (pre-existing: preview gives mermaid blocks copy-button chrome that render suppresses)."},"c-mtfnju2p":{"id":"c-mtfnju2p","author":"Carlos Scheidegger","created_at":"2026-08-10T21:54:06.662575Z","text":"Investigated 2026-08-10; plan skeleton at claude-notes/plans/2026-08-10-mermaid-cell-options.md (commit f44b3a81 on main), probes under claude-notes/plans/mermaid-cell-options-investigation/.\n\nKey finding: the figure/caption/crossref path already works for a ```mermaid fence — it just keys off the `#|` marker. A mermaid block with `#| label: fig-x` + `#| fig-cap:` renders a full quarto-float figure (figcaption, Figure N numbering, aria-describedby) at HEAD today; only the `%%|` spelling is unrecognized. cell_options::comment_syntax_for has no mermaid entry (Q1 keeps mermaid's `%%` on the handler + constants.lua, not in kLangCommentChars, which is why the port missed it).\n\nArchitectural constraint: the work must happen in PreEngineSugaringStage (via crossref/codeblock_shorthand.rs), not in transforms/mermaid.rs — that transform is on Q2_PREVIEW_TRANSFORM_EXCLUDED (so preview would diverge) and runs after crossref-render (so labels could never be numbered). Doing it pre-engine gives preview/render parity for free.\n\nThree pre-existing defects in the shared `#|` shorthand also surfaced, reproduced on a plain python cell so they are general, not mermaid-specific: quoted YAML values keep their quotes in the figcaption; markdown in captions is not parsed; fig-alt and fig-scap are consumed and silently dropped. fig-alt being dropped is on this strand's critical path.\n\nSix design questions are open in the plan (unlabelled fig-cap, fig-alt markup, scope of the three defects, honoured option keys + unknown-key policy, one marker or two, where the work lands)."}}}
 {"id":"bd-mermaid-runtime-not-bundled-vxejw159","title":"Mermaid runtime is imported from jsDelivr at page load, not bundled into the site","description":"WHAT HAPPENS. Every HTML page containing a diagram gets a module script that imports the mermaid runtime from a third-party CDN, and nothing is written into the rendered site:\n\n    <script type=\"module\">\n      import mermaid from 'https://cdn.jsdelivr.net/npm/mermaid@11.12.0/dist/mermaid.esm.min.mjs';\n      ...\n    </script>\n\n    $ find _site -iname '*mermaid*'\n    (nothing)\n    $ ls _site/site_libs\n    bootstrap  quarto\n\nQuarto 1 bundles it instead -- site_libs/quarto-diagram/mermaid.min.js, mermaid-init.js and mermaid.css, referenced relatively -- so a Q1 site is self-contained. Under q2 it is not. Load a rendered page with the network blocked (DevTools > Network > Offline, or a block rule for cdn.jsdelivr.net) and the pre class=\"mermaid\" simply stays as visible diagram *source text*: no SVG, and nothing surfaced to the reader explaining why.\n\nThree consequences worth separating, because only the first is about offline:\n\n- Air-gapped and closed-network hosting. The rendered site cannot be archived or served where outbound HTTPS is unavailable.\n- A third-party runtime dependency in published documentation. Pages served under a vendor's domain execute code fetched from a CDN at load time -- a review question in most organizations -- and reader IP addresses reach the CDN.\n- Availability. A CDN outage or a corporate block silently degrades every diagram page.\n\nROOT-CAUSE AREA. crates/quarto-core/src/transforms/mermaid.rs: MERMAID_VERSION is pinned to \"11.12.0\" and the emitted module script hard-codes the jsdelivr URL (the format string around line 170; a unit test around line 287 asserts that exact URL). The script is appended once per document to rendered.includes.after-body via the MERMAID_JS_SENTINEL. Unchanged on origin/main at v0.17.0.\n\nSUGGESTED FIX DIRECTION. The machinery already exists: crates/quarto-core/src/dependency.rs::store_html_dependencies writes assets to _site/site_libs/libs/{name}/{filename} and is what quarto.doc.add_html_dependency goes through. The mermaid transform currently bypasses it. Registering the runtime as an ordinary HTML dependency and emitting a relative import instead of the CDN URL would put it on the same footing as Bootstrap and the quarto JS.\n\nThe genuinely open question is where the bytes come from, and it deserves a decision rather than a default: vendor the mermaid ESM build into the repo the way Q1 vendors mermaid.min.js (simple, offline-correct, but adds a multi-megabyte binary asset to the tree and a manual bump whenever MERMAID_VERSION moves), or fetch it at build time into the resources dir. Note that crates/quarto-core/resources/ currently holds only styles.css, so there is no vendored-JS precedent in that crate to follow -- the hub-client's Monaco fix went through npm/vite, a different pipeline that does not apply to `q2 render`. If bundling by default is judged too heavy, an opt-in project/format key would at least give closed-network projects a supported answer.\n\nPRIOR ART IN THIS REPO. bd-yvz2xqrm (closed) was the same class of problem for the hub client: Monaco loaded from jsDelivr at runtime and the editor hung when the CDN was slow, blocked, or offline; fixed by bundling in PR #411. A comment on that work says explicitly, \"Mermaid's lazy jsDelivr import noted as out of scope\" -- so this dependency was seen at the time and deferred, not judged acceptable. This strand is that follow-up.\n\nREAL-WORLD HIT (Posit Connect docs). 14 pages, 33 diagrams. The sharpest case is admin/appendix/airgapped/index.qmd -- the guide to running Posit Connect *without internet access* -- which itself contains two mermaid diagrams that now need internet access to render. Next worst are the 8 admin/reference-architectures/* pages, where the diagram is essentially the entire content of the page; each renders as a block of diagram source when the CDN is unreachable.\n\nREPRO. /Users/cscheid/repos/github/cscheid/q2-connect-docs/llms-info/repros/mermaid-runtime-not-bundled/ -- run 'q2 render' there. One page with one diagram; the README shows the emitted CDN URL, the empty `find _site -iname '*mermaid*'`, the contrasting contents of site_libs, and how to verify the offline failure in DevTools. That repo is local-only, hence the absolute path.\n\nOrigin strand in the connect-docs porting skein: br-u5v95mpd. Observed with q2 0.16.0; the emitting code is unchanged on origin/main at v0.17.0.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-11T14:19:27.082881Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-11T18:32:31.589678Z","closed_at":"2026-08-11T18:32:31.589678Z","close_reason":"Implemented and merged in PR #504 (merge commit 808215fc). The runtime is vendored at resources/mermaid/mermaid.min.js (upstream npm 11.12.0, SHA-256 07e37dfa...), embedded via include_str!, registered as an ArtifactScope::Project artifact at mermaid/mermaid.min.js, and referenced from each page with a resolver-computed relative URL. Rendered sites are self-contained; no jsDelivr reference remains in q2 render output.\n\nKey correction to the original analysis: dist/mermaid.esm.min.mjs (the file we imported) is a 26 KB stub over a 146-file / 13 MB chunk tree with per-diagram-type lazy imports, so vendoring \"the file we already import\" would not have worked. The self-contained UMD dist/mermaid.min.js is byte-identical to Quarto 1's vendored copy.\n\nVerified end-to-end: 3-page website rendered with the real binary and loaded in Chrome; flowchart, gantt and class diagrams all drew as real SVG (three distinct chunks under the ESM build) with zero external network requests; nested pages resolve ../site_libs/ and share one copy.\n\nScope deliberately left open in follow-ups: bd-1vwtdwtq (q2 preview / hub-client still CDN-dependent), bd-43gpsd7c (compressed compile-time embedding, prompted by the 2.62 MiB binary cost and the measured +7.3% WASM growth), and mermaid theming, which is tracked by the pre-existing bd-sehm2rha / bd-nj25kgbu.","labels":["parity"],"dependencies":{"bd-yvz2xqrm:related":{"depends_on_id":"bd-yvz2xqrm","type":"related","created_at":"2026-08-11T14:19:27.082881Z","created_by":"Carlos Scheidegger"}},"comments":{"c-4vbvgrl9":{"id":"c-4vbvgrl9","author":"Carlos Scheidegger","created_at":"2026-08-11T14:48:37.117400Z","text":"Design settled 2026-08-11 (commit e9422bea). Decisions:\n\n1. 2.62 MiB unconditionally in the q2 binary: accepted for now. Compressed compile-time embedding (include_compressed_str!-style) filed as bd-43gpsd7c.\n2. Bundle by default -- no opt-in key.\n3. This strand covers the 'q2 render' path only. Preview/hub-client filed as bd-1vwtdwtq, with the direction NOT to bundle mermaid into the client (2.62 MiB, only some docs use it) but to serve the same vendored bytes as an HTML dependency to preview projects that contain diagrams.\n4. Mermaid CSS/theming parity with Q1 filed as bd-93ensbpn.\n5. Version-bump ergonomics follow the reveal.js precedent (vendored dir + README procedure + node_modules drift test), no bespoke xtask step, so all vendored assets stay on one mechanism.\n\nCorrection recorded in bd-1vwtdwtq: the direction in (3) was framed as 'like we do for revealjs', but revealjs does not do this today -- render embeds vendored resources/revealjs/ via include_str!, while preview gets its JS from the bundled npm packages (@revealjs/react + reveal.js) and only its CSS from the vendored files. So it is a new mechanism to design, not a pattern to copy.\n\nThis strand is now ready to implement."},"c-66a43ul3":{"id":"c-66a43ul3","author":"Carlos Scheidegger","created_at":"2026-08-11T14:37:30.581364Z","text":"Investigated on main @ 001cb6a5; plan skeleton at claude-notes/plans/2026-08-11-mermaid-runtime-not-bundled.md (commit 6d9ecc32). Verdict: ready to design.\n\nTwo premises in this description are wrong in ways that change the plan:\n\n1. There IS a vendored-JS precedent -- repo-root resources/revealjs/ (README + include_str! + register_reveal_assets() storing Project-scoped artifacts to site_libs/ + a node_modules drift test). The description checked only the crate-local crates/quarto-core/resources/.\n\n2. dist/mermaid.esm.min.mjs -- the file we import today -- is a 26 KB stub, not the runtime: 10 static chunk imports plus ~25 dynamic per-diagram-type ones over a 13 MB / 146-file tree. So vendoring 'the file we already import' would not fix anything, and today's CDN path already makes an extra round trip per diagram type (a flowchart-only offline test would pass while a gantt page fails).\n\nThe stated open question ('where do the bytes come from') resolves cleanly: upstream npm mermaid@11.12.0 dist/mermaid.min.js is self-contained (2,748,992 B, zero dynamic imports, exports globalThis.mermaid) and is SHA-256 identical (07e37dfa...) to the copy Quarto 1 vendors. So 'vendor the upstream UMD' and 'do what Q1 does' are the same decision -- no esbuild step, no build-time fetch.\n\nAlso confirmed AstTransformsStage bridges RenderContext.artifacts both directions (ast_transforms.rs:182 and :227), so the transform can register the asset with no new plumbing. That was the plan's only feasibility risk.\n\nOpen questions for the user are in the plan: (1) is 2.62 MiB unconditionally in the q2 binary acceptable; (2) bundle by default vs opt-in key; (3) does scope include the TS preview path (MermaidCodeBlock.tsx has its own jsDelivr import + a version-parity test); (4) also port Q1's mermaid CSS/theming; (5) version-bump ergonomics (xtask sync vs README procedure)."},"c-t9iczsk1":{"id":"c-t9iczsk1","author":"Carlos Scheidegger","created_at":"2026-08-11T15:52:11.525679Z","text":"Implemented on branch braid/bd-mermaid-runtime-not-bundled-vxejw159 (commits 095cb39c, eeb1eb7d). Plan: claude-notes/plans/2026-08-11-mermaid-runtime-not-bundled.md. Not yet pushed.\n\nWHAT LANDED. The runtime is vendored at resources/mermaid/mermaid.min.js (upstream npm 11.12.0, 2,748,992 B, SHA-256 07e37dfa...), embedded via include_str!, registered as an ArtifactScope::Project artifact at mermaid/mermaid.min.js, and referenced from each page with a resolver-computed relative URL. Rendered sites are now self-contained.\n\nTWO DESIGN POINTS WORTH KNOWING.\n\n1. The artifact is keyed `mermaid:runtime`, NOT `js:`. The first cut used js:mermaid:... following the dependency.rs/reveal convention, and the integration tests immediately caught two <script> tags per page: ApplyTemplateStage emits a head <script> for every js: artifact (apply_template.rs:167), so the 2.6 MiB bundle loaded twice. The fix is not \"let the template emit it\" either, because the reveal scaffold collects only js:revealjs:* (apply_template.rs:305) -- a js:-keyed runtime would emit NO tag in a deck, silently breaking every diagram in a presentation. So the transform owns its own emission into the after-body slot, which both the Bootstrap template and the reveal scaffold wire up. Pinned by runtime_is_not_keyed_as_a_template_script (unit) and revealjs_deck_bundles_the_runtime (integration).\n\n2. Staleness guarding deviates from reveal.js on one point, deliberately. Reveal compares embedded bytes against node_modules because it has two live sources that must agree; mermaid has one, and adding mermaid as a devDependency purely to anchor a comparison would pull 66 MB / 793 files into every npm install. Instead: two always-on guards (the bundle's embedded version:\"...\" must equal MERMAID_VERSION; the bundle must contain no `import(` and no `chunks/`), plus the reveal-style npm byte-compare that skips while mermaid is absent from node_modules, so it starts working automatically if bd-1vwtdwtq adds it. Mechanisms 1-4 of the precedent (vendored dir, README, include_str!, Project-scoped registration) are followed exactly.\n\nThe self-containment guard is the one that matters most: dist/mermaid.esm.min.mjs is a 26 KB stub that lazily imports a chunk PER DIAGRAM TYPE, so vendoring the wrong dist file would break offline rendering in a way a flowchart-only test would miss.\n\nVERIFICATION. 16 new tests (7 integration incl. a revealjs deck, 9 unit). Two smoke-all fixtures that asserted the CDN URL now assert a relative <script src> and forbid cdn.jsdelivr.net. Full `cargo xtask verify` including the WASM/hub leg passes. End-to-end: a 3-page website rendered with the real binary, served locally, loaded in Chrome -- flowchart, gantt and class diagrams all drew as real SVG (three distinct chunks under the ESM build), with zero external network requests; nested page resolved ../site_libs/ and got a 304 on the shared copy.\n\nCOST. 2.62 MiB in the q2 binary as accepted. Additionally, and not anticipated in the design discussion, the WASM client grew +2.82 MB (+7.3%) because quarto-core is linked into wasm-quarto-hub-client. Details and rationale for not cfg-gating it are in the plan's \"Measured cost\" section, and passed to bd-1vwtdwtq (comment c-00tq71p6), where it makes that strand's chosen design cheaper.\n\nDocs: docs/guides/authoring/diagrams.qmd said \"Viewing a page with diagrams requires network access\" -- now false for rendered output. Rewritten, with a callout that q2 preview still uses a CDN so it does not overclaim."}}}
 {"id":"bd-metadata-yml-error-swallowed-b8sjkxek","title":"Malformed _metadata.yml is silently ignored: metadata_merge swallows the directory-metadata parse error","description":"Found while fixing bd-x30aq7ae (plan: claude-notes/plans/2026-10-08-frontmatter-yaml-error-code.md).\n\ncrates/quarto-core/src/stage/stages/metadata_merge.rs (~L318) calls\n    directory_metadata_for_document(&ctx.project, &ctx.document.input, ctx.runtime.as_ref()).unwrap_or_default()\nso any Err, including a YAML syntax error in a directory's _metadata.yml, is dropped. The render continues with **no** directory metadata layers and **no** diagnostic, which leaves the user unaware that their settings were ignored. Since bd-x30aq7ae, directory_metadata_for_document returns QuartoError::Parse with a located Q-1-1 diagnostic and a SourceContext holding the file. The value is there; it just never reaches the user.\n\nFix: surface it. Decide whether a malformed _metadata.yml should abort the document (Q1 parity? check Q1's behavior) or warn and continue. In either case push the Q-1-1 diagnostic and make sure its span renders: _metadata.yml is registered in the doc SourceContexts only for successfully-read layers, so a failed file may need registering explicitly, or the ParseError's own SourceContext merged. Add a pipeline-level test: a project with a broken chapters/_metadata.yml must produce the diagnostic.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-08T18:24:47.600652Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-08T18:24:47.600652Z","dependencies":{"bd-x30aq7ae:discovered-from":{"depends_on_id":"bd-x30aq7ae","type":"discovered-from","created_at":"2026-10-08T18:24:47.600652Z","created_by":"Carlos Scheidegger"}}}
@@ -1314,6 +1325,7 @@
 {"id":"bd-mxa44voa","title":"Nested-parse source mapping drifts when the inner text was unescaped (attribute values, YAML scalars)","description":"SourceInfo::substring composes a purely AFFINE map -- resolve_byte_range does parent_start + start_offset (quarto-source-map-0.1.0/src/source_info.rs:388-403), with no validation and no clamping -- and the nested reader supplies byte offsets into the inner string (crates/pampa/src/pandoc/location.rs:213-218). That is only correct when the inner string is a byte-identical prefix-aligned slice of the parent span. It is not, in at least two places:\n\n1. Div attribute values. Attr.2 stores the value UNESCAPED and unquoted (extract_quoted_text + unescape_punctuation, crates/pampa/src/pandoc/treesitter_utils/text_helpers.rs:28-59, called from treesitter.rs:1207-1212), while AttrSourceInfo.attributes[i].1 spans the RAW quote-inclusive text (the grammar token includes the delimiters: crates/tree-sitter-qmd/tree-sitter-markdown/grammar.js:580-585; range recorded at commonmark_attribute.rs:39-51).\n\n2. YAML scalars. quarto-yaml's compute_scalar_len deliberately spans the quotes while the DECODED scalar is what gets handed to parse_config_string_as_markdown (crates/pampa/src/pandoc/meta.rs:59-66 and :240-330). Nothing compensates.\n\nMeasured on a real render: for title=\"Say \\\"hello\\\" now\" the stored value is 15 bytes and the span is 19. For title=\"Use `renv` today\" the code span sits at inner bytes 4..10, maps to 85..91, and is actually at 86..92 -- off by one before any escape is involved, and one more byte per collapsed escape. Fails silently: no panic, no clamp, just wrong rows/columns. Beyond diagnostics, SourceInfo::preimage_in feeds the incremental writer's verbatim-copy decision, so a drifted range can copy the wrong bytes.\n\nThe constraint is already documented: claude-notes/plans/2026-07-20-ipynb-surface-syntax-design.md:73-92 states that Substring/Concat compose only affine maps, notes that a Transformed variant existed and was removed as unused, and says the YAML analogue 'currently punts'. Its prescribed workaround is to unescape once at ingestion and register the logical text as its own ephemeral SourceFile. The one exemplary in-tree case is cell options (crates/quarto-core/src/cell_options/mod.rs:33-43), exact precisely because it never unescapes.\n\nProper fix: carry an offset map through the unescape and emit SourceInfo::concat pieces around each escape, or adopt the ipynb doc's ephemeral-SourceFile approach. Either way this is shared infrastructure needed by the attribute path, the YAML path, and ipynb.\n\nDiscovered while designing the callout title= fix (bd-callout-custom-title-dropped-9qi1p7iw), which works around it: it derives the exact parent from lengths alone when no escape was collapsed, and falls back to the whole-value span otherwise (bounded error, cannot leave the attribute). Evidence and worked examples: claude-notes/plans/callout-title-attribute-investigation/escape-drift-probe.md","status":"closed","priority":2,"issue_type":"epic","created_at":"2026-08-10T21:07:09.695951Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-23T18:30:46.273867Z","closed_at":"2026-08-23T18:30:46.273867Z","close_reason":"Provenance epic complete: Plans 1-3 all landed on feature/yaml-provenance. Plan 1 shipped quarto-source-map 0.1.2/0.1.3 (piecewise content provenance + ProvenanceBuilder); Plan 2 made q2's YAML and attribute decoders drive it (callout.rs workaround deleted, @quarto/annotated-qmd 0.2.0); Plan 3 audited the remaining surface and fixed the comrak NodeValue::Text drift, the codeblock-shorthand body search, the is_gapless narrowing and the diagnostic-render panic guard, and recorded the rest (cell_options constraint, quarto-xml dead-code precondition, use_cmd/config.rs's byte-equality refusal). All four children closed 2026-08-21. Follow-ups live on separately as discovered-from strands: bd-6392eba3, bd-8hrjqcx0, bd-lm75ion7. Evidence: claude-notes/plans/2026-08-20-provenance-{1,2,3}-*.md and claude-notes/research/2026-08-21-provenance-audit-findings.md.","labels":["parity"],"dependencies":{"bd-callout-custom-title-dropped-9qi1p7iw:discovered-from":{"depends_on_id":"bd-callout-custom-title-dropped-9qi1p7iw","type":"discovered-from","created_at":"2026-08-10T21:07:09.695951Z","created_by":"Carlos Scheidegger"}},"comments":{"c-if3b704m":{"id":"c-if3b704m","author":"Gordon Woodhull","created_at":"2026-08-21T01:24:18.678325Z","text":"PRIOR ART: q2 contains THREE independent hand-rolled defenses against this bug class. That is the strongest available argument for fixing it at the source with shared machinery rather than continuing to work around it per-site.\n\n1. crates/quarto-core/src/transforms/callout.rs:400-444 (attribute_value_source) -- the one this strand was discovered from. Derives the exact parent from lengths alone when no escape was collapsed (span_len == value_len, or == value_len + 2 for quotes), and falls back to the whole-value span otherwise. Bounded error, cannot leave the attribute.\n\n2. crates/quarto/src/commands/use_cmd/config.rs:229 (scalar_value_span) -- newly found. Byte-compares the raw span's text against the decoded value and returns None on mismatch rather than assuming alignment:\n\n       let start = entry.value_span.start_offset();\n       let end = entry.value_span.end_offset();\n       if self.text.get(start..end)? == parsed.as_str() { Some((start, end)) } else { None }\n\n   Its doc comment names the problem outright: \"source text is not literally its value -- the caller then reports that it cannot repoint the declaration, which is honest and leaves the user's file alone.\" So a third author independently hit this, understood it, and routed around it.\n\n3. crates/quarto-core/src/cell_options/mod.rs:196-228 -- avoids the bug structurally by never decoding: it concatenates raw byte slices and pairs each with a SourceInfo::concat piece of exactly matching length. Cited in this strand already as the exemplary case. Worth noting it is exemplary precisely BECAUSE it uses the concat-piece approach this strand proposes generalizing.\n\nEach of these costs real capability. (1) and (2) degrade to whole-span or to refusing the operation; (3) constrains its input format. All three would collapse to a single call against correct provenance.\n\n--\n\nQ2 IMPACT OF THE PLANNED quarto-yaml 0.2.0 (verified at HEAD, both by a subagent and by direct re-check):\n\n- Production call sites of YamlWithSourceInfo::new_scalar / new_scalar_with_tag in q2: ZERO.\n- Test-only call sites: 15 -- crates/quarto-config/src/convert.rs:116,120,197 (cfg(test) opens at :110) and 12 sites in crates/pampa/src/pandoc/meta.rs:501..772 (cfg(test) opens at :463). Each needs one extra SourceInfo argument. Mechanical.\n- No q2 crate re-exports YamlWithSourceInfo or YamlHashEntry, so nothing propagates to downstream consumers of q2's own crates.\n- No struct-literal construction of YamlHashEntry in q2 (only YamlHashEntry::new at convert.rs:199, test code).\n- The only production is_scalar() use on a quarto-yaml node (crates/quarto-core/src/crossref/codeblock_shorthand.rs:563) operates on parsed nodes, not self-built ones. Unaffected.\n\nBeware a grep decoy: ~50 hits for \"new_scalar\" in q2 are ConfigValue::new_scalar (crates/quarto-pandoc-types/src/config_value.rs:423), an unrelated method on q2's own type. quarto-pandoc-types has no quarto-yaml dependency at all.\n\nFORWARD-LOOKING: only TWO sites in q2 should switch to content_source_info(), and both were already known -- crates/pampa/src/pandoc/meta.rs:59 (parse_yaml_string_as_markdown_to_config) and its caller crates/quarto-core/src/transforms/config_markdown.rs:289. A sweep found no others. Explicitly ruled out as false positives, recorded so the work is not repeated: project_resources.rs:215/224 (whole-node span, no sub-offset arithmetic), use_cmd/config.rs:229 (already defensive, see above), lua/config_value.rs:610-618 (different provenance path -- see bd-x0o0pem3), theme_diagnostic.rs:331/405 (synthetic test spans).\n\nSEMVER NOTE: q2 pins quarto-yaml = \"0.1.2\" as a version dep, so ^0.1.2 applies. A breaking 0.1.3 would be pulled in silently by a routine cargo update and break q2's test build with no Cargo.toml signal; 0.2.0 requires an explicit edit and is therefore the SAFER choice in practice as well as the semver-correct one."},"c-ilq4gh5r":{"id":"c-ilq4gh5r","author":"Gordon Woodhull","created_at":"2026-08-21T01:59:36.308992Z","text":"PLAN POINTER UPDATE. Earlier comments on this epic reference claude-notes/plans/2026-08-20-diagnostic-span-integrity.md. That file has been DELETED — it was split into three plans and its material moved into them. Read these instead:\n\n  1. claude-notes/plans/2026-08-20-provenance-1-foundations.md\n     quarto-source-map 0.1.2 (ProvenanceBuilder + Location.offset floor) and\n     quarto-yaml 0.2.0 (comprehensive content_source_info). Also holds the\n     salvaged findings: the scalar-style drift table, why the length()-1\n     fallback is NOT implicated and must be left alone, why the bad offset\n     survived (Location floors column but not offset), and the rejected\n     span-plus-verbatim-flag design.\n\n  2. claude-notes/plans/2026-08-20-provenance-2-consumers.md\n     quarto-error-reporting 0.2.2 release, q2's YAML consumption, q2's div\n     attribute path driving the builder, deletion of the callout.rs\n     workaround, and the panic boundary. Holds the four-defect orientation\n     table and the shadowing audit.\n\n  3. claude-notes/plans/2026-08-20-provenance-3-audit-and-fix.md\n     comrak NodeValue::Text, the Lua liveness question, the ~132 untriaged\n     SourceInfo::original sites, quarto-xml attributes, quarto-csl and\n     quarto-citeproc (never examined), and the engine map_offset triad.\n     Audits AND fixes, with a failing test per instance.\n\nTogether these cover essentially all of this epic. The one deliberate\nexclusion is ipynb, which chose the ephemeral-SourceFile route and is not\nserved by the shared builder."},"c-wqkls0fr":{"id":"c-wqkls0fr","author":"Gordon Woodhull","created_at":"2026-08-21T01:06:31.534713Z","text":"SESSION 2026-08-19/20 — measurements from the Positron-docs ariadne crash, folded in here per the dedupe decision. bd-jfl1ba13 (filed this session before this strand was found) is CLOSED as a duplicate of this one; bd-jmquuiqh and bd-th2ah982 are re-parented under it.\n\nThis strand's analysis was correct and is confirmed independently. Adding four things it did not have.\n\n1. THE YAML BRANCH HAS A SECOND, WORSE SYMPTOM THAN ESCAPES. The quoted-scalar case is a constant -1 (base is the opening quote, not the first content byte). But MULTI-LINE BLOCK SCALARS drift by one indent-width per line crossed, and it compounds. Measured on a three-line page-footer.center, two Q-2-9 warnings that both belong on line 9 are reported at 8:10 and 9:14 -- the drift is large enough to attribute a diagnostic to the WRONG LINE. Single-line block scalars are correct (they never cross a newline); plain scalars are correct (no delimiter). So the YAML branch is wrong for quoted (constant) and multi-line block (accumulating), and right otherwise.\n\n2. IT CAN CRASH, NOT ONLY MISPOINT. When a drifted offset lands inside a multi-byte character, the diagnostic renderer panics and aborts the whole render:\n\n    thread 'main' panicked at ariadne-0.6.0/src/write.rs:84:59:\n    end byte index 37 is not a char boundary; it is inside U+2728 (bytes 35..38 of string)\n\nReal-world hit: one U+2728 in the Positron docs navbar made every full-project render exit 101 with a truncated diagnostic log, while _site/ was written correctly. Tracked as bd-ariadne-config-span-char-boundary-panic-rkqmhzrg and now FIXED defensively in quarto-error-reporting (char-boundary snapping in both renderer paths, commit 4da3385 on fix/diagnostic-span-char-boundary). That fix stops the abort but cannot recover the lost byte -- carets remain wrong until THIS strand is fixed. Note the crash risk applies to every drifting path here, not just YAML.\n\n3. A THIRD DECODER HAS THE SAME BUG, INDEPENDENT OF THE TWO LISTED HERE. crates/comrak-to-pandoc/src/inline.rs:49-52 (NodeValue::Text arm) passes comrak's RAW sourcepos as base into tokenize_text_with_source (crates/comrak-to-pandoc/src/text.rs:90-140), which does base_offset + byte_idx over comrak's DECODED text -- CommonMark unescapes backslash escapes and resolves entity references in text nodes. Every offset after the first escape in a text node is shifted. Reachable in production via pampa --from commonmark (crates/pampa/src/main.rs:334). Other NodeValue arms use whole-node spans and are safe; only the Text arm does per-offset arithmetic. Filed separately and linked.\n\n4. A FOURTH, PROBABLY-INERT SITE. crates/pampa/src/lua/config_value.rs:601-631 (the quarto.config.md Lua API) has the same call shape as meta.rs, but its base is filter_source_info() = SourceInfo::Generated with an empty anchor list, so resolve_byte_range()/map_offset() return None and the bad arithmetic never resolves to a byte position. UNVERIFIED: whether any append_anchor call site attaches an Invocation anchor to that Generated before it is read. Needs a liveness check before it is worth fixing.\n\nAGREED FIX DIRECTION (this strand's first option, chosen after independent rederivation): carry an offset map through the decode and emit SourceInfo::concat pieces. Factoring decided: a small run-accumulating builder in quarto-source-map (verbatim(src_range) / replacement(src_range, out_len) / finish() -> Concat), with each decoder driving it using its own escape rules -- YAML in quarto-yaml, backslash-escapes in pampa's attribute path, entities in comrak-to-pandoc. The escape rules are not shareable; the run-tiling machinery is.\n\nVERIFIED PREREQUISITES (against quarto-source-map 0.1.0): SourceInfo::concat(Vec<(SourceInfo, usize)>) is PUBLIC and SourcePiece's fields are all pub, so no new enum variant and no upstream API is needed -- consistent with the ipynb doc's finding that the closed SourceInfo enum need not change. concat() assigns offset_in_concat cumulatively, so pieces tile contiguously by construction; this also disposes of a latent wrinkle where Concat::length() sums piece lengths while the exclusive-end branch of map_offset tests last.offset_in_concat + last.length.\n\nWHY THE EPHEMERAL-SOURCEFILE ALTERNATIVE WAS NOT CHOSEN for these paths: it is right for ipynb (users want positions relative to the CELL) and wrong for config (users want _quarto.yml:7:16, not a position inside an anonymous decoded-scalar file). Same mechanism, opposite desired user-visible outcome. Both approaches can coexist per-path.\n\nSUGGESTED PRIORITY BUMP: this is currently P2. The description's own note that preimage_in feeds the incremental writer's verbatim-copy decision means a drifted range can copy the WRONG BYTES into output -- a correctness bug in rendered output, not only in diagnostics. crates/pampa/src/writers/incremental.rs does copy byte ranges verbatim from original_qmd. I did not trace the exact wiring end to end, so this is a recommendation rather than a confirmed escalation.\n\nPlan: claude-notes/plans/2026-08-20-diagnostic-span-integrity.md (being restructured into three plans)."}}}
 {"id":"bd-mxysy0tr","title":"Selection-sync: loop-safety, focus, reveal, debounce, settings","description":"Echo guards both directions (isSyncingRef window; editorHasFocusRef); never steal Monaco focus; optional revealRange; debounce chatty selectionchange; respect existing scrollSyncEnabled toggle. Phase 5 of bd-6dib1198.","status":"open","priority":2,"issue_type":"task","created_at":"2026-06-25T21:27:58.623295Z","created_by":"shikokuchuo","updated_at":"2026-06-25T21:27:58.623295Z","dependencies":{"bd-6dib1198:parent-child":{"depends_on_id":"bd-6dib1198","type":"parent-child","created_at":"2026-06-25T21:27:58.623295Z","created_by":"shikokuchuo"}}}
 {"id":"bd-my0o5","title":"q2 preview end-to-end verification for mermaid blocks","description":"Verify the MermaidEngine survives the q2-preview round trip (capture-splice path).\n\n**This task effectively closes PR #238's bd-iq0hp** — PR #238 documented that a browser E2E of multi-engine preview was not possible at PR time because the default registry uses real engines, FixtureEngine is test-only, and knitr/jupyter don't compose cleanly (knitr claims `{python}` via reticulate). Mermaid + knitr is the first real-engine pair with disjoint cell-class ownership (`{r}` vs `{mermaid}` never overlap).\n\nTest matrix:\n\n1. Document with mermaid block only — q2 preview renders the diagram.\n2. Document with engine: [knitr, mermaidjs], knitr block + mermaid block, with recorded captures — both render in preview.\n3. Edit the mermaid block — preview updates without re-running knitr's capture (capture invariance on engine 1's input).\n4. Confirm the jsdelivr script include arrives in the preview's HTML body. If C1 (inline RawBlock in engine markdown) is the chosen approach (per plan v2), this should Just Work because capture-splice preserves result.markdown. If C2/C3 is chosen instead, bd-cp3em becomes a blocker.\n\nPlan: claude-notes/plans/2026-05-28-mermaidjs-engine-design.md","status":"closed","priority":2,"issue_type":"task","created_at":"2026-05-28T13:45:19.961230Z","created_by":"cscheid","updated_at":"2026-06-08T18:28:47.455967Z","closed_at":"2026-06-08T18:28:47.455967Z","close_reason":"Audit 2026-06-08: q2 preview mermaid end-to-end verified (5c377134); close-commit 9a3e0fcd.","dependencies":{"bd-gwfdo:blocks":{"depends_on_id":"bd-gwfdo","type":"blocks","created_at":"2026-05-28T13:45:37.473723Z","created_by":"cscheid"},"bd-je48v:parent-child":{"depends_on_id":"bd-je48v","type":"parent-child","created_at":"2026-05-28T13:45:19.961230Z","created_by":"cscheid"}}}
+{"id":"bd-my5fqvok","title":"Should span-only documents auto-route to a computational engine?","description":"Question raised during the 2026-10-29 review of the execution-records epic (claude-notes/plans/2026-10-28-execution-records-epic.md; see EXR-2, EXR-3, JOL-1). Out of scope for the epic, which matches q1.\n\nToday a document with inline expressions (IEE) but no executable cells resolves to the markdown engine, so the spans render as literal code. q2's resolver walks only CodeBlocks (crates/quarto-core/src/engine/resolution.rs, computational_languages / walk_block_for_langs), and q1 is the same: languagesWithClasses is fence-regex only (external-sources/quarto-cli/src/core/pandoc/pandoc-partition.ts:144-166). Workaround that already exists, as in q1: an engine-selecting front matter key such as jupyter: python3 or engine: jupyter (JOL-2a makes the jupyter: key give span-only documents a kernel).\n\nShould q2 detect spans and route span-only documents automatically?\n\nPros: author intent is unambiguous; consistent with cells routing by language; mostly cheap because the unit loop is EXR-3 and JOL-2a work.\nCons: small payoff (a document with no cells can only evaluate self-contained expressions); false positives on tutorials or bug reports that show a literal inline {python} expression in prose, which would start demanding a runtime (same class as bd-3tq5u8vm); renders that succeed today could fail, since span errors are fatal; some engines cannot run spans yet (knitr before EXR-12, marimo never), so routing needs a per-engine supports-inline check; diverges from q1.\n\nVerified 2026-10-29: pampa represents a span as an inline Code node whose text starts with {lang} and a space; an escaped {{python}} stays literal text and could be excluded. Not verified: whether an inline visitor already exists to reuse, and that a zero-cell document flows through the EXR-3 unit loop.\n\nIf revisited: a resolver-side AST walk of Code inlines (roughly 100-150 lines plus tests), brace spelling only (never the classic r spelling), gated per language on the claiming engine supporting spans, plus a conformance test that the AST walk and the EXR-2 scanner agree (except spans in unbraced fences). Open policy question: with no runtime installed and only spans as evidence, error or warn and leave literal.","status":"open","priority":3,"issue_type":"question","created_at":"2026-10-29T19:15:07.599145Z","created_by":"Gordon Woodhull","updated_at":"2026-10-29T19:15:07.599145Z","labels":["execution-records","iee"]}
 {"id":"bd-myfwwmki","title":"hephaestus phase 4: light/dark .hep rendering","description":"Plan: claude-notes/plans/2026-09-17-hephaestus-hep-integration.md (Phase 4, decision 7). Native: render two SVG artifacts per .hep (Theme::invert / paper-ink swap) and emit a light/dark image pair the theme toggle understands. Preview: PlotView.setColorScheme driven by the preview's theme data-mode (preview-renderer has no color-scheme plumbing today) or 'auto' as a first cut. Blocked on a light/dark image-pair convention for Q2 (Q1's .light-content / .dark-content), which does not exist yet — see the prerequisite strand linked as blocks.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-09-17T16:15:08.738500Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T16:15:40.715285Z","labels":["hephaestus"],"dependencies":{"bd-3qych45b:parent-child":{"depends_on_id":"bd-3qych45b","type":"parent-child","created_at":"2026-09-17T16:15:29.528046Z","created_by":"Carlos Scheidegger"},"bd-74sxnthr:blocks":{"depends_on_id":"bd-74sxnthr","type":"blocks","created_at":"2026-09-17T16:15:40.715273Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-myls3dvu","title":"quarto-hub-mcp vitest: random tests hit the 5s default timeout locally","description":"Observed 2026-10-22 on macOS during cargo xtask verify --skip-hub-build, and on three standalone 'npx vitest run' in ts-packages/quarto-hub-mcp: 1-4 tests fail with 'Test timed out in 5000ms', a different set each run (conformance.test.ts BP-1 net, create-only.test.ts, expected-hash.test.ts, synced-write.test.ts). All 575 pass with --testTimeout=30000, so these are slow, not broken. Fix options: raise testTimeout for the hub-round-trip suites, or find what makes the round trip slow (debug hub binary? wasm init?). Fails verify locally.","status":"open","priority":3,"issue_type":"bug","created_at":"2026-10-22T20:48:41.702084Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-22T20:48:41.702084Z","dependencies":{"bd-rfz4e3bw:discovered-from":{"depends_on_id":"bd-rfz4e3bw","type":"discovered-from","created_at":"2026-10-22T20:48:41.702084Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-myoj9kp5","title":"Decide pipeline-wide SVG ingestion/serving posture (active content in image/svg+xml)","description":"Follow-up to bd-706b0ixu (whose paste path is raster-only by design, S1). SVG ingestion pre-exists paste: the drag/drop + asset dialog flows accept image/svg+xml today. SVG scripts do not run in <img> contexts (preview renders <img src=blob:...>), but do run when an SVG URL is opened as a document — and preview blob URLs are minted in the hub-client origin, so 'open image in new tab' on a malicious collaborator's SVG executes script on the hub origin. Also applies to any future hub endpoint serving raw project files (mitigation: CSP sandbox / Content-Disposition, cf. GitHub's sandboxed-domain approach) and to published q2 render output (standard static-site posture). Decide: sanitize on ingest (DOMPurify), restrict, or accept + document trust model for collaborators (who can already inject raw HTML blocks). See claude-notes/plans/2026-08-27-paste-image-clipboard.md §3.","status":"open","priority":2,"issue_type":"task","created_at":"2026-08-27T17:53:40.522140Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-27T17:53:40.522140Z","dependencies":{"bd-706b0ixu:related":{"depends_on_id":"bd-706b0ixu","type":"related","created_at":"2026-08-27T17:53:40.522140Z","created_by":"Carlos Scheidegger"}}}
@@ -1358,6 +1370,7 @@
 {"id":"bd-no0b6sfd","title":"Windows: q2 preview --static shutdown untested and ignores Ctrl-Break (clippy dead-code on preview_static_e2e)","description":"cargo xtask verify fails on Windows since PR #712 (commit 0e67a44c): clippy -D warnings reports Server.stdout never read in crates/quarto/tests/integration/preview_static_e2e.rs, because its only reader, sigint_exits_cleanly, is #[cfg(unix)]. The field must stay (it holds the child's stdout pipe open). Underlying gap: the Windows ShutdownSignal branch in crates/quarto/src/commands/preview_static.rs handles only ctrl_c, so Ctrl-Break / console close kill the process without graceful shutdown (unix handles SIGINT+SIGTERM), and no test exercises the Windows shutdown path. Chosen fix: also handle ctrl_break on Windows, make the shutdown e2e test portable (SIGINT on unix; CTRL_BREAK_EVENT to a CREATE_NEW_PROCESS_GROUP child on Windows via windows-sys dev-dep). Liveness check: cargo clippy -p quarto --tests -- -D warnings on Windows.","status":"in_progress","priority":1,"issue_type":"bug","created_at":"2026-09-24T10:31:35.550508Z","created_by":"cderv","updated_at":"2026-09-24T11:56:35.137400Z","comments":{"c-fykrazyf":{"id":"c-fykrazyf","author":"cderv","created_at":"2026-09-24T11:08:20.809166Z","text":"Fix committed 48323753 on braid/bd-no0b6sfd-windows-q2-preview-static: ShutdownSignal handles ctrl_break on Windows; shutdown e2e test now portable (interrupt_exits_cleanly). RED on Windows (exit 0xC000013A) -> GREEN. clippy -D warnings + fmt clean on Windows; unix side unbuilt locally (CI). Found: bd-n048wj8t (CRLF fixture no-op), bd-ymd9up1v (lazy exec no reload), audit bd-5mdm8iln. Not pushed."},"c-v6e5q0gc":{"id":"c-v6e5q0gc","author":"cderv","created_at":"2026-09-24T11:56:35.137400Z","text":"PR https://github.com/quarto-dev/q2/pull/720 opened (branch bugfix/bd-no0b6sfd-windows-q2-preview-static, commits 5b0cbe65 + 40e999cd)."},"c-vnzsl0nf":{"id":"c-vnzsl0nf","author":"cderv","created_at":"2026-09-24T11:39:03.401475Z","text":"Manual check on Windows (2026-09-24, debug build, examples/websites/01-minimal copy): Ctrl-C in Windows Terminal -> 'Received Ctrl-C, shutting down the static preview', exit 0. Ctrl-Break via Python Popen(CREATE_NEW_PROCESS_GROUP) + os.kill(CTRL_BREAK_EVENT) -> same message, exit 0. Review fix 82c9d2a6 (AllocConsole for headless runs) closed roborev 2753. Console-close (ctrl_close) still unhandled; commit 48323753 message implies otherwise - pending decision."}}}
 {"id":"bd-nqcv","title":"Glob support in project.nav-dependencies declarations","description":"Today `project.nav-dependencies` accepts a flat list of paths. Some users may want `project.nav-dependencies: [posts/*.qmd]` or `[chapters/**/*.qmd]`. v1 doesn't support globs (open question 5 in the Phase 8 plan). Wire glob expansion through DocumentProfileStage when populating the field. Defer until a real user need surfaces.","status":"open","priority":4,"issue_type":"feature","created_at":"2026-04-28T00:42:51.182984Z","created_by":"cscheid","updated_at":"2026-04-28T00:42:51.182984Z","dependencies":{"bd-0tr6:parent-child":{"depends_on_id":"bd-0tr6","type":"parent-child","created_at":"2026-04-28T00:42:51.182984Z","created_by":"cscheid"},"bd-fegm:discovered-from":{"depends_on_id":"bd-fegm","type":"discovered-from","created_at":"2026-04-28T00:42:51.182984Z","created_by":"cscheid"}}}
 {"id":"bd-nqj8gmrb","title":"q2 preview: find_project_root_above ignores _quarto.yaml","description":"crates/quarto/src/commands/preview.rs:940 find_project_root_above checks only _quarto.yml, unlike core find_project_config / render.rs find_project_root_upward which also accept _quarto.yaml. A file-arg preview in a _quarto.yaml project falls back to single-file mode. May be folded into bd-nested-projects-xyb28wnl if a shared project-marker predicate lands there (plan Q7).","status":"closed","priority":3,"issue_type":"bug","created_at":"2026-09-23T21:49:21.856049Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-24T14:01:28.526387Z","closed_at":"2026-09-24T14:01:28.526387Z","close_reason":"Fixed on braid/bd-nested-projects-xyb28wnl-nested-project-boundaries in its own commit (find_project_root_above now uses PROJECT_CONFIG_FILENAMES).","dependencies":{"bd-nested-projects-xyb28wnl:discovered-from":{"depends_on_id":"bd-nested-projects-xyb28wnl","type":"discovered-from","created_at":"2026-09-23T21:49:21.856049Z","created_by":"Carlos Scheidegger"}}}
+{"id":"bd-nqq41zgx","title":"Config-path resolution class: project-level filters: resolve per document dir (issue #945); converge on provenance-based late binding","description":"GH #945: a relative path in a project-level filters: list (_quarto.yml top-level or format.html.filters) is joined onto each consuming document's directory (filter_resolve.rs resolve_filter_path + the two existence probes), so pages in subdirectories fail with 'Failed to read filter <subdir>/<filter>.lua'. Known instance of the config-path-resolution class (claude-notes/designs/path-resolution-model.md VIOLATIONS row for filters; scoped out of bd-oejuizi9 on 2026-08-19 into bd-rdcvjy2s/bd-hjv5o). This strand carries (a) the #945 fix and (b) the class assessment + convergence plan requested 2026-10-28: survey of the coexisting mechanisms (SourceInfo late binding, !path/adjust_paths_to_document_dir, FORMAT_PATH_KEYS merge-time marking, BaseDirContext, extension read-time tables) and a plan to converge on one blessed resolver that late-binds the base dir from each value's SourceInfo. Triage record: claude-notes/issue-reports/945/triage.md. Plan: claude-notes/plans/2026-10-28-config-path-late-binding.md. Branch: issue-945 (topic branch in the main checkout).","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-10-28T14:08:32.239398Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T16:01:05.855519Z","closed_at":"2026-10-28T16:01:05.855519Z","close_reason":"Phase 0 merged in PR #954 (GH #945 fixed: project-level filters resolve against the declaring file; resolve_config_path introduced). Class work continues: bd-lwvumpfk (Phase 1), bd-okqrjh8v (Phase 1b), bd-40luf359 (lint), bd-0x1znhgb.","labels":["extensions"],"dependencies":{"bd-0x1znhgb:related":{"depends_on_id":"bd-0x1znhgb","type":"related","created_at":"2026-10-28T14:08:32.239398Z","created_by":"Carlos Scheidegger"},"bd-40luf359:related":{"depends_on_id":"bd-40luf359","type":"related","created_at":"2026-10-28T14:08:32.239398Z","created_by":"Carlos Scheidegger"},"bd-hjv5o:related":{"depends_on_id":"bd-hjv5o","type":"related","created_at":"2026-10-28T14:08:32.239398Z","created_by":"Carlos Scheidegger"},"bd-oejuizi9:related":{"depends_on_id":"bd-oejuizi9","type":"related","created_at":"2026-10-28T14:08:32.239398Z","created_by":"Carlos Scheidegger"},"bd-rdcvjy2s:related":{"depends_on_id":"bd-rdcvjy2s","type":"related","created_at":"2026-10-28T14:08:32.239398Z","created_by":"Carlos Scheidegger"}},"comments":{"c-6cno873t":{"id":"c-6cno873t","author":"Carlos Scheidegger","created_at":"2026-10-28T14:17:01.634570Z","text":"Pre-flight follow-up: the hub-client test:wasm failures were the stale pkg/ artifact (dated 2026-10-21). After npm run build:wasm, npm run test:wasm passed: 44 files, 462 tests. main @ 49cb226de is fully green; nothing in this strand's area is pre-broken."},"c-88ktm1ui":{"id":"c-88ktm1ui","author":"Carlos Scheidegger","created_at":"2026-10-28T15:00:45.379157Z","text":"Phase 0 implemented and pushed: PR https://github.com/quarto-dev/q2/pull/954 (branch issue-945 -> origin/bugfix/issue-945). New crates/quarto-core/src/config_paths.rs (resolve_config_path / resolve_config_file / PathCtx); resolve_filters takes a PathCtx at both call sites. 8 failing-first integration tests in tests/integration/filter_paths.rs, 18 resolver unit tests, 4 new filter_resolve unit tests; E2E on the four fixtures renders 2 of 2 each. cargo xtask verify --skip-hub-build green (17,463 Rust tests; hub-client 1,741 + 229 + 462 wasm after a WASM rebuild). Two findings during implementation recorded in the plan's Phase 0 handoff: a Path-kind value with a leading / is OS-absolute (outside-project rebase), and an authored OS-absolute filter path keeps working through the existence policy in resolve_config_file. Follow-ons filed: bd-lwvumpfk (Phase 1), bd-okqrjh8v (Phase 1b, blocked by Phase 1). Strand stays in_progress until the PR merges."},"c-gdqdkznn":{"id":"c-gdqdkznn","author":"Carlos Scheidegger","created_at":"2026-10-28T14:15:57.861008Z","text":"Triage done 2026-10-28 on branch issue-945 (topic branch, main checkout). Reproduced at main @ 49cb226de in all spellings: _quarto.yml filters: [hello.lua] and format.html.filters: [./hello.lua] fail for sub/page.qmd; filters: [/hello.lua] fails for every page (OS-absolute); front-matter ../hello.lua works. Root cause: filter_resolve.rs resolve_filter_path + both existence probes join document_dir; Path::is_absolute for the leading slash. Per-item SourceInfo survives the merge and the declaring files are registered in the doc SourceContext, so late binding is feasible without new plumbing. Pre-flight verify: Rust 17435 tests green, hub-client unit+integration green; test:wasm failed with the stale-artifact signature (pkg wasm dated Oct 21); rebuild + rerun in progress, result to follow. Records: claude-notes/issue-reports/945/{triage.md, consumer-sweep-2026-10-28.md, repro/, exp-*/}; plan claude-notes/plans/2026-10-28-config-path-late-binding.md (draft, awaiting review). Sweep found 12 consumption sites not in the contract inventory (template-partials typst, shortcodes, highlight-style, language file, font-paths, listing.template, revealjs logo, brand: from merged metadata, book chapter leading /, listing image prefix list, !path /x never anchored). Recommendation: one resolver resolve_config_path binding the base from SourceInfo (Path-kind = already rebased; leading / = project root; fallback doc dir); Phase 0 pilots it on filters; FORMAT_PATH_KEYS row is the fallback."},"c-r9v5yo5d":{"id":"c-r9v5yo5d","author":"Carlos Scheidegger","created_at":"2026-10-28T14:27:08.390194Z","text":"Plan review round 1 (2026-10-28): Q1 pilot the resolver on filters (decided); Q2 span-less fallback = document dir (decided); Q3 leading-/ spelling for rebased extension values, option (a) (decided); Q5 WASM unsynced-VFS degradation accepted. Q4 (stored form of marked Path values: keep doc-relative vs switch to project-root pivot) still open; plan §4.6 now lists the ~10 consumers that join the stored string onto the doc dir and proposes the switch as Phase 1b after Phase 1 routes them through the resolver. Execution not yet authorized."},"c-zkhaed1o":{"id":"c-zkhaed1o","author":"Carlos Scheidegger","created_at":"2026-10-28T16:00:49.080987Z","text":"PR #954 (Phase 0) merged to main 2026-10-28 after a clean review. Phase 1 continues on bd-lwvumpfk."}}}
 {"id":"bd-nrbxc8di","title":"Vendored: reveal.js 6.0.0 → 6.0.2","description":"Surfaced by the vendored survey 2026-10-21 (claude-notes/plans/2026-10-21-vendored-survey.md).\n\nAssets (group reveal-js): revealjs-dist 6.0.0 → 6.0.2 (patch). Follower: revealjs-scss-template (reveal-template/ split from css/theme/template). reveal-menu-css is current (2.1.0).\n\nCoupling: the vendored dist must match the reveal.js npm pin in the root, hub-client and preview-renderer package.json (exact '6.0.0' pins) — bump npm first, then re-copy dist (vendored_reveal_assets_match_npm_package enforces it). Diff reveal's css/theme/template between 6.0.0 and 6.0.2 to see whether the SCSS template split needs updating.\nProcedure: resources/revealjs/README.md, resources/scss/revealjs/README.md.\nGuards: quarto-core vendored_reveal_assets_match_npm_package; quarto-sass tests; revealjs preview parity specs.","status":"closed","priority":3,"issue_type":"chore","created_at":"2026-10-21T16:00:38.003027Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-21T16:38:17.428857Z","closed_at":"2026-10-21T16:38:17.428857Z","close_reason":"reveal.js 6.0.2 on braid/bd-xm7l-vendored-updates (a3b6b4f79); template sources unchanged upstream.","labels":["deps","vendored"],"dependencies":{"bd-xm7l:discovered-from":{"depends_on_id":"bd-xm7l","type":"discovered-from","created_at":"2026-10-21T16:00:38.003027Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-nrywksil","title":"preview parity: included list item content wrapped in <p> on preview, bare text on render","description":"Fixture: includes/nested/nested.qmd. `{{< include >}}` inside a tight bullet-list item. Render: `<li>` → bare text `\"NESTED-LIST-MARKER-XYZ included item content.\"` (no wrapper). Preview: `<li>` → `<p>` → same text.\n\n## Evidence\n\nrender.norm.txt:\n```\n<ul>\n  <li>\n    \"NESTED-LIST-MARKER-XYZ included item content.\"\n  <li>\n    \"plain second item\"\n```\n\npreview.norm.txt:\n```\n<ul>\n  <li>\n    <p>\n      \"NESTED-LIST-MARKER-XYZ included item content.\"\n  <li>\n    \"plain second item\"\n```\n\n## Root cause: unknown — two hypotheses, needs investigation\n\nThis is NOT an excluded-transform case (unlike strands 1/2/4 above) — the include machinery runs identically in both native and WASM builds. Two candidate splice sites both take the same shape: splice whatever `Block`s the included content parsed as directly into the list item slot, with no explicit downgrade from `Paragraph` to `Plain`:\n- `crates/quarto-core/src/transforms/shortcode_resolve.rs:1656` — `blocks.splice(i..=i, new_blocks)` inside `resolve_blocks`, the generic block-context shortcode splice (`ShortcodeResult::Blocks` arm).\n- `crates/quarto-core/src/stage/stages/include_expansion.rs:412` — `blocks.splice(i..i + 1, children)` inside the stage-level include expansion (doc comment at `:801-803` explicitly notes tight list items hold `Plain`, not `Paragraph`).\n\nNeither site's code differs between render and preview, and `crates/pampa/src/writers/html.rs`'s `write_list_item`/`write_blocks_inline` (around lines 1366-1389) render blocks verbatim with no tight-list Paragraph→Plain downgrade either — so if render truly shows bare text, the AST block at that position must literally be `Block::Plain`, not `Block::Paragraph`, by the time the HTML writer sees it. It is not yet established whether:\n  (a) the AST is actually the same (`Paragraph`) in both pipelines, and the divergence is in the React rendering leaves — `ts-packages/preview-renderer/src/q2-preview/blocks/Plain.tsx` (renders no wrapper — inlines flow into the surrounding context) vs `.../blocks/Para.tsx` (always renders `<p>{...}</p>`), i.e. preview renders a `Paragraph` where it should render as a `Plain`; or\n  (b) the include splice genuinely yields different Block types (`Paragraph` on one path, `Plain` on the other) between the native and WASM/preview include-resolution code paths, before either the writer or React leaf ever sees it.\n\nNeeds actual AST inspection (e.g. `q2 render --show-ast` or debug-printing the WASM AST) to determine which.\n\nFix should end with `dom-parity: true` on the fixture(s). Found by the DOM parity harness (whole-corpus survey, 2026-08-24). Plan: claude-notes/plans/2026-08-24-preview-render-dom-parity-harness.md","status":"open","priority":2,"issue_type":"bug","created_at":"2026-08-25T00:47:42.523295Z","created_by":"Gordon Woodhull","updated_at":"2026-08-25T00:49:44.743507Z","labels":["preview-parity"],"dependencies":{"bd-j3764r9a:parent-child":{"depends_on_id":"bd-j3764r9a","type":"parent-child","created_at":"2026-08-25T00:49:44.743504Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-ntnx","title":"Investigate tree-sitter grammar CRLF handling on Windows","description":"Cargo xtask verify step 4 (tree-sitter test) fails 5/451 pipe-table tests on Windows but passes on ubuntu-latest CI. Surfaced during PR 109 Windows verification on 2026-04-24. Root cause is ambiguous between a test-corpus CRLF issue (Scenario A, .gitattributes fix) and a grammar CRLF-handling bug (Scenario B, grammar.js fix). Must investigate before remediating — fixing the test would mask a real Windows-user bug if Scenario B applies. See design for reproducer and both scenarios. Related to bd-tjbr (sibling Windows dev-ergonomics work). Target branch feature/dev-setup-improvements or its own branch.","design":"Failure mode. cd crates/tree-sitter-qmd/tree-sitter-markdown && tree-sitter test reports 451 parses, 446 successful, 5 failed on Windows. All 5 failures are pipe-table tests: Example 201 (GFM), Pipe table can precede content (two variants), Pipe table can follow content, Pipe table with blank-line caption. file test/corpus/pipe_table.txt shows CRLF line terminators (426 CR paired with 426 LF). git ls-files --eol confirms i/lf w/crlf — stored as LF in the index, materialized as CRLF on Windows checkout via core.autocrlf=true. Same mechanism caused the quarto-doctemplate failures catalogued in q2-windows-verify-tests memory note.\n\nScenario A — test-corpus CRLF. Grammar is CRLF-agnostic. Parse trees for LF and CRLF inputs are identical. Tests fail only because the corpus .txt file embeds both the input text AND the expected S-expression AST, and the expected AST was written assuming LF parse output. Remediation is a scoped .gitattributes entry: crates/tree-sitter-qmd/tree-sitter-markdown/test/corpus/*.txt eol=lf. One line, followed by git rm --cached then git checkout to normalize existing files. Real Windows user qmd content continues to parse correctly regardless of line endings.\n\nScenario B — grammar CRLF bug. Grammar does not treat carriage-return-linefeed as a block boundary where it should. Real Windows user qmd content with CRLF endings parses incorrectly — pipe tables do not detect captions as separate blocks, block boundaries drift. This is a user-facing bug; only a Windows contributor running pampa against CRLF content catches it. Remediation is in crates/tree-sitter-qmd/tree-sitter-markdown/grammar.js — typically adding carriage-return to newline character classes or the external scanner logic. .gitattributes would mask this bug instead of fixing it.\n\nReproducer. Two layers, both diagnostic. Layer 1 tests grammar.js directly via tree-sitter parse. Layer 2 tests the full pampa pipeline that real users hit. Fixtures are written into the tree-sitter-markdown directory so tree-sitter parse picks up the configured grammar without an init-config step.\n\nLayer 1 (grammar.js):\n\nprintf 'before\\r\\n\\r\\n| a | b |\\r\\n|---|---|\\r\\n| 1 | 2 |\\r\\n\\r\\nafter\\r\\n' > crates/tree-sitter-qmd/tree-sitter-markdown/test-crlf.qmd\nprintf 'before\\n\\n| a | b |\\n|---|---|\\n| 1 | 2 |\\n\\nafter\\n' > crates/tree-sitter-qmd/tree-sitter-markdown/test-lf.qmd\ncd crates/tree-sitter-qmd/tree-sitter-markdown\ntree-sitter parse test-crlf.qmd > test-crlf.parse\ntree-sitter parse test-lf.qmd > test-lf.parse\ndiff test-crlf.parse test-lf.parse\n\nLayer 2 (pampa end-to-end):\n\ncargo run --quiet -p pampa --bin pampa -- crates/tree-sitter-qmd/tree-sitter-markdown/test-crlf.qmd > pampa-crlf.out\ncargo run --quiet -p pampa --bin pampa -- crates/tree-sitter-qmd/tree-sitter-markdown/test-lf.qmd > pampa-lf.out\ndiff pampa-crlf.out pampa-lf.out\n\nNotes on the prior reproducer in this design field. pampa has no parse subcommand — it is cargo run -p pampa --bin pampa -- input.qmd, with default -f markdown -t native. Git Bash /tmp resolves to %TEMP% but native Windows binaries resolve /tmp to C:\\tmp\\, so the previous version would have created files Git Bash could not find later. Use repo-local paths.\n\nIf both diffs are empty then Scenario A applies. If either diff is non-empty then Scenario B applies. Layer 2 also tells us whether the bug propagates past pampa's input handling into the Pandoc AST that real users render from.\n\nBroader git cross-platform context. The 2026 consensus is that most projects do not need .gitattributes. core.autocrlf=true plus LF-tolerant tooling handles 99 percent of cross-platform cases. quarto-cli is an example — no .gitattributes and no CRLF-related failures. .gitattributes is for files that must be a specific line ending (shell scripts, batch files), binary files misdetected as text, and whitespace-sensitive test corpora where the consuming tool is itself correct. Over-restrictive patterns like * text=auto eol=lf force LF everywhere and are worse for cross-platform dev than no config. The target for this project is minimal, scoped .gitattributes only where the test harness cannot be fixed — and only in Scenario A.\n\nOut of scope. Broader .gitattributes sweep for the whole repo. Fixing the 8 quarto-doctemplate CRLF failures already catalogued in q2-windows-verify-tests (separate work). Tree-sitter CLI changes (upstream, not ours).","acceptance_criteria":"Reproducer runs and classifies failure as Scenario A or Scenario B. If Scenario A then .gitattributes entry added for test/corpus/*.txt and cargo xtask verify step 4 passes on Windows. If Scenario B then follow-up beads issue filed for grammar.js fix at higher priority and this investigation issue closes with evidence. q2-windows-verify-tests memory note updated with resolution and link to fix PR or follow-up issue. No regression in CI — tree-sitter tests continue to pass on Linux.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-04-24T16:38:18.426543400Z","created_by":"cderv","updated_at":"2026-04-27T14:29:38.899054500Z","closed_at":"2026-04-27T14:29:38.898439400Z","close_reason":"Scenario B classified. Follow-up bd-0gsj filed at P1 for the grammar.js fix; full evidence in that issue's design and in the closing comment here.","labels":["tree-sitter","windows"],"dependencies":{"bd-tjbr:related":{"depends_on_id":"bd-tjbr","type":"related","created_at":"2026-04-24T16:39:17.513779500Z","created_by":"cderv"}},"comments":{"c-d152rz75":{"id":"c-d152rz75","author":"cderv","created_at":"2026-04-27T14:28:37Z","text":"Investigation complete. Result: Scenario B (real grammar CRLF bug). Layer 1 (tree-sitter parse on grammar.js) showed pipe_table extents diverge between CRLF and LF inputs — CRLF case absorbs the trailing paragraph as a malformed table row. Layer 2 (cargo run -p pampa) confirmed pampa does not normalize line endings before parsing, so the bug propagates to the Pandoc AST that drives every output format. Real Windows users with default core.autocrlf=true and the common pattern of a pipe table followed by content render structurally wrong documents. .gitattributes alone would have masked this.\n\nFollow-up filed at higher priority: bd-0gsj — Fix CRLF handling in tree-sitter-qmd grammar.js for pipe tables (P1, bug). Full evidence and reproducer carried over to bd-0gsj design field."}}}
@@ -1376,7 +1389,7 @@
 {"id":"bd-nzdm1wry","title":"Surface a cause-naming diagnostic when an extension needed by a render fails to load (not a silent WARN)","description":"When read_extension returns Err, discover.rs (scan_extension_entry) logs only tracing::warn!('Failed to read extension ...') and discover_extensions returns Vec (not Result), so nothing surfaces to the user. The document then renders with the extension's engine/filter/shortcode missing — e.g. a marimo/julia cell renders as raw code, or a shortcode renders as the unknown-shortcode fallback '<strong>?greet</strong>' — misattributed to the user's own document with no actionable cause. This remains true even after bd-8b0af414 relaxed over-required fields: a genuinely malformed _extension.yml (or a real quarto-required mismatch) still fails silently. Proposal: when an extension that a render actually depends on (referenced format/engine/shortcode/filter) fails to read, emit a real cause-naming diagnostic (Q-* error or at least a user-visible warning naming the extension, the file, and the specific parse/validation failure) instead of a buried tracing WARN. Discovered while fixing bd-8b0af414; out of scope for that bug (which only relaxed the field requirements).","status":"open","priority":2,"issue_type":"feature","created_at":"2026-07-08T19:16:11.186439Z","created_by":"Gordon Woodhull","updated_at":"2026-07-08T19:16:11.186439Z","dependencies":{"bd-8b0af414:discovered-from":{"depends_on_id":"bd-8b0af414","type":"discovered-from","created_at":"2026-07-08T19:16:11.186439Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-nzmwy6uc","title":"Typst compile ignores extension-vendored typst/packages (network fetch instead)","description":"Found during book-projects P2 (template-partials fix): an extension can ship its own Typst packages at <ext>/typst/packages/preview/<name>/<version>/ (orange-book's vendored subtree does exactly this for orange-book 0.7.1 + marginalia 0.3.1), but TypstCompileStage only stages the globally-vendored resources/typst-packages tree (extract_typst_packages) into --package-cache-path. Extension packages never reach the cache, so typst-gather downloads them from the network at render time — renders of orange-book books are non-hermetic and fail offline. Wire extension typst/packages trees into the compile package cache (or vendor orange-book into resources/typst-packages).","status":"open","priority":2,"issue_type":"bug","created_at":"2026-09-24T17:58:34.181811Z","created_by":"Gordon Woodhull","updated_at":"2026-09-24T17:58:34.181811Z"}
 {"id":"bd-o1f58huc","title":"hub-client eslint (npm run lint) is red and not in any CI gate (e2e spec debt)","description":"npm run lint -w hub-client (eslint .) fails with ~10 errors, all in e2e/ specs and helpers (no-unused-vars, no-explicit-any) — e.g. e2e/forced-colors.harness.spec.ts, e2e/helpers/assertAutomerge.ts, e2e/menu-keyboard.harness.spec.ts. CI only runs lint:css (ts-test-suite.yml step 'Lint hub-client CSS'), so eslint debt accumulates invisibly — same shape as the ci-test-suite-unwired class. Options: fix the ~10 errors and add an eslint step to CI, or explicitly decide eslint stays advisory. Found 2026-09-01 while fixing bd-v51cly8i.","status":"open","priority":3,"issue_type":"chore","created_at":"2026-09-01T16:08:39.673545Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-01T16:08:39.673545Z","dependencies":{"bd-v51cly8i:discovered-from":{"depends_on_id":"bd-v51cly8i","type":"discovered-from","created_at":"2026-09-01T16:08:39.673545Z","created_by":"Carlos Scheidegger"}}}
-{"id":"bd-o1meelim","title":"Listing `template:` treats a leading `/` as filesystem-absolute, contradicting the path-resolution contract (project root)","description":"load_custom_template (crates/quarto-core/src/transforms/listing_render.rs, ~:398-401) does `if template_rel.is_absolute() { template_rel.to_path_buf() }`, so `template: /templates/card.template` resolves against the filesystem root, not the project root. claude-notes/designs/path-resolution-model.md says a leading `/` means the project root, never the filesystem root. Pre-existing; noticed while documenting the new 'Custom templates' section of docs/guides/projects/listings.qmd (which says template: resolves relative to the declaring page and is silent on `/`). Deliberate scope-out from bd-custom-template-not-templated-e5t6m0i0 per the CLAUDE.md path-resolution rule — fix should use the shared resolution helper the contract names, and add the `/` case to the docs sentence.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-08-25T18:50:15.834779Z","created_by":"Gordon Woodhull","updated_at":"2026-08-25T18:50:15.834779Z","labels":["listings"],"dependencies":{"bd-custom-template-not-templated-e5t6m0i0:discovered-from":{"depends_on_id":"bd-custom-template-not-templated-e5t6m0i0","type":"discovered-from","created_at":"2026-08-25T18:50:15.834779Z","created_by":"Gordon Woodhull"},"bd-oejuizi9:related":{"depends_on_id":"bd-oejuizi9","type":"related","created_at":"2026-08-25T18:50:15.834779Z","created_by":"Gordon Woodhull"}}}
+{"id":"bd-o1meelim","title":"Listing `template:` treats a leading `/` as filesystem-absolute, contradicting the path-resolution contract (project root)","description":"load_custom_template (crates/quarto-core/src/transforms/listing_render.rs, ~:398-401) does `if template_rel.is_absolute() { template_rel.to_path_buf() }`, so `template: /templates/card.template` resolves against the filesystem root, not the project root. claude-notes/designs/path-resolution-model.md says a leading `/` means the project root, never the filesystem root. Pre-existing; noticed while documenting the new 'Custom templates' section of docs/guides/projects/listings.qmd (which says template: resolves relative to the declaring page and is silent on `/`). Deliberate scope-out from bd-custom-template-not-templated-e5t6m0i0 per the CLAUDE.md path-resolution rule — fix should use the shared resolution helper the contract names, and add the `/` case to the docs sentence.","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-08-25T18:50:15.834779Z","created_by":"Gordon Woodhull","updated_at":"2026-10-28T23:17:31.019750Z","labels":["listings"],"dependencies":{"bd-custom-template-not-templated-e5t6m0i0:discovered-from":{"depends_on_id":"bd-custom-template-not-templated-e5t6m0i0","type":"discovered-from","created_at":"2026-08-25T18:50:15.834779Z","created_by":"Gordon Woodhull"},"bd-oejuizi9:related":{"depends_on_id":"bd-oejuizi9","type":"related","created_at":"2026-08-25T18:50:15.834779Z","created_by":"Gordon Woodhull"}},"comments":{"c-dxngut23":{"id":"c-dxngut23","author":"Carlos Scheidegger","created_at":"2026-10-28T23:17:31.019750Z","text":"Code half landed in Phase 1 (2079c8c82, PR #955; test rooted_listing_template_anchors_at_project_root). Docs sentence added in PR #961: https://github.com/quarto-dev/q2/pull/961 — close on merge."}}}
 {"id":"bd-o1yn1fqy","title":"Transition quarto-hub from stable actor IDs to automerge author IDs","description":"Execution epic for the H4 root-cause fix (production duplicate-seq wedge investigated in bd-6f21d4c6; Gordon approved the production fix 2026-09-30). Plan: claude-notes/plans/2026-09-30-automerge-author-id-transition.md. Automerge 3.5.0 / 0.12.0 change-level author metadata carries the stable per-user identity (server-minted HMAC, byte-identical to the legacy actor ID per D5); actor IDs return to the automerge random per-document default (D1), removing the shared-actor collision at the source. Attribution consumers key off change.author with change.actor fallback for pre-transition history.","status":"in_progress","priority":1,"issue_type":"epic","created_at":"2026-09-30T10:49:04.460309Z","created_by":"shikokuchuo","updated_at":"2026-09-30T10:49:28.590655Z","dependencies":{"bd-6f21d4c6:discovered-from":{"depends_on_id":"bd-6f21d4c6","type":"discovered-from","created_at":"2026-09-30T10:49:04.460309Z","created_by":"shikokuchuo"}}}
 {"id":"bd-o20jxpfc","title":"Single-variant adaptive highlight resolution: compiled-CSS darkness sentinel","description":"Item (3) of bd-hl-theme-translator-2mdgh4k6, deferred by design decision 6 (2026-08-18): for custom-SCSS single-variant themes, adaptive highlight names resolve via BuiltInTheme::is_dark (builtin_darkness in quarto-sass config.rs), an approximation. Q1 instead greps the compiled CSS for a darkness sentinel (blackness($body-bg) > threshold), which handles custom SCSS that darkens a light base. Requires moving the adaptive decision after (or into) the variant compile, or a pre-compile probe. See claude-notes/plans/2026-08-18-highlight-theme-translator.md § Deferred.","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-18T18:26:23.176260Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-18T18:26:23.176260Z","dependencies":{"bd-hl-theme-translator-2mdgh4k6:discovered-from":{"depends_on_id":"bd-hl-theme-translator-2mdgh4k6","type":"discovered-from","created_at":"2026-08-18T18:26:23.176260Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-o2pm4nzt","title":"wasm-quarto-hub-client/Cargo.lock stale on main (missing quarto-node-discovery)","description":"main's quarto-core/Cargo.toml:126 depends on quarto-node-discovery (added in b889c4662 'Run TS engine extensions on Node'), but the committed crates/wasm-quarto-hub-client/Cargo.lock predates it. Any local WASM build (npm run build:all / build:wasm) dirties the file with the quarto-node-discovery entry. CI's WASM Tests leg passes anyway (no cleanliness gate). Fix: commit the refreshed lockfile on main. Discovered while restacking the hub-mcp phase 0-7 PR stack onto main.","status":"closed","priority":3,"issue_type":"chore","created_at":"2026-10-14T09:27:26.170748Z","created_by":"shikokuchuo","updated_at":"2026-10-14T11:17:09.542890Z","closed_at":"2026-10-14T11:17:09.542890Z","close_reason":"Fixed in d6674b9d4 (PR #832): committed the refreshed wasm-quarto-hub-client/Cargo.lock with the quarto-node-discovery entry"}
@@ -1400,6 +1413,7 @@
 {"id":"bd-ohvl879u","title":"Jupyter engine ignores error policy: embeds cell errors and continues where knitr fails the render","description":"Discovered during bd-gthycd33 parity-suite work (2026-07-02). With a plain failing cell (no #| error: true), q2's knitr path fails the pipeline ('Execution failed in knitr: R process failed', no capture produced) — matching Q1's default execute.error=false policy. q2's jupyter path (text_execute.rs format_outputs) instead embeds the error as .cell-output-error output and reports success, unconditionally. The engines should agree: a cell error without error: true should fail the render for jupyter too; with #| error: true both should embed. Note: q2 jupyter currently doesn't parse #| cell directives at all, so fixing this requires directive awareness in the jupyter text path. The cross-engine parity suite (crates/quarto-core/tests/integration/engine_output_parity.rs, parity_error_output) pins the error-output SHAPE under #| error: true; this strand is about the POLICY divergence. See claude-notes/plans/2026-07-01-bd-gthycd33-jupyter-cell-wrapper.md.","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-07-02T12:47:46.203497Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-02T18:51:14.386081Z","closed_at":"2026-07-02T18:51:14.386081Z","close_reason":"Fixed in PR #363 (merged): shared cell_options facility (language-aware #| partition with exact source attribution + scoped ConfigValue resolution) + jupyter engine error policy — un-annotated cell errors now fail the render like knitr/Q1; #| error: true or execute: error: true embed the error; option lines stripped from kernel input and echo. Full evidence in claude-notes/plans/2026-07-02-bd-ohvl879u-cell-options-error-policy.md. Follow-ups: bd-eizgnxlx, bd-1gty7f7o, bd-2xkpy5ra, bd-moef1ec4, bd-2lc8qu6e.","dependencies":{"bd-gthycd33:blocks":{"depends_on_id":"bd-gthycd33","type":"blocks","created_at":"2026-07-02T14:02:47.025975Z","created_by":"Carlos Scheidegger"},"bd-gthycd33:discovered-from":{"depends_on_id":"bd-gthycd33","type":"discovered-from","created_at":"2026-07-02T12:47:46.203497Z","created_by":"Carlos Scheidegger"}},"comments":{"c-0ktxw34b":{"id":"c-0ktxw34b","author":"Carlos Scheidegger","created_at":"2026-07-02T16:51:10.604150Z","text":"Decisions 1-7 locked with Carlos (2026-07-02). Notable: (3) upgraded from 'read error off the YAML' to SCOPED RESOLUTION — convert cell options YamlWithSourceInfo -> ConfigValue (pampa's yaml_to_config_value, source-info preserving) and merge against the document's already-merged metadata execute scope (merge_with_diagnostics); cell error: overrides doc execute: error:, absent both = false. This starts infrastructure for scoped document-metadata resolution in cell options, a longstanding Q1 limitation. Verified the pieces exist: yaml_to_config_value (pampa meta.rs:139), merge_with_diagnostics (quarto-config materialize.rs:188), and the engine input's front matter carries merged metadata (MetadataMergeStage runs before EngineExecutionStage). (4) accepted for v1 with corrected framing — locations in the serialized engine input are the same coordinate system the pipeline already reconciles at the AST level (quarto-ast-reconcile); documented in the plan as mechanism, not problem. (5) strip #| from kernel input + echo. (6) malformed options = hard error with mapped location. (2,7) error-only scope; crossref migration is follow-up. Plan updated; awaiting go-ahead."},"c-6t9xek6b":{"id":"c-6t9xek6b","author":"Carlos Scheidegger","created_at":"2026-07-02T18:25:34.447846Z","text":"Committed (62125080, rebased to 65eb576e onto post-#360 main) and PR opened: https://github.com/quarto-dev/q2/pull/363 (branch bugfix/bd-ohvl879u-jupyter-engine-ignores-error). Rebase note: main had moved (quarto-yaml stack extracted to posit-dev/quarto-yaml as a version dep, validate-yaml deleted, strict-mode warnings PR) — re-verified on the rebased tree: external quarto-yaml API compatible, 48 unit + 15 engine integration tests green, clippy 0, workspace 9924/9924. CI watch running; close strand after merge."},"c-bcf4pw4g":{"id":"c-bcf4pw4g","author":"Carlos Scheidegger","created_at":"2026-07-02T17:17:51.160396Z","text":"Phases 1-3 complete (2026-07-02). Facility: quarto-core/src/cell_options/ (registry ported from Q1 kLangCommentChars; partition with SourceInfo::concat per-line substrings -> quarto_yaml::parse_with_parent; options_to_config + merge_cell_over_scope for decision-3 scoped resolution). Engine: text_execute.rs partitions each cell, strips #| lines from kernel input AND echo, resolves error: via cell-over-document ConfigValue merge, aborts on disallowed cell error (new JupyterError::CellExecutionFailed / InvalidCellOptions). Better than planned: ExecutionContext already carries source_info (Concat from write_with_source_info) + SourceContext, so diagnostics render path:line:col in ORIGINAL-file coordinates — no ephemeral-file fallback needed. Tests: 28 facility unit tests, 20 text_execute unit tests, 15/15 engine integration tests (7 error-policy + behavior parity + all bd-gthycd33 fences unchanged). Workspace suite 10217/10217 passed; clippy/lint/fmt clean. Remaining: verify (full), CLI e2e, close-out follow-ups."},"c-d8or22aq":{"id":"c-d8or22aq","author":"Carlos Scheidegger","created_at":"2026-07-02T17:26:26.270468Z","text":"All phases complete (2026-07-02). Phase 4: workspace 10217/10217 (x2), full cargo xtask verify passed, clippy/lint/fmt clean. Phase 5 e2e through the real binary: plain failing cell -> exit 1 with diagnostic at fail.qmd:9:2 (ORIGINAL-file line of the raise, via ExecutionContext source-map chain) + actionable hint; #| error: true -> renders with .cell-output-error and zero #| in HTML; doc-level execute: error: true -> renders with embedded error; q2 preview spot-check (fresh WASM chain) -> healthy directive-bearing cell previews with stripped echo and spliced output. Phase 6: follow-ups filed bd-eizgnxlx (crossref migration), bd-1gty7f7o (LSP reuse), bd-2xkpy5ra (body-only CodeBlock SourceInfo), bd-moef1ec4 (eval), bd-2lc8qu6e (guessChunkOptionsFormat); the original-file-coordinates follow-up was unnecessary (already works). Staged, pre-commit checklist done; awaiting commit approval. Close after merge."},"c-hu6qsjam":{"id":"c-hu6qsjam","author":"Carlos Scheidegger","created_at":"2026-07-02T18:37:50.063991Z","text":"CI green on PR #363: test suite macos+ubuntu (both runs), WASM tests, Hub-Client E2E, snyk. Ready for review/merge; close strand after merge."},"c-vpzonm0v":{"id":"c-vpzonm0v","author":"Carlos Scheidegger","created_at":"2026-07-02T14:08:11.749345Z","text":"Design plan written after source study: claude-notes/plans/2026-07-02-bd-ohvl879u-cell-options-error-policy.md (worktree .worktrees/bd-ohvl879u-jupyter-engine-ignores-error, branch based on bd-gthycd33's — PR #360 must merge first; blocks dep recorded). Core design: shared quarto-core/src/cell_options/ facility — Q1 kLangCommentChars registry ported (prefix or prefix+suffix pairs, default #), leading-run partition per Q1's optionCommentPattern, YAML reassembled with SourceInfo::concat of per-line substrings and parsed via quarto_yaml::parse_with_parent so node spans map back through map_offset. Jupyter engine consumes it: error:true policy (un-annotated cell error aborts the render like knitr/Q1), option lines stripped from kernel input + echo (Q1 parity). Survey found 4 ad-hoc #| sites in q2 (crossref shorthand naive parser, LSP directive_tokens = best precedent, both engines opaque) and no shared facility. 7 open decisions flagged in the plan (scope eval?, global execute.error, error-location fidelity given text-in/text-out engine input, malformed-YAML severity, crossref migration timing). Awaiting plan review."}}}
 {"id":"bd-ojtq","title":"xtask create-worktree: auto-detect integration branch (or warn) instead of defaulting to main","description":"Today `cargo xtask create-worktree <bd-id>` defaults `--base main`. For sub-task work inside an active epic (the dominant use case for create-worktree per .claude/rules/worktrees.md), main is almost always the wrong base — the convention is to branch off `feature/<epic-name>`. Silent default leads to a worktree at the wrong base that must be `git reset --hard`ed after creation, which is what happened during bd-kw93.1 setup on 2026-05-14.\n\nOptions:\n1. Auto-detect: if the beads issue has a parent-child to an epic whose branch can be inferred (e.g. the epic title or labels record it), use that branch as the base.\n2. Warn-on-default: when `--base` is not passed and the issue has a parent epic, print 'Using default base `main`; sub-tasks of <epic-id> typically branch off feature/<name>. Re-run with --base if needed.' and prompt or exit.\n3. Make `--base` required when the issue is a sub-task (has parent-child to a non-closed epic).\n\nOption 2 is the lowest-friction nudge. Option 1 needs a way to recover the integration branch name from the epic (could be a beads label, an external_ref, or just a convention scan).\n\nRelated friction: `switch-task` already does the right thing for sequential sub-task work but is under-discoverable when you've reached for create-worktree first. Cross-reference in the rules file or in xtask help text might help.\n\nAcceptance: pick one option, ship, no longer silently defaults to main when context says otherwise.","status":"closed","priority":3,"issue_type":"task","created_at":"2026-05-14T14:40:07.105617Z","created_by":"cscheid","updated_at":"2026-05-14T20:03:15.368560Z","closed_at":"2026-05-14T20:03:15.368432Z","close_reason":"fetch_beads_metadata now reports parent_epic from the dependencies array; create-worktree warns when --base is at its default and the issue has an open parent epic. 5 new unit tests + 3-case binary smoke (warning fires; --base main silences; no parent = no warning). Workspace nextest 8864/8864. Merged to main as commit 28f1a18b.","dependencies":{"bd-kw93.1:discovered-from":{"depends_on_id":"bd-kw93.1","type":"discovered-from","created_at":"2026-05-14T14:40:07.105617Z","created_by":"cscheid"}}}
 {"id":"bd-okfxemnj","title":"pandoc.read from Lua leaks '[TOP-LEVEL MISSING NODE] Unhandled node kind: -' for pipe-table delimiter dashes","description":"GitHub issue: https://github.com/quarto-dev/q2/issues/847\n\nCalling pandoc.read(text, \"markdown\") from an {lmd} cell or Lua filter on text containing a pipe table prints one '[TOP-LEVEL MISSING NODE] Warning: Unhandled node kind: -' line per delimiter-row dash to stderr. Output is correct; the noise is unconditional, unformatted, and not routed through the diagnostic collector.\n\nTwo suspected causes: (1) the tree-sitter dispatch fallback in pampa/src/pandoc/treesitter.rs fires for the anonymous '-' children of pipe_table_delimiter_cell even though process_pipe_table_delimiter_cell handles them; (2) pandoc_read in pampa/src/lua/readwrite.rs passes std::io::stderr() as the reader's debug buffer while every pipeline caller passes a sink.","status":"in_progress","priority":2,"issue_type":"bug","created_at":"2026-10-15T18:49:53.250888Z","created_by":"cscheid","updated_at":"2026-10-16T13:41:16.623370Z","labels":["gh-847"],"comments":{"c-0jsflwk0":{"id":"c-0jsflwk0","author":"cscheid","created_at":"2026-10-16T13:23:51.541756Z","text":"Design decisions recorded in the plan (2026-10-16): land sink fix + explicit marker arms + reader-level and e2e tests. Follow-ups filed: bd-fwswbxnl (forward pandoc.read reader diagnostics), bd-yi613i1r (make catch-all test-visible). Awaiting go-ahead to implement."},"c-n43eygnh":{"id":"c-n43eygnh","author":"cscheid","created_at":"2026-10-15T18:59:26.784219Z","text":"Pre-flight cargo xtask verify --skip-hub-build at main@e308835bd: green (all 14 steps passed)."},"c-rmbe0ocl":{"id":"c-rmbe0ocl","author":"cscheid","created_at":"2026-10-16T13:41:16.623370Z","text":"Implementation complete; verify green. PR: https://github.com/quarto-dev/q2/pull/849 (Fixes #847). Strand stays in_progress until the PR merges."},"c-tlorcsi2":{"id":"c-tlorcsi2","author":"cscheid","created_at":"2026-10-15T18:55:39.533194Z","text":"Investigation plan: claude-notes/plans/2026-10-15-pandoc-read-delimiter-dash-noise.md (branch braid/bd-okfxemnj-pandoc-read-delimiter-dash-noise, topic branch in main checkout). Reproduced at main@e308835bd. Verdict: ready to design. Two confirmed causes: (1) lua/readwrite.rs pandoc_read passes std::io::stderr() as the reader's verbose trace buffer, the only production caller that does; (2) native_visitor has no arms for '-', pipe_table_align_left/right, '![' so they hit the catch-all before the parent consumes them by name. Proposed: sink the buffer (closes #847) + add explicit marker arms in the existing house pattern."}}}
+{"id":"bd-okqrjh8v","title":"Config paths Phase 1b: store marked Path values in project-root pivot form; document the Lua-visible change","description":"Follow-on to bd-nqq41zgx (plan claude-notes/plans/2026-10-28-config-path-late-binding.md, §4.6 and Phase 1b; decided with the user 2026-10-28). After Phase 1 (bd-lwvumpfk) routes every Path-kind consumer through resolve_config_path, switch the stored form written by mark_entry (project/format_paths.rs) and adjust_paths_recursive (project/mod.rs) from document-relative to the project-root pivot spelled with a leading / (absolute only for files outside the project), update the resolver's step 1, fix adjust_paths_recursive's http/https prefix list (use is_external_url) and anchor !path /x at the project root, regenerate goldens (smoke-all/metadata/dir-metadata-paths, hub-client pandoc-request goldens if affected). Removes the per-document rebase that caused the SCSS cache-key misses (bd-79c4do6g) and the .. chains (bd-s0jupmmv). DOCS TASK (accepted behaviour change): a Lua filter that reads a marked path key sees /x where Quarto 1 showed it the input-relative form; add a note to the Lua filter / extension docs with the recommended way to resolve it.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-10-28T14:41:44.344871Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-28T19:53:23.370490Z","closed_at":"2026-10-28T19:53:23.370490Z","close_reason":"Merged via PR #956 (62601b9ec). Marked Path values now stored in a single project-root pivot form on all platforms; Lua-visible change documented.","dependencies":{"bd-0x1znhgb:related":{"depends_on_id":"bd-0x1znhgb","type":"related","created_at":"2026-10-28T14:41:44.344871Z","created_by":"Carlos Scheidegger"},"bd-79c4do6g:related":{"depends_on_id":"bd-79c4do6g","type":"related","created_at":"2026-10-28T14:41:44.344871Z","created_by":"Carlos Scheidegger"},"bd-lwvumpfk:blocks":{"depends_on_id":"bd-lwvumpfk","type":"blocks","created_at":"2026-10-28T14:42:06.420509Z","created_by":"Carlos Scheidegger"},"bd-nqq41zgx:related":{"depends_on_id":"bd-nqq41zgx","type":"related","created_at":"2026-10-28T14:41:44.344871Z","created_by":"Carlos Scheidegger"}},"comments":{"c-h9md9axu":{"id":"c-h9md9axu","author":"Carlos Scheidegger","created_at":"2026-10-28T17:19:46.293619Z","text":"Phase 1b implemented: PR https://github.com/quarto-dev/q2/pull/956 (single commit on main after #955 merged). Marked Path values now hold the project-root pivot (no leading slash; absolute outside the project) via config_paths::pivot_form; adjust_paths_to_document_dir renamed adjust_paths_to_project_root; !path /x anchors at the project root; is_external_url replaces the prefix list; front-matter !path rewritten at the merge; ThemeContext and pampa citeproc built on the project root; FormatCssTransform relativizes entries in source-served mode (the WASM smoke-all run caught that); Lua docs note added; test lua_sees_marked_paths_in_project_root_form. Spelling deviation from plan §4.6 (no leading slash on marked values) recorded in the plan for review. Verify green with WASM rebuilt; clippy clean."},"c-xtaovpwk":{"id":"c-xtaovpwk","author":"Carlos Scheidegger","created_at":"2026-10-28T19:53:20.319955Z","text":"PR #956 merged into main as 62601b9ec (2026-10-28). Phase 1b lands: marked ConfigValueKind::Path values are stored in project-root pivot form (forward slashes, no leading slash, lexically normalized on every platform; absolute only outside the project), adjust_paths_to_project_root runs on every layer including document front matter, format-css relativizes entries in source-served mode, and docs/guides/authoring/lua-filters.qmd records the Lua-visible change. Review follow-up e390ede35 removed the cfg!(windows) spelling fork from pivot_form and its tests."}}}
 {"id":"bd-ol45i8oe","title":"Normalize EOL in revealjs vendored-asset drift test (Windows CRLF false-positive)","description":"vendored_reveal_assets_match_npm_package test fails on Windows: resources/revealjs/*.css checked out CRLF (core.autocrlf), byte-compared against LF node_modules/reveal.js copy in crates/quarto-core/src/revealjs/assemble.rs:576. Same class of fix as bd-mv2ggmr5 (pampa) and quarto-citeproc/.gitattributes: add repo-root .gitattributes entry pinning resources/revealjs/** (or the specific vendored files) to text eol=lf.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-07-08T09:37:32.193673Z","created_by":"cderv","updated_at":"2026-07-10T14:55:13.900226Z","closed_at":"2026-07-10T14:55:13.900226Z","external_ref":"https://github.com/quarto-dev/q2/pull/386","dependencies":{"bd-eehxwr29:parent-child":{"depends_on_id":"bd-eehxwr29","type":"parent-child","created_at":"2026-07-08T09:37:32.193673Z","created_by":"cderv"}},"comments":{"c-4eqy2aym":{"id":"c-4eqy2aym","author":"cderv","created_at":"2026-07-08T15:03:34.840990Z","text":"Reconsidered the approach; did NOT pin resources/revealjs to LF. Root distinction: the reveal assets are served VERBATIM (include_str -> artifact bytes -> file), so line endings survive to output; but a CRLF reveal.css is functionally identical to every browser, and nothing in q2 states bit-reproducible builds as a goal (the other ~9 embedded vendored dirs are all unpinned and would already violate it). The failing test's real job is content/version drift (re-sync when reveal.js bumps), and it was over-strict: byte-eq where it meant content-eq. Line endings were noise in a drift test.\n\nFix shipped: vendored_reveal_assets_match_npm_package now normalizes CRLF->LF on both sides before comparing (crates/quarto-core/src/revealjs/assemble.rs). Removes the false-positive at its source with zero per-file .gitattributes maintenance and no need for a coverage guardrail. Verified TDD on Windows: converted the 4 vendored files to CRLF (simulating a core.autocrlf checkout) -> test RED on EOL; added the normalize -> GREEN with the CRLF files still on disk; restored files to LF. Full revealjs module 108/108 green.\n\nCompanion finding (SCSS, the original worry that prompted the review): user-CRLF SCSS mixed with internal-LF SCSS is a non-issue. It is normalized twice already -- q2's parse_layer() strips CRLF (layer.rs:112, with a test at :407-427 and a comment matching this same 'SCSS keeps no byte-offset map' reasoning) and grass's lexer collapses CRLF/CR/form-feed to LF at the char level before parsing (grass_compiler 0.13.4 lexer.rs:128-146; empirically confirmed on windows-msvc: 8 CR in -> 0 CR out, expanded + compressed). No SCSS change needed."},"c-b25l7gtk":{"id":"c-b25l7gtk","author":"cderv","created_at":"2026-07-08T09:49:52.960370Z","text":"Root cause confirmed: REAL cross-platform build bug, not a test-only artifact.\n\nresources/revealjs/{reset.css,reveal.css,reveal.js,theme/white.css,quarto-reveal.css}\nare pulled in via include_str! at crates/quarto-core/src/revealjs/assemble.rs:29-35,\nembedded verbatim into the compiled binary as artifact content served by both\n`q2 render` and `q2 preview` (assemble.rs:817-824 explicitly asserts they are NOT\ninlined into HTML - they ship as separate served files, byte-for-byte from the\nconstant). Windows checkout with core.autocrlf=true (default) rewrites these\ncommitted-LF files to CRLF on disk; a Windows-built q2 binary then embeds\ndifferent bytes than a Linux/Mac-built binary for the same source tree. This\nbreaks build reproducibility across platforms, which is exactly the invariant\nthe failing test asserts (\"q2 render and q2 preview must use identical reveal\nassets\").\n\nWeb research (per codebase-research.md \"research ecosystem precedent\" rule):\ninclude_str! deliberately does NOT normalize line endings. rust-lang/rust#63681\nproposed \\r\\n -> \\n normalization in include_str! and was REJECTED by the lang\nteam - rationale: include_str! is used to embed non-Rust binary-sensitive\ncontent (Windows batch files needing literal \\r\\n, JSON/YAML, hash/signature\npayloads), so any implicit conversion is unacceptable breakage. Official\nposition: fix line-ending consistency at the git-checkout boundary\n(.gitattributes), not inside the macro or the consuming code.\n\nThis confirms our fix is the ecosystem-canonical answer, and matches existing\nrepo precedent for the exact same failure class:\n- crates/pampa/.gitattributes (test fixtures/snapshots)\n- crates/quarto-citeproc/.gitattributes (CSL conformance lockfile, same\n  byte-compare-against-generated-string shape as this bug)\n- crates/quarto-doctemplate/.gitattributes (template fixtures)\n\nFix applied (uncommitted, awaiting go-ahead per CLAUDE.md push policy):\n- Added to root .gitattributes: `resources/revealjs/** text eol=lf`\n  (pins the whole vendored reveal.js asset dir - covers all 5 files consumed\n  by include_str!, plus reset-scoped.css/LICENSE/README for consistency,\n  though those aren't currently embedded).\n- `git add .gitattributes && git add --renormalize resources/revealjs`\n  staged with ZERO content diff vs HEAD for the resources/ files - confirms\n  the committed blobs were already LF (npm-sourced); only the Windows\n  checkout conversion was ever wrong. Nothing to fix upstream in the blobs\n  themselves.\n- `git restore --worktree` did NOT rewrite the already-present CRLF files\n  (git skipped write, presumably because content hash matched cached stat).\n  Had to `rm` the 5 files (including theme/white.css, easy to miss on first\n  pass) + `git checkout --` to force a fresh smudge-filtered checkout to LF.\n- Verified: `cargo nextest run -p quarto-core vendored_reveal_assets_match_npm_package`\n  passes. Full `revealjs::` module (62 tests) passes.\n\nState at handoff:\n- Only `.gitattributes` is staged (+7 lines). resources/revealjs/* show\n  no git diff (expected - see above). NOT committed, NOT pushed.\n- Still needed before commit per CLAUDE.md: cargo build --workspace,\n  cargo nextest run --workspace, cargo xtask verify (at least\n  --skip-hub-build, since quarto-core is on the WASM dependency path -\n  wasm-quarto-hub-client also include_str!s nothing extra here but consumes\n  quarto-core, so the hub-build leg is worth including if time allows).\n- No other test/snapshot found referencing these constants' literal bytes\n  (grepped for .snap files under revealjs - none; hub-client only\n  string-references \"resources/revealjs\" in comments, no embed).\n- This strand sits under the line-ending epic (bd-eehxwr29) as a\n  parent-child, but is a DISTINCT mechanism from that epic's document-\n  processing-engine CRLF work (qmd source preservation) - this is a\n  vendored-build-asset reproducibility bug, fixed purely at the git\n  attributes layer, no q2 code changes."}}}
 {"id":"bd-olz91r4v","title":"Lua parity G: __toinline/__toblock metamethods","description":"Coercion should consult __toinline/__toblock metamethods per pandoc-lua-marshal semantics (incl. ignored-if-not-function, non-Inline-return-ignored). Catalog: claude-notes/research/2026-07-13-lua-api-mismatch-catalog.md. Epic plan: claude-notes/plans/2026-07-13-lua-api-pandoc-parity.md.","status":"closed","priority":3,"issue_type":"feature","created_at":"2026-07-13T17:09:27.592263Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-13T23:15:00.583783Z","closed_at":"2026-07-13T23:15:00.583783Z","close_reason":"__toinline/__toblock hooks in all four fuzzy peekers with hslua recoverable-failure semantics (ignored-if-not-function, non-element-return-ignored pinned). Track-1 25->21 xfail; differential toinline-toblock-hooks; e2e identical to pandoc 3.9.0.2. Same commit as bd-0g2yp61w (shared types.rs edits).","dependencies":{"bd-grkrb9nj:parent-child":{"depends_on_id":"bd-grkrb9nj","type":"parent-child","created_at":"2026-07-13T17:09:27.592263Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-omrnwgeh","title":"Sync wasm-quarto-hub-client/Cargo.lock to workspace version","description":"The excluded WASM crate has its own Cargo.lock that 'cargo update --workspace' doesn't touch; it has lagged at 0.1.0 (through 0.1.1 and now 0.2.0). build-wasm.js builds without --locked so CI regenerates it and the release is unaffected, but the committed file is cosmetically stale. Bump it in a future PR. Discovered cutting v0.2.0.","status":"closed","priority":3,"issue_type":"chore","created_at":"2026-06-16T13:49:11.284942Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-29T16:51:44.168+00:00","closed_at":"2026-07-29T16:51:44.168+00:00","dependencies":{"bd-56hbab5q:discovered-from":{"depends_on_id":"bd-56hbab5q","type":"discovered-from","created_at":"2026-06-16T13:49:11.284942Z","created_by":"Carlos Scheidegger"}}}
@@ -1457,6 +1471,7 @@
 {"id":"bd-ppu6xmvv","title":"exit-drain test times out on main since automerge-repo 2.6.0-alpha.5 (TS Test Suite red)","description":"TS Test Suite has failed on every push to main since PR #685 (automerge npm upgrade, bd-d08gpqvu) merged: runs 35245424175 (macos leg) and 35249754405 (ubuntu leg). Failing test: ts-packages/quarto-hub-mcp/src/exit-drain.test.ts > 'create_project then immediate stdin EOF must not lose the created docs' — vitest per-test timeout (30 s). PR #685's own CI passed at 19-26 s; on main the test takes 36-39 s. Before the upgrade it took 10-13 s on CI.\n\nDiagnosis (same machine, same probe, dependency trees swapped; drain-to-exit ms per round, 64 x 64 KB files):\n- automerge-repo 2.5.6 hub + 2.5.6 client: 1.7 / 2.0 / 4.6 s\n- 2.6 hub + 2.5.6 client: 9.8 / 8.0 / 9.5 s\n- 2.5.6 hub + 2.6 client: 7.4 / 22.8 / 26.2 s\n- 2.6 hub + 2.6 client: 10.3 / 20.9 / 33.1 s\nChecking the hub's repo directly shows the 64 docs only finish arriving ~13 s after create_project returns on the 2.6/2.6 tree; once they have landed the drain exits in <1 s. So this is raw sync throughput between two automerge-repo 2.6 JS peers, not a drain-signal bug. Zeroing saveDebounceRate on the test hub does not help.\n\nProduction relevance: against the real Rust hub (samod, target/debug/hub) the drain takes ~5.0 s with both the 2.5.6 and the 2.6 client — no regression observed. The slow path needs a JS automerge-repo peer as the *server*, which only the in-process startTestHub() stand-in is.\n\nOptions: (a) raise the per-test timeout for this one test (test already sets QUARTO_MCP_SHUTDOWN_DRAIN_MS=30000 and waits up to 35 s for exit, both longer than the 30 s vitest timeout, so the hang-detector bound can never fire today — fix that inconsistency at the same time); (b) chase the throughput regression upstream in automerge-repo 2.6 (DocSynchronizer rewrite) and report it; (c) both. Not a candidate: shrinking the payload (see the measured floor in the test's header comment).","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-09-17T17:41:49.952440Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T18:31:46.369577Z","closed_at":"2026-09-17T18:31:46.369577Z","close_reason":"Fixed by PR #691 (merged): exit-drain test budgets sized to automerge-repo 2.6 JS-to-JS sync throughput","labels":["ci","flaky"],"dependencies":{"bd-10deu8h4:related":{"depends_on_id":"bd-10deu8h4","type":"related","created_at":"2026-09-17T17:41:49.952440Z","created_by":"Carlos Scheidegger"},"bd-d08gpqvu:discovered-from":{"depends_on_id":"bd-d08gpqvu","type":"discovered-from","created_at":"2026-09-17T17:41:49.952440Z","created_by":"Carlos Scheidegger"},"bd-yw3mcdkg:related":{"depends_on_id":"bd-yw3mcdkg","type":"related","created_at":"2026-09-17T17:41:49.952440Z","created_by":"Carlos Scheidegger"}},"comments":{"c-j044d610":{"id":"c-j044d610","author":"Carlos Scheidegger","created_at":"2026-09-17T17:59:35.479399Z","text":"Fix PR: https://github.com/quarto-dev/q2/pull/691 (branch bugfix/bd-ppu6xmvv-exit-drain-timeout, commit a06eb1c4). Raises DRAIN_BUDGET_MS to 60 s and derives EXIT_BOUND_MS / TEST_TIMEOUT_MS from it. Upstream throughput regression in automerge-repo 2.6 JS-to-JS sync remains open here."}}}
 {"id":"bd-pq72bplh","title":"Q-5-12 render-scripts warning wrongly shown in q2 preview","description":"The one-time Q-5-12 warning ('Project render scripts do not run in the hub preview') is pushed by render_project_active_page_to_response in crates/wasm-quarto-hub-client/src/lib.rs for ANY project that configures project.pre-render / project.post-render scripts. That shared function serves both the hub-client preview (render_page_in_project*) and the q2 preview SPA (render_page_for_preview). For q2 preview the claim is false: the native host runs pre-render scripts once at boot (run_boot_pre_render_scripts in crates/quarto-preview/src/lib.rs, plan D7 of 2026-07-29-pre-post-render-scripts.md). Suppress the warning for the q2 preview host. Plan: claude-notes/plans/2026-08-08-q5-12-suppress-in-q2-preview.md","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-08T15:08:04.058128Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-08T17:13:17.570664Z","closed_at":"2026-08-08T17:13:17.570664Z","close_reason":"Fixed in PR #472 (merged): RenderHost enum threaded through the WASM project-render path; Q-5-12 now emits only for the hub host. q2 preview no longer shows the false warning (verified end-to-end). Plan: claude-notes/plans/2026-08-08-q5-12-suppress-in-q2-preview.md","dependencies":{"bd-w348iu63:caused-by":{"depends_on_id":"bd-w348iu63","type":"caused-by","created_at":"2026-08-08T15:08:10.369142Z","created_by":"Carlos Scheidegger"}},"comments":{"c-bcxqoh5g":{"id":"c-bcxqoh5g","author":"Carlos Scheidegger","created_at":"2026-08-08T15:09:06.367934Z","text":"Reproduced 2026-08-08 with a scratch website project (pre-render: pre.sh) under 'q2 preview --port 7799 --no-browser': terminal printed 'Running pre-render script: pre.sh', pre-ran.txt was created, and the browser overlay still showed the Q-5-12 warning. Assessment + fix plan (Option A: thread a RenderHost enum through render_project_active_page_to_response and gate the warning on HubClient) written to claude-notes/plans/2026-08-08-q5-12-suppress-in-q2-preview.md. Awaiting Carlos's review before implementation."},"c-m2240t56":{"id":"c-m2240t56","author":"Carlos Scheidegger","created_at":"2026-08-08T15:32:13.987670Z","text":"Fix implemented and PR opened: https://github.com/quarto-dev/q2/pull/472 (branch bugfix/bd-pq72bplh-q5-12-preview-warning, commit 87393d27). Awaiting CI."}}}
 {"id":"bd-pq9k90z2","title":"Epic: native docx writer (clean-room OOXML emitter + redoc-style .qmd/source-map sidecars)","description":"Native, pandoc-free docx output for Quarto 2: Pandoc AST -> OOXML written directly by pampa, with Word-specific XML knowledge sourced from python-docx (MIT) and ECMA-376 (not pandoc, GPL). Embeds the source .qmd and source map as OPC parts related from document.xml (the redoc mechanism, verified to survive a Word round trip) so docx comments can be mapped back to the .qmd. Requires an OMML math writer, which comes from the quarto-math crate (child strand). This is a parallel experiment to the pandoc-hybrid leg (Lua filters + Pandoc for docx/pptx/typst); both proceed until we decide which is easier/cleaner/faster to land. Plan: claude-notes/plans/2026-09-21-quarto-math-and-native-docx.md","status":"open","priority":2,"issue_type":"epic","created_at":"2026-09-21T15:10:55.433291Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-21T15:13:46.501067Z","comments":{"c-lazjhfel":{"id":"c-lazjhfel","author":"Carlos Scheidegger","created_at":"2026-09-21T15:13:46.501067Z","text":"Plan: claude-notes/plans/2026-09-21-quarto-math-and-native-docx.md. Phases 2-4 (OPC layer + validation harness, docx writer, sidecars/comment round trip) get their own child strands when Phase 1 (bd-entbg6x3) lands."}}}
+{"id":"bd-pqj37chw","title":"Lua API: Quarto 1 parity gaps from the Connect docs port (GH #939-#942)","description":"Umbrella for four Lua API gaps reported by atheriel while porting the Posit Connect docs from Quarto 1.9.32 to q2 0.36.0: quarto.project + QUARTO_PROFILE via os.getenv (#939), quarto.doc.input_file/output_file (#940), os.exit killing the whole process (#941), quarto.version as a plain {0,1,0} table with no pandoc.types.Version (#942). All reproduced on main at 3f7d4b0e8 on 2026-10-27.\n\nPlan: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md\n\n#939 and #940 share new plumbing (a per-document HostContext in pampa's LuaOptions, filled by quarto-core's UserFiltersStage and ShortcodeResolveTransform). Order: #942 -> #939 (plumbing) -> #940 -> #941, on one integration branch feature/lua-api-q1-parity.","status":"open","priority":2,"issue_type":"epic","created_at":"2026-10-27T20:21:54.201030Z","created_by":"claude","updated_at":"2026-10-27T23:44:28.480908Z","external_ref":"https://github.com/quarto-dev/q2/pull/949","labels":["lua","pampa","parity"],"comments":{"c-3ci9ftxp":{"id":"c-3ci9ftxp","author":"claude","created_at":"2026-10-27T21:54:32.675578Z","text":"PR: https://github.com/quarto-dev/q2/pull/949 (feature/lua-api-q1-parity -> main; closes GH #939-#942 on merge). Full cargo xtask verify green incl. hub-client tests on a rebuilt WASM; Windows cross-check not possible from macOS, flagged in the PR."},"c-hj2pch21":{"id":"c-hj2pch21","author":"claude","created_at":"2026-10-27T22:08:30.987734Z","text":"Follow-up bd-irkfpip1 (env shortcode on the snapshot) is up as a stacked PR: https://github.com/quarto-dev/q2/pull/950"},"c-pyv9gpxz":{"id":"c-pyv9gpxz","author":"claude","created_at":"2026-10-27T21:37:04.697692Z","text":"All four children implemented on feature/lua-api-q1-parity (local branch, not yet pushed): 942 -> 939 -> 940 -> 941 as one --no-ff merge each, plus docs (docs/guides/authoring/lua-filters.qmd: 'Where the filter is running', 'Checking versions', 'Failing a render') and the plan's implementation notes. Running cargo xtask verify --skip-hub-build and the WASM Lua tests next; then push + PR. Windows cross-check is blocked from macOS by C-building deps (aws-lc-sys, wasmtime, tree-sitter); flagging for a Windows run in the PR."},"c-uo39ujt7":{"id":"c-uo39ujt7","author":"claude","created_at":"2026-10-27T23:44:28.480908Z","text":"PR #949 merged into main after CI came back clean (the only CI failure was the Q-2-7 apostrophe in docs/guides/authoring/lua-filters.qmd, fixed in 7c9ae921a). Closes GH #939-#942. Follow-up PR #950 (bd-irkfpip1) now targets main."},"c-uyukyqg0":{"id":"c-uyukyqg0","author":"claude","created_at":"2026-10-27T21:04:10.936118Z","text":"Review round 1 (Carlos, 2026-10-27), all three open decisions resolved and folded into the plan: (1) os.getenv reads from an environment snapshot the host builds on the orchestrator thread before any filter runs (real env, then _environment file values for missing keys, then QUARTO_PROFILE forced to the active list) -- avoids libc getenv/setenv races from filter worker threads; (2) minimal PANDOC_STATE {input_files, output_file} is in scope under bd-diu7rtes; (3) one PR at the end on feature/lua-api-q1-parity. Still awaiting the go-ahead to execute."},"c-ya4rg9xo":{"id":"c-ya4rg9xo","author":"claude","created_at":"2026-10-27T20:23:04.244769Z","text":"Plan written: claude-notes/plans/2026-10-27-lua-api-q1-parity-gh-939-942.md (uncommitted, on main, awaiting review). All four issues reproduced on main at 3f7d4b0e8 (2026-10-27). Children: bd-5qepzwst (#942, do first), bd-ip1lrgra (#939, carries the HostContext plumbing), bd-diu7rtes (#940, blocked on bd-ip1lrgra), bd-9u10sb4b (#941). Open decisions for review: (1) os.getenv also falling back to _environment file values (small extension past #939); (2) whether to add a minimal PANDOC_STATE now or as a follow-up; (3) one PR per sub-task vs one PR at the end."}}}
 {"id":"bd-pqmjwtdn","title":"FTS Phase 1.1: choose search library (spike)","description":"Spike comparing MiniSearch / FlexSearch / Orama for hub-client full-text search: bundle size, relevance (TF + prefix/fuzzy), tokenizer flexibility. Record the decision in the design doc. Blocks the provider implementation. Plan: claude-notes/plans/2026-06-23-hub-client-full-text-search.md","status":"closed","priority":2,"issue_type":"task","created_at":"2026-06-23T18:23:10.134272Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-23T18:25:02.585388Z","closed_at":"2026-06-23T18:25:02.585388Z","close_reason":"MiniSearch v7.2.0 chosen and added to hub-client; rationale recorded in plan.","dependencies":{"bd-mdet0yoy:parent-child":{"depends_on_id":"bd-mdet0yoy","type":"parent-child","created_at":"2026-06-23T18:23:10.134272Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-prfbxrth","title":"Move hub-client/vscode-sync-experiment out from under hub-client/ (nested npm workspace)","description":"hub-client/vscode-sync-experiment (quarto-hub-vscode, added 909eada30) is registered as a root npm workspace nested inside the hub-client workspace. npm resolves `-w hub-client` to every workspace at/under that path, so any `-w hub-client` invocation fans out to it: lint:css failed (fixed in bd-verify-css-lint-nested-ws-r0dd5vdk via --prefix), and hub-client-e2e.yml:189 `npm run build --workspace hub-client` also esbuilds the extension. The nested-npm-workspace xtask lint allowlists this pair pending this strand. Fix: move the experiment (e.g. ts-packages/ or a top-level experiments/), update root package.json workspaces, drop the ALLOWED entry in crates/xtask/src/lint/nested_npm_workspaces.rs. Coordinate with Elliot (owner) — to be raised on the PR for the parent strand.","status":"open","priority":2,"issue_type":"task","created_at":"2026-10-08T17:51:00.643409Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-08T17:51:00.643409Z","dependencies":{"bd-verify-css-lint-nested-ws-r0dd5vdk:discovered-from":{"depends_on_id":"bd-verify-css-lint-nested-ws-r0dd5vdk","type":"discovered-from","created_at":"2026-10-08T17:51:00.643409Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-proxy-econnreset-sxelg87m","title":"local-prod-server.mjs proxy crashes on unhandled ECONNRESET in WebSocket upgrade, failing author-attribution E2E","description":"scripts/local-prod-server.mjs handleUpgrade (lines ~94-101) joins the client socket and proxySocket with proxySocket.pipe(socket).pipe(proxySocket) but attaches no 'error' handler to either. When a proxied WebSocket is reset (e.g. a tab closes abruptly), Node throws 'Unhandled error event ... read ECONNRESET' and the whole proxy process exits.\n\nObserved in CI (Hub-Client E2E Tests, run 36887990034, PR #770): after test 2 of hub-client/e2e-author/author-attribution.spec.ts (the two-tab H4 scenario) the log shows '[proxy:3999] node:events ... throw er; // Unhandled error event ... Error: read ECONNRESET'. Test 4 (legacy doc, auth-on) then fails in ~2ms on every retry with 'fetch failed ... ECONNREFUSED 127.0.0.1:3999' because the auth-ON proxy is gone. Tests 1-3 pass. The failure is intermittent in nature (timing race on socket close) but reproduced on every rerun for that PR; other branches and main pass. #770 itself is Rust-only and unrelated.\n\nExpected: a reset on one proxied connection tears down only that connection pair; the proxy keeps serving. Likely fix: socket.on('error', () => proxySocket.destroy()) and proxySocket.on('error', () => socket.destroy()) in handleUpgrade, plus checking the plain-HTTP path for the same gap. Add a regression check if practical (e.g. abruptly reset an upgraded connection and confirm the proxy still answers).","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-01T21:01:18.847948Z","created_by":"Gordon Woodhull","updated_at":"2026-10-01T21:01:30.870756Z","comments":{"c-ylzzy63i":{"id":"c-ylzzy63i","author":"Gordon Woodhull","created_at":"2026-10-01T21:01:30.870756Z","text":"Dispatched to agent local-prod-server on workspace-1 (branch braid/bd-proxy-econnreset-sxelg87m-proxy-econnreset)"}}}
@@ -1660,7 +1675,7 @@
 {"id":"bd-t0av3ge8","title":"smoke-all runner doesn't clean up _site/ for type: website fixtures","description":"The crates/quarto/tests/smoke-all runner cleans up only the output file and <stem>_files/ after each fixture run. It has no cleanup path for type: website project fixtures, which render into a _site/ directory instead.\n\nbd-repo-actions-missing-99ezd2fe added the first three type: website smoke-all fixtures in the repo (the only three type: website _quarto.yml files anywhere under smoke-all/). Each run leaves a ~708 KB _site/site_libs/ tree behind in the source tree. The branch worked around this correctly for git with a .gitignore glob (crates/quarto/tests/smoke-all/**/_site/, with a comment explaining why), confirmed to cover all three fixture depths and leave the working tree clean — but that's a workaround, not a fix.\n\nFix: extend the smoke-all runner's cleanup step to also remove _site/ (or more generally, whatever output directory a type: website fixture's _quarto.yml resolves to) after each run, the same way it already removes the single-file output and its _files/ sibling.\n\nFiled per the branch's final review (M7): \"gitignoring is a workaround for a runner that does not clean up after project-type fixtures, and this branch is the first to need it. Worth a strand against the smoke-all runner, not against this branch.\"","status":"open","priority":3,"issue_type":"task","created_at":"2026-08-25T01:48:23.219663Z","created_by":"Gordon Woodhull","updated_at":"2026-08-25T01:48:23.219663Z","dependencies":{"bd-repo-actions-missing-99ezd2fe:discovered-from":{"depends_on_id":"bd-repo-actions-missing-99ezd2fe","type":"discovered-from","created_at":"2026-08-25T01:48:23.219663Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-t0qt409i","title":"Document caption_long canonicalization (Plain, not Para) as an intentional Q1 divergence for Lua filter authors","description":"Follow-up to bd-n3sark9b Phase 2 (approved 2026-09-17). Q2 canonicalizes FloatRefTarget.caption_long to a single Plain block at the crossref sugar boundary for every float form — div-form trailing paragraph, ![cap](img){#fig-x}, : cap {#tbl-x}, and #| fig-cap cells. Quarto 1 does NOT canonicalize: its filters see the div-form caption as a Para (refCaptionFromDiv returns the trailing Para verbatim) and the other forms as Plain. Rendered HTML converges on Q1 (bare inlines inside <figcaption>, no <p>), but the filter-visible IR diverges: a Lua filter that matches Para inside a float caption (e.g. walking float.caption_long, or a Para handler that expected to see div-form captions) will see Plain in q2. Document this for advanced users: (1) in docs/guides/authoring/lua-filters.qmd, alongside the existing 'Quarto 1 accepts…' notes — state the caption_long contract ([Plain[...]]), why (Pandoc's own Figure/Table caption convention; uniform figcaption markup), and the porting recipe (match Plain or Para, or use the inlines of caption_long[1]); (2) in the cross-references authoring guide once it exists — note docs/guides/authoring/cross-references.cmd is currently a 'TBD' placeholder with a .cmd extension. Blocked on bd-n3sark9b so the docs describe shipped behavior; source material: claude-notes/plans/2026-09-17-float-caption-prefix-plain.md (Phase 2 + design decision).","status":"open","priority":2,"issue_type":"docs","created_at":"2026-09-17T17:11:12.632382Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T17:11:29.642800Z","labels":["crossref","docs"],"dependencies":{"bd-n3sark9b:blocks":{"depends_on_id":"bd-n3sark9b","type":"blocks","created_at":"2026-09-17T17:11:25.761299Z","created_by":"Carlos Scheidegger"},"bd-n3sark9b:discovered-from":{"depends_on_id":"bd-n3sark9b","type":"discovered-from","created_at":"2026-09-17T17:11:29.642791Z","created_by":"Carlos Scheidegger"},"bd-tr81:parent-child":{"depends_on_id":"bd-tr81","type":"parent-child","created_at":"2026-09-17T17:11:20.187329Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-t0wpbq59","title":"Jupyter execution environment: honest detection, dev machines, CI","description":"Groups the work uncovered by the Windows jupyter_kernel_cleanup_e2e failure (2026-09-28). Plan: claude-notes/plans/2026-09-28-jupyter-windows-detection-and-env.md (shared, project-scope only; per-machine dev setup lives outside the repo). NOW (directly workable): bd-1eu34vpy is_available probe, bd-ce4fftg8 + bd-zvf3e8n2 test-guard hardening. LATER (needs design/team discussion): bd-elhgm6ff q2 check jupyter (blocked by bd-4qflzhwh coordination), bd-z09fx92g CI python provisioning (blocked by bd-ek7njozm strategy decision).","status":"open","priority":1,"issue_type":"epic","created_at":"2026-09-28T10:02:32.645018Z","created_by":"cderv","updated_at":"2026-09-28T10:04:38.609160Z","labels":["jupyter"]}
-{"id":"bd-t18384ia","title":"hub-client useEffect refactor: Phase 5 useEffectEvent sweep","description":"Replace ~12 ref-mirror effects with useEffectEvent where behavior-preserving. Plan: claude-notes/plans/2026-10-24-hub-client-useeffect-refactor.md (Phase 5). Stacks on PR #936.","status":"in_progress","priority":3,"issue_type":"task","created_at":"2026-10-27T16:01:06.044411Z","created_by":"shikokuchuo","updated_at":"2026-10-27T16:16:12.935206Z","comments":{"c-k9m1izxs":{"id":"c-k9m1izxs","author":"shikokuchuo","created_at":"2026-10-27T16:16:12.935206Z","text":"Triaged all 12 sites; converted 3 (useAuthProbe, useSessionKeepAlive, useWindowFileDrop). 9 sites fail the useEffectEvent contract (v7 rules-of-hooks enforces lexical effect-scope references only; render-phase calls throw): useRouting (setState updater), useDownloadAs (click handler), Preview/ReactPreview previewStateRef (render-scope useCallback), useSlideThumbnails (recursive useCallback), Editor currentFilePathRef (passed down to hooks), Editor getSlideForLineRef + editorDndHandlersRef (Monaco onMount callback). Characterization tests added pre-conversion; test:ci + build:all green; eslint clean."}}}
+{"id":"bd-t18384ia","title":"hub-client useEffect refactor: Phase 5 useEffectEvent sweep","description":"Replace ~12 ref-mirror effects with useEffectEvent where behavior-preserving. Plan: claude-notes/plans/2026-10-24-hub-client-useeffect-refactor.md (Phase 5). Stacks on PR #936.","status":"closed","priority":3,"issue_type":"task","created_at":"2026-10-27T16:01:06.044411Z","created_by":"shikokuchuo","updated_at":"2026-10-27T16:18:49.574370Z","closed_at":"2026-10-27T16:18:49.574370Z","close_reason":"Done: 3 of 12 sites converted to useEffectEvent (useAuthProbe, useSessionKeepAlive, useWindowFileDrop); 9 documented as failing the effect-event contract. test:ci + build:all green.","comments":{"c-k9m1izxs":{"id":"c-k9m1izxs","author":"shikokuchuo","created_at":"2026-10-27T16:16:12.935206Z","text":"Triaged all 12 sites; converted 3 (useAuthProbe, useSessionKeepAlive, useWindowFileDrop). 9 sites fail the useEffectEvent contract (v7 rules-of-hooks enforces lexical effect-scope references only; render-phase calls throw): useRouting (setState updater), useDownloadAs (click handler), Preview/ReactPreview previewStateRef (render-scope useCallback), useSlideThumbnails (recursive useCallback), Editor currentFilePathRef (passed down to hooks), Editor getSlideForLineRef + editorDndHandlersRef (Monaco onMount callback). Characterization tests added pre-conversion; test:ci + build:all green; eslint clean."}}}
 {"id":"bd-t18wrsqo","title":"Add autonomously-fix-small-issue skill","description":"Codify the GH #930 / PR #934 workflow (issue -> strand -> plan -> branch -> fix + tests -> PR) as a user-invoked-only skill with a scope gate that bails to /triage when the issue is not small and unambiguous. Plan: claude-notes/plans/2026-10-27-autonomously-fix-small-issue-skill.md","status":"in_progress","priority":3,"issue_type":"chore","created_at":"2026-10-27T15:50:05.355383Z","created_by":"t","updated_at":"2026-10-27T15:51:02.759542Z","labels":["skills"],"comments":{"c-u04efzlv":{"id":"c-u04efzlv","author":"t","created_at":"2026-10-27T15:51:02.759542Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/935"},"c-wwb6bcgf":{"id":"c-wwb6bcgf","author":"t","created_at":"2026-10-27T15:50:58.412535Z","text":"PR opened: see feature/bd-t18wrsqo-autonomous-small-fix-skill"}}}
 {"id":"bd-t1avfz0n","title":"q2-preview: comment chrome for block-level ::: >> comments","description":"Follow-up to bd-an9gkxnp. Block comments (::: >> {author=… date=…}) reach the AST as Div.quarto-edit-comment. The q2-preview CommentBlock chrome (ts-packages/preview-renderer/src/q2-preview/custom/CommentBlock.tsx) assumes comments are inline [>> ] spans inside a host block, plus the .quarto-edit-comment-container Div used for code blocks. Decide how a block comment is shown (its own bubble? attached to the neighboring block?), how resolve/add work for it, and make sure it at least renders without breaking the chrome. See claude-notes/plans/2026-09-24-block-editorial-marks.md.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-09-24T15:25:54.281943Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-24T15:25:54.281943Z","labels":["hub-client","preview-renderer"],"dependencies":{"bd-an9gkxnp:blocks":{"depends_on_id":"bd-an9gkxnp","type":"blocks","created_at":"2026-09-24T15:25:54.281943Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-t2t2zspe","title":"Document inline executable expressions in q2's docs/","description":"q2 now implements both Quarto inline expression spellings with their documented, deliberately-different semantics (bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe): `{r} expr` escapes markdown specials in the value; knitr's native `r expr` inserts the value as live markdown, equivalent to `{r} I(expr)`.\n\nNothing in q2's own docs/ says so. The strand's provenance search found inline expressions are undocumented there — only Q-2-47 and Q-2-50 mention {r} at all — so a q2 user has to read quarto.org (which describes Q1) or the source to learn which spelling to reach for.\n\nWorth covering:\n- the two spellings and the escaping split, with the I() equivalence\n- that a value inserted via the classic spelling is live markdown, and therefore live HTML: untrusted input should use the brace spelling\n- how to *write about* the syntax without executing it (doubled brace, or an extra enclosing backtick). This is a live trap: a knitr document displaying an unresolvable brace expression inside a fence now fails the render, matching Q1.\n- the one q2/Q1 divergence found while fixing this: an empty brace expression `{r}   ` in prose renders as a literal code span in q2 rather than erroring, because q2's pass runs over the serialized AST (which normalizes the trailing whitespace away) where Q1's runs over raw source. It errors in both in an attribute value.\n\nContract and rationale: crates/quarto-core/src/engine/knitr/preprocess.rs module docs.\n\nNote bd-u996g8g2 is the jupyter half of the spelling gap ({python} expr); if that lands, this page should cover both engines rather than being R-specific.","status":"closed","priority":2,"issue_type":"docs","created_at":"2026-09-04T16:23:06.390915Z","created_by":"Gordon Woodhull","updated_at":"2026-09-04T16:33:33.446543Z","closed_at":"2026-09-04T16:33:33.446543Z","close_reason":"Done in braid/bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe: docs/guides/authoring/computations.qmd gained an 'Inline expressions' section covering the two spellings, the escaping split and the I() equivalence, where they work, and how to write about the syntax. The jupyter half stays with bd-u996g8g2; the display-block escape gap is bd-0gwekaem.","dependencies":{"bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe:discovered-from":{"depends_on_id":"bd-inline-r-brace-spelling-not-evaluated-lk9s3iwe","type":"discovered-from","created_at":"2026-09-04T16:23:06.390915Z","created_by":"Gordon Woodhull"}}}
@@ -1721,6 +1736,7 @@
 {"id":"bd-tux4m6od","title":"Improve project import error UX: generic 'Document ... is unavailable' when opening a shared project collection","description":"Sharing a project collection on quarto-hub.com from one browser to another fails with the generic message 'Document 2Agx7kENjysHSujsVgirvykVKECf is unavailable'. The message gives users no indication of the cause (auth? sync server? missing doc? permissions?) or a next step. Investigate the failure in the production deployment, match against hub-client source, and design specific, actionable error messages for the project-import path.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-07-29T14:57:38.338605Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-15T11:05:59.224185Z","closed_at":"2026-10-15T11:05:59.224185Z","close_reason":"Landed as PR #439 (merged 2026-07-30): join-collection failures now classified (CollectionConnectError: auth-expired | offline | sync-unreachable | not-found | unknown) instead of surfacing automerge-repo's raw 'Document ... is unavailable'; bounded find + wait-for-peer fixes the 1s forceReady race, /auth/me probe disambiguates 401.","comments":{"c-4pkk7gb6":{"id":"c-4pkk7gb6","author":"Carlos Scheidegger","created_at":"2026-07-29T14:58:01.837179Z","text":"Plan: claude-notes/plans/2026-07-29-project-import-error-ux.md — starting live diagnosis in production via DevTools-controlled Chrome."},"c-bmm9jvhu":{"id":"c-bmm9jvhu","author":"Carlos Scheidegger","created_at":"2026-07-29T15:08:01.894082Z","text":"Phase 1 diagnosis complete (live against production). Error string is automerge-repo's raw Repo.find() rejection, surfaced verbatim by JoinCollectionLanding.tsx:74 via projectSetService.connectCollection (find at projectSetService.ts:237). At least 3 failure modes converge on the one message: (1) server lacks the doc — reproduced live with a random valid doc id on a healthy authed connection; (2) ws upgrade 401 (expired quarto_hub_token cookie, server.rs ws_handler) + adapter forceReady-after-1s => find() with zero peers => unavailable; (3) transient cold-ws race, same 1s forceReady — matches observed flakiness (user's click failed, retry seconds later succeeded). Custom error branches in connectCollection are dead code for the no-cache path. Full notes: claude-notes/plans/2026-07-29-project-import-error-ux.md"},"c-dq536efl":{"id":"c-dq536efl","author":"Carlos Scheidegger","created_at":"2026-07-30T17:48:59.529019Z","text":"Phase 3 implemented (TDD). New CollectionConnectError (auth-expired | offline | sync-unreachable | not-found | unknown) thrown by projectSetService.connectCollection; findCollectionDoc mirrors quarto-sync-client's findDoc pattern (bounded find, wait-for-peer fixes the 1s forceReady race, bounded re-request window, /auth/me probe to disambiguate 401-ws from network-down, VITE_GOOGLE_CLIENT_ID gates the auth-expired verdict). JoinCollectionLanding shows classified copy + 'Sign in again' action (reload; pre-auth-hash restore returns to the join route). 23 new tests green, full test:ci green, build:all green. E2E-verified in local-prod: phantom doc id now yields the actionable not-found message (screenshot inspected). Gotcha: stale PWA service worker initially served the old bundle during verification."},"c-fbmbkkf9":{"id":"c-fbmbkkf9","author":"Carlos Scheidegger","created_at":"2026-07-30T18:29:21.626548Z","text":"CI: both suites GREEN on the PR branch via workflow_dispatch (TS Test Suite run 30569132769, Test Suite run 30569134654, both on commit 65f52199 = PR head). GitHub webhook event delivery for the repo has been stalled since ~17:57Z (pull_request open/reopen/close/synchronize events all unprocessed; githubstatus claims operational), so the PR checks box only shows Snyk for now — the dispatched runs cover the same commit. Empty commit a6340044 pushed to fire synchronize once the backlog clears; watcher armed."},"c-jdv97tvk":{"id":"c-jdv97tvk","author":"Carlos Scheidegger","created_at":"2026-07-30T19:07:43.927623Z","text":"PR #439 fully green: after the GitHub webhook backlog cleared and Carlos merged main into the branch (abaafa6b, includes Comments v1 #441), the pull_request-triggered Test Suite, TS Test Suite, and E2E — interactive all passed, alongside both Snyk checks. Ready for review/merge."},"c-s0l9fyim":{"id":"c-s0l9fyim","author":"Carlos Scheidegger","created_at":"2026-07-30T17:52:14.099210Z","text":"Open diagnostic question (which mode hit the original failing browser: ws 401 vs slow 101) is now unanswerable — the user lost the tab. Moot for the fix: both modes are classified and the race mode is eliminated. Committed on braid/bd-tux4m6od-join-collection-error-ux: d947fab2 (fix) + a02cde3f (changelog)."},"c-uht26g2b":{"id":"c-uht26g2b","author":"Carlos Scheidegger","created_at":"2026-07-30T17:56:16.472334Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/439 (branch feature/bd-tux4m6od-join-collection-error-ux). cargo xtask verify --skip-hub-build green before push; hub-client legs (test:ci, build:all, test:wasm) green by hand. Watching CI."}}}
 {"id":"bd-tv2s","title":"Cargo: upgrade automerge v0.8.0 → v0.9.0","description":"Major upgrade surfaced by cargo-upgrade survey 2026-05-04. Current 0.8.0 is range-pinned in workspace; latest is 0.9.0. Type: pre-1.0 minor (semver-breaking). Review changelog and bump deliberately. See claude-notes/plans/2026-05-04-cargo-upgrade-survey.md and bd-hb8h.","status":"closed","priority":3,"issue_type":"chore","created_at":"2026-05-04T18:15:54.659246Z","created_by":"cscheid","updated_at":"2026-07-27T21:28:50.578415Z","closed_at":"2026-07-27T21:28:50.578415Z","close_reason":"SUPERSEDED — the upgrade happened via other work and outran the target. The strand asked for automerge 0.8.0 -> 0.9.0 (filed 2026-05-04); the tree today declares automerge = { version = \"0.10.0\" } in all three consumers (quarto-hub, quarto-hub-provider, quarto-preview), and 0.10.0 is the current crates.io latest stable. Nothing left to do. Found by the in-progress/chore strand audit (bd-a0eyjshu).","labels":["cargo","deps"],"dependencies":{"bd-0a3b:blocks":{"depends_on_id":"bd-0a3b","type":"blocks","created_at":"2026-05-04T19:06:15.793870Z","created_by":"cscheid"},"bd-hb8h:discovered-from":{"depends_on_id":"bd-hb8h","type":"discovered-from","created_at":"2026-05-04T18:16:04.613332Z","created_by":"cscheid"}}}
 {"id":"bd-tvtknbhx","title":"Interactive task-list checkboxes: toggle in hub-client/q2 preview --allow-edit writes [ ]/[x] back to qmd source","description":"Once bd-obkvhlam lands (task markers as Str U+2610/U+2612 with source-info covering the [ ]/[x] bytes, rendered as input[type=checkbox]), make the checkboxes live in edit-capable surfaces: hub-client and q2 preview --allow-edit (flag already exists). Toggle = minimal source splice of the marker's byte range through the existing block-editing substrate (apply_node_edit -> reconcile -> incremental_write; see claude-notes/designs/2026-06-06-block-editing-design.md and claude-notes/plans/2026-07-21-task-list-rendering.md Phase 7). Static render keeps disabled checkboxes (Q1 parity); edit surfaces un-disable and wire the click. Design questions for the session: where the DOM->marker-range mapping lives (render-component API vs data attribute), concurrent-edit self-heal semantics for the 1-byte splice, and whether preview without --allow-edit shows disabled or hidden interactivity.","status":"open","priority":2,"issue_type":"feature","created_at":"2026-07-21T19:11:31.557540Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-10T18:22:47.705293Z","dependencies":{"bd-eias3e39:discovered-from":{"depends_on_id":"bd-eias3e39","type":"discovered-from","created_at":"2026-07-21T19:11:31.557540Z","created_by":"Carlos Scheidegger"},"bd-obkvhlam:blocks":{"depends_on_id":"bd-obkvhlam","type":"blocks","created_at":"2026-07-21T19:11:31.557540Z","created_by":"Carlos Scheidegger"}},"comments":{"c-1bmim7gl":{"id":"c-1bmim7gl","author":"Carlos Scheidegger","created_at":"2026-07-21T19:46:25.870510Z","text":"Implemented in 5852d0ee (same session as the bd-obkvhlam fix): checkbox toggle in q2 preview --allow-edit flips [ ]/[x] on disk via the subtree edit channel; verified live with chrome-devtools (both directions, editor-activation guard, label-click guard, rich-text commit preserves marker). Remaining before close: (1) live verification in hub-client proper (shares PreviewRoot, only q2-preview surface was browser-tested); (2) polish: loose/Para-leading task items still render the ballot glyph as text (label can't wrap <p>); (3) the rich-text block editor displays the raw ballot glyph for task items; (4) subtree commit canonicalizes '-' bullets to '*' (property of whole-list rewrite; a marker-only splice would need inline-level apply_node_edit)."},"c-5lo1vnw9":{"id":"c-5lo1vnw9","author":"Carlos Scheidegger","created_at":"2026-09-10T18:22:47.705293Z","text":"bd-qif9l4cx (2026-09-10) fixes polish item (2): loose/Para-leading task items now render li > p > label > input in q2-preview and toggle through the same subtree channel. Items (3) rich-text editor shows the raw ballot glyph and (4) bullet canonicalisation '-' -> '*' on toggle remain open."}}}
+{"id":"bd-twfep08m","title":"Auto-discover linked local files as page resources (links copy to output, like Q1)","description":"Forwarded issue: a website page linking a local non-document file ([x](data.csv), [y](dashboard.json){download=...}) renders the href unchanged but never copies the file to _site/, silently. Q1 copies these (HTML postprocessor over a[href] etc.). q2 copies Markdown images (ResourceCollectorTransform) but skips Link targets. Workaround: project.resources / resources:.\n\nPlan: add a Rust AST walk over Link targets that resolves existing in-project, non-document files and appends them to meta.resources (so post filters can remove them), tagged with a new ResourceOrigin::AutoDiscovery; gated by a merged-metadata key (proposed resource-discovery: false). Plan doc: claude-notes/plans/2026-10-29-linked-resource-discovery.md. Repros in the plan's appendix.","status":"in_progress","priority":1,"issue_type":"feature","assignee":"claude","created_at":"2026-10-29T18:22:42.170989Z","created_by":"Carlos Scheidegger","updated_at":"2026-10-29T19:11:45.845712Z","external_ref":"gh-967","labels":["resources","website"],"dependencies":{"bd-o8pr:related":{"depends_on_id":"bd-o8pr","type":"related","created_at":"2026-10-29T18:22:42.170989Z","created_by":"Carlos Scheidegger"}},"comments":{"c-li2ytjyu":{"id":"c-li2ytjyu","author":"Carlos Scheidegger","created_at":"2026-10-29T18:32:14.115789Z","text":"Plan reviewed 2026-10-29. All 7 decisions resolved (images on collector; Finalization placement, document post-filter-only pruning; author entries defended; exclusions as proposed; all formats; no missing-target warning; key = resource-discovery). Proposed values: boolean only in v1; non-boolean warns + defaults true; map form reserved for the navigation follow-up. Awaiting go-ahead."},"c-my937kx9":{"id":"c-my937kx9","author":"Carlos Scheidegger","created_at":"2026-10-29T19:11:42.769300Z","text":"Implemented in 982985db5; PR https://github.com/quarto-dev/q2/pull/967 (branch feature/bd-twfep08m-linked-resource-discovery). cargo xtask verify green. Follow-ups to file after review: nav hrefs (AutoDiscovery via navigation), missing-target lint."}}}
 {"id":"bd-typst-skylighting-av61oid8","title":"Typst: wire Skylighting syntax highlighting (12 of 13 syntax-highlighting fixtures fail)","description":"Finding #12 of claude-notes/plans/2026-09-29-typst-smoke-all-followup.md (workspace-4, typst-testing/workspace-4), corrected by a research pass on 2026-10-01. The plan said Skylighting for Typst has no implementation at all. That is wrong: pandoc 3.11's Typst writer already emits #Skylighting(...) / #KeywordTok(...) plus a highlighting-definitions block, and Q2's template (resources/pandoc-filters/typst-template/definitions.typ) already has the $highlighting-definitions$ slot. What is missing is the plumbing.\n\nVerified facts:\n- Pandoc 3.11 Typst writer defaults to idiomatic output (bare fenced blocks) unless told otherwise. `highlight-style: tango` in a qmd already produces real Skylighting output via q2 render, through the deprecated --highlight-style flag (pandoc prints a deprecation warning).\n- crates/quarto-core/src/pandoc_filters/format_defaults.rs forwards only highlight-style. Nothing reads syntax-highlighting:. There is no default style when the user sets nothing (Q1: kDefaultHighlightStyle, pandoc.ts ~1790-1860). No .theme file support, no light/dark selection, no brand monospace-* tokens.\n- Upstream has not fixed the block styling: skylighting-format-typst (last commit Feb 2026) and pandoc 3.11 still emit `block(fill: bgcolor, blocks)` with no width/inset/radius. Q1 patches this with a regex postprocessor in src/format/typst/format-typst.ts (quarto-cli#14126), labelled temporary and still present.\n- Pandoc's Typst writer sets highlighting-definitions with defField, which does not override an existing value. A value supplied from outside wins (tested as markdown metadata, where pandoc escaped it; the real route needs a verbatim path such as -V or a raw Typst inline).\n\nProposed approach (a design decision for the implementer to confirm, not a mandate):\n1. Map syntax-highlighting: and highlight-style: to --syntax-highlighting=, including Q1's default style, none and idiomatic. Small change in format_defaults.rs; should fix most fixtures.\n2. Avoid the regex postprocessor. This repo's policy is no text postprocessors; instead supply our own highlighting-definitions, generated from the theme data Q2 already has (crates/quarto-sass/src/highlight_theme.rs translates pandoc highlight styles for HTML). That gives correct block styling, the brand monospace-block background, line numbers and .theme files without patching pandoc's output. The generated block must define every token function pandoc can emit.\n3. Light/dark: Typst output has no dark mode, so probably the light variant only.\n\nFixtures: .worktrees/workspace-4/crates/quarto/tests/smoke-all/typst/syntax-highlighting is only idiomatic.qmd on that branch; the other 12 (custom-theme, highlight-style-alias, none, skylighting, skylighting-default, skylighting-line-numbers, 6 brand-monospace-*) were excluded as bug #12 and live in the Q1 smoke-all tree.\nRelated: claude-notes/plans/2026-09-18-pandoc-hybrid-typst.md notes Q2 has no equivalent of #14126.","status":"closed","priority":2,"issue_type":"feature","created_at":"2026-10-01T17:29:16.470103Z","created_by":"Gordon Woodhull","updated_at":"2026-10-02T17:21:25.076970Z","closed_at":"2026-10-02T17:21:25.076970Z","close_reason":"Fixed in PR #772 (merged)","comments":{"c-w73o0v79":{"id":"c-w73o0v79","author":"Gordon Woodhull","created_at":"2026-10-01T17:29:26.737549Z","text":"Dispatched to agent typst-skylighting on workspace-7 (branch braid/bd-typst-skylighting-av61oid8-typst-skylighting)"}}}
 {"id":"bd-tyvt","title":"Open Graph / Twitter card / social meta tags","description":"Q1 has metadataHtmlPostProcessor for og:* and twitter:* meta tags. Out of Phase 7 MVP. Q1 reference: external-sources/quarto-cli/src/project/types/website/website-meta.ts. Filed as a separate transform alongside Phase 7's metadata transforms. Originating phase: bd-b9mz.","status":"open","priority":1,"issue_type":"feature","created_at":"2026-04-27T15:03:22.023239Z","created_by":"cscheid","updated_at":"2026-07-29T16:10:48.159+00:00","dependencies":{"bd-0tr6:parent-child":{"depends_on_id":"bd-0tr6","type":"parent-child","created_at":"2026-04-27T15:03:22.023239Z","created_by":"cscheid"},"bd-b9mz:discovered-from":{"depends_on_id":"bd-b9mz","type":"discovered-from","created_at":"2026-04-27T15:03:22.023239Z","created_by":"cscheid"}}}
 {"id":"bd-tzwcof0n","title":"Lua parity B: Attr argument shapes (parse_attr, pandoc.Attr, AttributeList)","description":"Support Pandoc positional triple and HTML-like map (class space-split) in parse_attr and pandoc.Attr(table); accept list-of-pairs and AttributeList userdata for attributes; give attr.classes the List metatable; add AttributeList constructor. Keep q2 named-key form (Decision 2). Differential cases: attr-positional-triple, attr-html-like-map. Catalog: claude-notes/research/2026-07-13-lua-api-mismatch-catalog.md. Epic plan: claude-notes/plans/2026-07-13-lua-api-pandoc-parity.md.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-07-13T17:09:20.466776Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-13T19:20:13.665801Z","closed_at":"2026-07-13T19:20:13.665801Z","close_reason":"Full Pandoc attr shape support + AttributeList + classes as pandoc List with cache+readback. Track-1 xfail 60->43, differential 4->2 (attr silent-empty class eliminated), e2e verified vs pandoc.","dependencies":{"bd-grkrb9nj:parent-child":{"depends_on_id":"bd-grkrb9nj","type":"parent-child","created_at":"2026-07-13T17:09:20.466776Z","created_by":"Carlos Scheidegger"}}}
@@ -1743,6 +1759,7 @@
 {"id":"bd-u7zj5p9w","title":"workspace-ts-suites: first red step still hides every later suite","description":"The workspace-ts-suites job in .github/workflows/ts-test-suite.yml runs ~12 suite tiers as sequential steps with no step-level continue-on-error (fail-fast:false is on the OS matrix only). A failing step skips all later steps, so every red run only ever names the FIRST flake in step order and the suites behind it produce no evidence. When wasm-js-bridge went red, the 5 suites after it never ran, and 'unproven' was read as 'clean'.\n\nA working prototype was built and verified to parse during bd-yw3mcdkg: give each suite step an id plus continue-on-error:true, split the two steps that chain two tiers with && (q2-preview-spa, kanban) so the unit tier stops hiding the integration tier, and add a final gate step that reads each step's outcome, writes a per-suite table to GITHUB_STEP_SUMMARY and exits 1 if any failed. The gate must be bash-3.2 safe (no associative arrays) because macOS runners ship bash 3.2. It was reverted before commit as out of scope for that strand.\n\nNote this changes nothing about merge gating either way: main has no required status checks, so these jobs are advisory already.","status":"open","priority":2,"issue_type":"task","created_at":"2026-08-23T23:34:38.236826Z","created_by":"Gordon Woodhull","updated_at":"2026-08-23T23:34:38.236826Z","dependencies":{"bd-yw3mcdkg:discovered-from":{"depends_on_id":"bd-yw3mcdkg","type":"discovered-from","created_at":"2026-08-23T23:34:38.236826Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-u996g8g2","title":"Jupyter inline expression evaluation ({python} expr) missing from the production text path","description":"Q1 supports inline executable expressions in jupyter documents. q2's production jupyter path (text_execute.rs) has never supported them; the only implementation was a prototype in the retired AST-path JupyterTransform (deleted in bd-gthycd33 because the parallel emitter was a shape-divergence source with no production consumer — see claude-notes/plans/2026-07-01-bd-gthycd33-jupyter-cell-wrapper.md, decision 4). knitr handles inline r expressions natively. When implementing, add the emission to text_execute.rs (single emission path) and extend the cross-engine parity suite (engine_output_parity.rs) with an inline-expression case so knitr and jupyter inline output shapes stay congruent. The deleted prototype (git history: crates/quarto-core/src/engine/jupyter/transform.rs, extract_inline_expressions / parse_inline_expression) is a useful reference.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-07-02T13:00:04.800653Z","created_by":"Carlos Scheidegger","updated_at":"2026-07-02T13:00:04.800653Z","dependencies":{"bd-gthycd33:discovered-from":{"depends_on_id":"bd-gthycd33","type":"discovered-from","created_at":"2026-07-02T13:00:04.800653Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-ubjo","title":"L8 follow-up: broader path resolution for YAML-declared paths","description":"L8 D2 deferred. The planned !path YAML tag + Q2 metadata-merging design should unify path resolution across _quarto.yml, per-page frontmatter, and listing templates. Listing template paths would resolve relative to the YAML they were defined in. L8 ships host-page-only resolution; this issue records the linkage so when the broader design lands, listing template paths fold in cleanly. See claude-notes/plans/2026-05-07-listings-L8-custom-templates.md §Decisions log D2.","status":"open","priority":3,"issue_type":"feature","created_at":"2026-05-08T14:57:39.790641Z","created_by":"cscheid","updated_at":"2026-05-08T14:57:39.790641Z","dependencies":{"bd-rqgx:discovered-from":{"depends_on_id":"bd-rqgx","type":"discovered-from","created_at":"2026-05-08T14:57:39.790641Z","created_by":"cscheid"}}}
+{"id":"bd-ucxaqjn2","title":"smokeAll.wasm: extensions/builtin-version-shortcode/test.qmd fails on main","description":"The smoke-all fixture 'extensions/builtin-version-shortcode/test.qmd [html]' fails locally on a pristine main checkout: ensureFileRegexMatches expects pattern '0\\.1\\.0' NOT to match the rendered HTML, but it matches. Repro: cd hub-client && npx vitest run --config vitest.wasm.config.ts src/services/smokeAll.wasm.test.ts. Found during bd-k8ptcrd9 Phase 0 gates (2026-10-29); appears environment/version-dependent since CI is green.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-29T19:38:41.182916Z","created_by":"shikokuchuo","updated_at":"2026-10-29T19:38:41.182916Z","dependencies":{"bd-k8ptcrd9:discovered-from":{"depends_on_id":"bd-k8ptcrd9","type":"discovered-from","created_at":"2026-10-29T19:38:41.182916Z","created_by":"shikokuchuo"}}}
 {"id":"bd-udlt","title":"L9 follow-up: title placeholder substitution from rendered HTML","description":"Q1 substitutes the rendered post title (post-engine) into RSS items, because engine output may contain math etc. that the metadata title doesn't. v1 uses item.title from the profile directly. Subscribers see the metadata title; if a feed item has math in its title, it'll show up as source markup, not the rendered form. File if a user reports this.","status":"deferred","priority":4,"issue_type":"feature","created_at":"2026-05-08T17:33:32.947255Z","created_by":"cscheid","updated_at":"2026-07-29T14:51:54.109+00:00","dependencies":{"bd-o90m:discovered-from":{"depends_on_id":"bd-o90m","type":"discovered-from","created_at":"2026-05-08T17:33:32.947255Z","created_by":"cscheid"}}}
 {"id":"bd-udrn0q47","title":"Nested {{< include >}} retargets relative paths to the including file (should be no-retargeting, relative to original)","description":"VERIFIED E2E 2026-06-16. q2's include expansion resolves a NESTED include's relative path against the including file's directory, not the original (top-level) document. This violates the documented Quarto 1/2 'no path retargeting' design for include shortcodes (https://quarto.org/docs/authoring/includes.html — Carlos authored that feature).\n\nRepro: main.qmd (root) {{< include sub/part.qmd >}}; sub/part.qmd {{< include other.qmd >}}; other.qmd at BOTH root and sub/. q2 render main.qmd includes sub/other.qmd (SUB-OTHER) — i.e. retargeted to sub/. Expected per design: root other.qmd.\n\nNote the ASYMMETRY: image/resource paths ARE correctly un-retargeted (ResourceCollectorTransform resolves over the expanded AST relative to the ORIGINAL deck dir; expand_includes_in_blocks never rewrites URL strings). Only nested INCLUDE paths are wrongly retargeted.\n\nRoot cause: crates/quarto-core/src/stage/stages/include_expansion.rs:248 recurses with current_file = &resolved (the just-included file), so the nested level's base_dir = include_expansion.rs:90 becomes the including file's parent. Fix: thread the ORIGINAL top-level document dir as the include-resolution anchor through the recursion instead of the including file's dir.\n\nSeparate-but-related: base_dir.join(\"/foo.qmd\") yields filesystem root /foo.qmd, so recommended project-root-relative leading-slash include notation may not resolve to project root through this path — verify and handle.\n\nReference: claude-notes/research/2026-06-16-include-shortcode-path-resolution.md. Single-level includes from the deck root are unaffected; only nested includes whose including file is in a subdir with a relative arg. Preview parity work (bd-9cyza5vy) should reuse the real IncludeExpansionStage so it inherits whatever this fix lands on.","status":"open","priority":2,"issue_type":"bug","created_at":"2026-06-17T00:16:03.464047Z","created_by":"Carlos Scheidegger","updated_at":"2026-06-17T00:17:09.677219Z","dependencies":{"bd-9cyza5vy:discovered-from":{"depends_on_id":"bd-9cyza5vy","type":"discovered-from","created_at":"2026-06-17T00:16:03.464047Z","created_by":"Carlos Scheidegger"}},"comments":{"c-jlcoavtw":{"id":"c-jlcoavtw","author":"Carlos Scheidegger","created_at":"2026-06-17T00:17:09.677219Z","text":"Confirmed against Quarto 1 source: external-sources/quarto-cli/src/core/handlers/include-standalone.ts (standaloneInclude). Every include resolves via the SAME fixed handlerContext.resolvePath(filename); the nested retrieveInclude(params[0]) reuses the same handlerContext, anchored at the original document target.source. So Q1 anchors ALL include paths (top-level + nested) at the original file = no retargeting. Q1 also splices raw TEXT (mappedConcat) and parses the concatenation ONCE, so resource paths are un-retargeted by construction. q2 AST-level splicing (expand_includes_in_blocks, parses each include separately) is where the nested-include asymmetry was introduced. Fix: anchor q2 recursion at the original top-level document dir to match Q1."}}}
 {"id":"bd-ue80chl0","title":"q2-preview-spa does not load render-components (hub-client-only feature)","description":"Discovered during bd-5m4ga0s1 Phase 4 e2e: the CLI preview SPA (q2-preview-spa/src/PreviewApp.tsx) never transpiles or passes customComponentsCode to Q2PreviewIframe, so YAML render-components: entries are silently ignored under 'q2 preview' — the feature only works in hub-client (hub-client/src/components/render/ReactRenderer.tsx does the babel transpile + LOAD_CUSTOM_COMPONENTS). Iframe side already supports it (entry.tsx loadCustomComponents), so the gap is parent-side only: read render-components from meta, resolve paths against the preview file server, transpile (needs @babel/standalone in the SPA bundle), pass through. Decide: implement, or document as hub-client-only.","status":"closed","priority":2,"issue_type":"task","created_at":"2026-07-20T16:55:41.567102Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-26T13:41:11.906775Z","closed_at":"2026-08-26T13:41:11.906775Z","close_reason":"Implemented in PR #604 (merged 2026-08-25): q2 preview now loads render-components in project and single-file mode — shared transpiler/meta-walk in preview-renderer, SPA parent half with lazy babel chunk + tsxTick cadence + overlay warnings, iframe re-render-after-LOAD fix, single-file closure pickup. e2e green against the real binary. Plan: claude-notes/plans/2026-08-25-preview-spa-render-components.md","dependencies":{"bd-5m4ga0s1:discovered-from":{"depends_on_id":"bd-5m4ga0s1","type":"discovered-from","created_at":"2026-07-20T16:55:41.567102Z","created_by":"Carlos Scheidegger"}},"comments":{"c-ob2ch7xb":{"id":"c-ob2ch7xb","author":"Carlos Scheidegger","created_at":"2026-08-25T20:24:38.323616Z","text":"PR opened: https://github.com/quarto-dev/q2/pull/604 (full verify green: 13,426 Rust tests + all TS suites + hub-client build:all). Awaiting review/CI; strand stays in_progress until merge."},"c-v0sp4o9k":{"id":"c-v0sp4o9k","author":"Carlos Scheidegger","created_at":"2026-07-20T17:42:21.532742Z","text":"Escalated to GitHub for the feature's primary author: https://github.com/quarto-dev/q2/issues/402 (filed 2026-07-20 at user request; issue body carries the full parent/iframe analysis and the two resolution options)."},"c-yrzc7tzx":{"id":"c-yrzc7tzx","author":"Carlos Scheidegger","created_at":"2026-08-25T20:18:43.449141Z","text":"Implemented (resolution #1 of GH #402) on branch braid/bd-ue80chl0-preview-spa-render-components. Phase 1: shared transpiler + meta walk in @quarto/preview-renderer/utils (hub-client refactored, behavior unchanged). Phase 2: SPA parent half (lazy babel chunk 2.9MB/664KB gz; tsxTick cadence; warnings in overlay) + iframe re-render-after-LOAD fix in entry.tsx. Phase 3: single-file closure syncs render-components .tsx (resolve_single_file_deps). e2e green against the real binary in both modes incl. live .tsx edits. Plan: claude-notes/plans/2026-08-25-preview-spa-render-components.md. PR link to follow after full verify."}}}
@@ -1862,6 +1879,7 @@
 {"id":"bd-wuiu1of7","title":"qmd writer emits decoded zero-width/format characters raw; output fails to re-parse (GH #672)","description":"https://github.com/quarto-dev/q2/issues/672. Since the named-entity decode (bd-named-entities-w6xbfftj, PR #488) the reader folds &ZeroWidthSpace; (and &shy;, &lrm;, &rlm;, &NoBreak;, &af;, &it;, &ic;, and numeric refs like &#x200B;) into a Str holding the literal Cf-category codepoint. The qmd writer emits that codepoint verbatim, but the grammar's prose regexes (pandoc_str in grammar.js) accept no Cf character except ZWNJ/ZWJ, so pampa's own output is a parse error; inside a heading superscript it cascades into Q-2-16 Unclosed Superscript. In the wild: quarto-web index.qmd '# Welcome to Quarto^&ZeroWidthSpace;[®]{.trademark}^'. Verified at 7ef59618: every Cf codepoint probed (U+00AD, U+200B, U+200E/F, U+2060-2064, U+FEFF, U+061C, U+202A-E, U+2066-9) is a hard parse error in prose; code spans, code blocks, link titles and attribute values accept them. Two-sided fix: (1) writer re-encodes Cf chars as named entities (numeric fallback); (2) grammar accepts \\p{Cf} raw in prose, mirroring the \\p{M} fix in bd-96fswwce (Pandoc parity; deliberately non-canonical at the text level). Plan: claude-notes/plans/2026-09-11-zero-width-chars-writer-parser-gh672.md","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-09-11T13:49:24.320395Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-11T15:51:22.351907Z","closed_at":"2026-09-11T15:51:22.351907Z","close_reason":"Fixed in caa29d6c (writer: Cf chars -> preferred WHATWG entity or &#xXXXX;, ZWNJ/ZWJ raw; grammar: PANDOC_COMBINING_MARKS widened to \\p{M}\\p{Cf}, parser regenerated) + b15868ed (bd-i18zoy4n, & escaping) on branch braid/bd-wuiu1of7-zero-width-entities. Full cargo xtask verify green for both commits; end-to-end q2 render of the quarto-web heading inspected. Plan: claude-notes/plans/2026-09-11-zero-width-chars-writer-parser-gh672.md","labels":["pampa","parity","tree-sitter-qmd"],"dependencies":{"bd-96fswwce:related":{"depends_on_id":"bd-96fswwce","type":"related","created_at":"2026-09-11T13:49:28.974483Z","created_by":"Carlos Scheidegger"},"bd-named-entities-w6xbfftj:discovered-from":{"depends_on_id":"bd-named-entities-w6xbfftj","type":"discovered-from","created_at":"2026-09-11T13:49:26.661741Z","created_by":"Carlos Scheidegger"}}}
 {"id":"bd-wv4e","title":"Tree-sitter scanner over-eagerly excludes *, -, +, #, digits from soft-break","description":"The same class of bug as bd-af1e affects 5 more characters. The scanner's line-break handler at scanner.c:2286-2291 and 2371-2376 excludes these from soft-line-break candidates whenever they appear at the start of a continuation line:\n- '*' (could be emphasis, list, thematic break)\n- '-' (could be list, thematic break, setext header)\n- '+' (could be list)\n- '#' (could be ATX header)\n- digits 0-9 (could be ordered list)\n\nBut each of these only opens a block in specific contexts (most need a trailing space, '#' needs space, '*'/'-'/'+' need space for list, '***'/'---' need 3+ for thematic break, digits need '.' or ')' followed by space). A bare 'X' followed by content is just inline text and should soft-break.\n\nVerified pampa vs pandoc divergence on:\n  foo bar\n  *emph* baz                 → pampa splits, pandoc soft-breaks\n  -text more                 → pampa splits, pandoc soft-breaks\n  +text more                 → pampa splits, pandoc soft-breaks\n  #hashtag more              → pampa splits, pandoc soft-breaks\n  123abc more                → pampa splits, pandoc soft-breaks\n\nCorrect exclusions (pampa already matches pandoc for):\n  * list item                → BulletList opens (correct)\n  - list item                → BulletList opens (correct)\n  # heading                  → Header opens (correct)\n  1. item                    → OrderedList opens (correct)\n\nFix approach: same as bd-af1e (peek-count). For each char, peek the trailing context to decide if it actually opens a block:\n- '*', '-': count run; if >=3 with all whitespace/eol after, thematic break (interrupt). If '* ' or '*\\t', list (interrupt). Otherwise inline.\n- '+': only list if followed by whitespace.\n- '#': only ATX if followed by whitespace or eol.\n- digits: only ordered list if followed by '.' or ')' AND space.\n\n':' (fenced div / definition list) was excluded from this issue because its semantics are more entangled with qmd-specific features and pandoc-mode-specific rules — file separately if regression observed.\n\nSibling of bd-af1e. See claude-notes/plans/2026-04-30-tree-sitter-backtick-paragraph-split.md for the bd-af1e implementation pattern.","status":"closed","priority":1,"issue_type":"bug","created_at":"2026-04-30T18:40:17.998897Z","created_by":"cscheid","updated_at":"2026-04-30T18:45:29.625753Z","closed_at":"2026-04-30T18:45:29.625599Z","close_reason":"Re-scoped: see bd-wv4e replacement for asterisks-only fix. The other characters (-, +, #, digits, :) have a backslash-escape workaround (\\-text, \\#hashtag, etc.) so are not severe enough to fix in this pass.","dependencies":{"bd-af1e:related":{"depends_on_id":"bd-af1e","type":"related","created_at":"2026-04-30T18:40:25.341409Z","created_by":"cscheid"}}}
 {"id":"bd-wzhsf","title":"CVE-prevention: cookie+Bearer dual-credential MUST return 400","description":"Highest-impact item flagged in plan §Phase 2. A request carrying BOTH a session cookie AND an Authorization: Bearer header MUST be rejected with HTTP 400 and body {\"error\":\"conflicting_credentials\"} BEFORE any auth-decision logic runs.\n\nTest name: cookie_and_bearer_returns_400 in crates/quarto-hub/tests/auth_bearer.rs.\n\nThe dual-credential 400 rule must win over CSRF, WS-Origin, and per-credential auth checks (covered by sibling test dual_credential_400_wins_over_csrf_and_origin).\n\nPlan §Phase 2 — Cross-cutting security requirements: claude-notes/plans/2026-05-05-hub-mcp-device-flow-implementation.md","status":"closed","priority":0,"issue_type":"bug","created_at":"2026-05-20T14:27:46.526825Z","created_by":"shikokuchuo","updated_at":"2026-05-20T21:22:18.258188Z","closed_at":"2026-05-20T21:22:18.258155Z","close_reason":"Dual-credential 400 enforced. cookie_and_bearer_returns_400 (REST), ws_upgrade_rejects_dual_credentials (WS), and dual_credential_400_wins_over_csrf_and_origin all pass. Body shape locked as {\"error\":\"conflicting_credentials\"} in extract_credential and the WS handler; auth_refresh re-runs the same check before CSRF. See plan §Phase 2 — completion notes (2026-05-20).","dependencies":{"bd-0ufq5:parent-child":{"depends_on_id":"bd-0ufq5","type":"parent-child","created_at":"2026-05-20T14:27:46.526825Z","created_by":"shikokuchuo"}}}
+{"id":"bd-wzz90981","title":"Jupyter figure output: unpadded base64 image/png (Ark R kernel) fails decode, falls back to '[Image output]' placeholder","description":"write_image (crates/quarto-core/src/engine/jupyter/text_execute.rs ~L685) decodes image payloads with base64 STANDARD, which requires '=' padding. The Ark R kernel (posit-dev/ark 0.1.252) sends image/png base64 with the trailing padding stripped, so every plot is dropped: warn 'jupyter figure: invalid base64; skipping' (Invalid padding), the cell falls back to the '[Image output]' placeholder, and that triggers Q-2-49 ('has no attribute block...').\n\nReproduced by probing the kernel with jupyter_client (plot(1:3)): payload is 12959 chars (len %4 == 3), no whitespace, no non-base64 chars, only mime image/png (no text/plain). Adding the missing '=' decodes to a valid PNG. Padding is optional in practice, so any kernel may do this; Ark is just the first seen.\n\nFix: build the decoder with decode_padding_mode(DecodePaddingMode::Indifferent) (base64 0.22 GeneralPurposeConfig) and strip all whitespace rather than only '\\n'. Test: feed write_image a valid payload with the final '=' removed and assert it decodes and writes the file. Also worth a parity case in engine_output_parity.rs if an R kernel is available.\n\nRepro: q2 render of an {r} cell with plot(...) using the ark kernelspec registered as 'ir' (q2 currently hardcodes r -> ir; see bd-9pvolzwl / bd-db68qtuf). Follow-up to bd-5t6wvu7m (closed, PR #412 introduced FigureWriter).","status":"open","priority":2,"issue_type":"bug","created_at":"2026-10-28T17:38:56.772445Z","created_by":"Gordon Woodhull","updated_at":"2026-10-28T17:38:56.772445Z","labels":["jupyter"],"dependencies":{"bd-5t6wvu7m:related":{"depends_on_id":"bd-5t6wvu7m","type":"related","created_at":"2026-10-28T17:38:56.772445Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-x0o0pem3","title":"Determine whether the quarto.config.md Lua path's offset arithmetic is live or inert","description":"Liveness question discovered while sizing the audit for bd-mxa44voa. This may be a no-op; the point of the strand is to determine that cheaply rather than leave it unknown.\n\nTHE SITE. crates/pampa/src/lua/config_value.rs:601-631 implements the quarto.config.md Lua API (the filter-facing analog of the !md tag). It calls:\n\n    crate::readers::qmd::read(s.as_bytes(), ..., Some(filter_source_info(lua)))\n\nThis is the same call shape as parse_config_string_as_markdown in crates/pampa/src/pandoc/meta.rs, which is the confirmed instance of the bd-mxa44voa bug: a parent SourceInfo supplied as the offset base for a recursive parse of a string, where the string's bytes are not a prefix-aligned byte-identical slice of the parent's span. Of roughly 50 readers::qmd::read call sites in the tree, these two are the ONLY ones in production that pass Some(parent).\n\nWHY IT MAY BE INERT. filter_source_info (crates/pampa/src/lua/types.rs:2291) returns SourceInfo::Generated { by: By::filter(path, line), from: [] } -- a Generated marker with an EMPTY anchor list. SourceInfo::Substring::resolve_byte_range() on a Generated parent delegates to its Invocation anchor; with from: [] there is none, so resolve_byte_range() and map_offset() return None. The defective arithmetic is therefore performed but never resolves to a byte position, and nothing user-visible comes of it.\n\nTHE UNKNOWN. Whether any code calls append_anchor to attach an Invocation anchor to a filter_source_info()-derived Generated BEFORE the resulting Substring is read for diagnostics or for preimage_in. If something does, the arithmetic becomes live and the site is a real instance of the class. The audit did not trace all append_anchor call sites across quarto-core and pampa.\n\nTASK.\n1. Enumerate append_anchor call sites and determine whether any can reach a filter_source_info()-derived Generated.\n2. If NO: close this, and add a comment at config_value.rs:601 recording WHY it is safe, so the next auditor does not have to re-derive it. The safety is non-obvious and depends on an empty anchor list several crates away.\n3. If YES: fix it the same way as the rest of the class (see bd-mxa44voa), and write a test that exercises quarto.config.md from a Lua filter with an escaped string.\n\nEither outcome is cheap and removes an open question from the audit.","status":"closed","priority":3,"issue_type":"question","created_at":"2026-08-21T01:07:22.153387Z","created_by":"Gordon Woodhull","updated_at":"2026-08-21T01:46:23.770791Z","closed_at":"2026-08-21T01:46:23.770791Z","close_reason":"Absorbed into claude-notes/plans/2026-08-20-provenance-3-audit-and-fix.md (Phase 2). Plan 3 now audits and fixes, so the liveness question plus its two outcomes (document why inert, or fix if live) are checklist rows there.","labels":["diagnostics"],"dependencies":{"bd-mxa44voa:parent-child":{"depends_on_id":"bd-mxa44voa","type":"parent-child","created_at":"2026-08-21T01:07:22.153387Z","created_by":"Gordon Woodhull"}}}
 {"id":"bd-x113wg9v","title":"Doc-level resource_error_to_parse_error binds the .qmd to a _metadata.yml/_quarto.yml-anchored SourceInfo (sibling of bd-p86nlm92)","description":"The doc-level twin of bd-p86nlm92, explicitly deferred by PR #478's docstring. project_resources.rs resource_error_to_parse_error (doc-level callers pass doc_source_abs) binds the document's own path/content to whatever FileId the ResourceError's SourceInfo resolves to — but doc-level resources: patterns come from MERGED metadata (document_profile.rs:117-124 reads doc.ast.meta post-MetadataMergeStage), so a pattern inherited from blog/_metadata.yml, _quarto.yml, or an extension's contributes.metadata carries that file's filename-hash FileId. Symptoms identical to bd-m6wmztln: confidently wrong span when the offsets fit inside the .qmd, dropped snippet when they don't. Fix: route through config_sources::bind_config_source with candidates [doc path, config_path] ++ extension_manifest_paths ++ the doc's directory _metadata.yml layer paths (enumerable via directory_metadata_for_document). TDD: fixture with resources: [\"../outside/*\"] declared in a _metadata.yml, assert the Q-5-1 snippet names the _metadata.yml. Assessment: claude-notes/research/2026-08-09-fileid-span-integrity-audit.md (finding D3).","status":"closed","priority":2,"issue_type":"bug","created_at":"2026-08-09T17:11:01.013617Z","created_by":"Carlos Scheidegger","updated_at":"2026-08-10T14:29:28.212087Z","closed_at":"2026-08-10T14:29:28.212087Z","close_reason":"Merged to main via PR #484 (quarto-dev/q2, merge commit 485e864c); full CI matrix green (macOS/Ubuntu Rust+TS suites, WASM, Hub E2E, Snyk).","dependencies":{"bd-nv4p0eb1:discovered-from":{"depends_on_id":"bd-nv4p0eb1","type":"discovered-from","created_at":"2026-08-09T17:11:01.013617Z","created_by":"Carlos Scheidegger"}},"comments":{"c-3n3sxma5":{"id":"c-3n3sxma5","author":"Carlos Scheidegger","created_at":"2026-08-10T00:20:44.868866Z","text":"Done on braid/bd-x113wg9v-doc-resource-binding, merged to feature/bd-nv4p0eb1-span-hardening (phase A part 2). TDD red-first e2e (Q-5-1 underlined post.qmd at _metadata.yml offsets); fix via new bind_source_candidates (FileId,Path pairs spanning dense doc ids + hash ids) + directory_metadata_paths_for_document (same spelling as the hashed parse paths). E2E inspected: anchors blog/_metadata.yml:2:5 + inherited-from info line. Lint blessings for render_scripts.rs/project_resources.rs removed. Workspace 11217 green."}}}
 {"id":"bd-x17470jx","title":"Perf: TemplateContext::child deep-clones the whole parent variable table on every $for$ iteration","description":"quarto-doctemplate context.rs child(): 'parent: Some(Box::new(self.clone()))' — a deep clone of the parent's HashMap<String, TemplateValue> (and, recursively, its parent chain) for every loop iteration (evaluator.rs for-loop: ctx.variables.child() per item). With the Connect docs' large sidebar/navbar metadata the per-document template render spends 4.2% of whole-project CPU inside this clone (514 samples; it is the second-largest memmove caller after map_offset). Fix: share the parent by reference/Rc (child holds &TemplateContext or Rc<TemplateContext>) so scope entry is O(1). Profile: claude-notes/research/2026-09-17-connect-docs-render-profile.md","status":"open","priority":2,"issue_type":"bug","created_at":"2026-09-17T20:07:24.473452Z","created_by":"Carlos Scheidegger","updated_at":"2026-09-17T20:07:24.473452Z","labels":["perf"],"dependencies":{"bd-5yektmwt:discovered-from":{"depends_on_id":"bd-5yektmwt","type":"discovered-from","created_at":"2026-09-17T20:07:24.473452Z","created_by":"Carlos Scheidegger"}}}
diff --git a/claude-notes/instructions/performance-profiling.md b/claude-notes/instructions/performance-profiling.md
index 69acf6423f..0e5a8adc94 100644
--- a/claude-notes/instructions/performance-profiling.md
+++ b/claude-notes/instructions/performance-profiling.md
@@ -305,6 +305,22 @@ node hub-client/scripts/perf/analyze-cpuprofile.mjs \
   the real bottleneck is inside the shimmed dep, you'll need a
   browser profile to see it.
 
+**Driving the real render path.** A driver that exercises the production
+render-path modules (`wasmRenderer.ts`, `intelligenceService.ts`, …)
+needs three shims beyond the template. First, extend the resolve hook:
+map `wasm-quarto-hub-client` and `/src/wasm-js-bridge/*` into the
+workspace, `@quarto/*` to their `src/`, and any `*.wasm?url` import to
+a stub string module — otherwise Node tries to load the `.wasm` as ESM
+and fails on its WASI import. Second, shim `fetch` for `file:` URLs so
+`initWasm()`\'s own `.wasm` load succeeds — that keeps the production
+init path in the profile instead of a test-only bypass. Third, import
+`fake-indexeddb/auto` **before** `initWasm()`: the SASS cache bridge
+(`wasm-js-bridge/cache.js`) needs IndexedDB, and without it every
+render recompiles the theme, roughly doubling the measured
+per-keystroke cost. A full driver + fixture built this way is Phase 0
+of `claude-notes/plans/2026-10-24-hub-client-main-thread-render-path.md`
+(bd-k8ptcrd9).
+
 ### 4. Read the numbers before designing the fix
 
 Put the numbers in a table with expected shapes next to them. If you
diff --git a/claude-notes/plans/2026-10-24-hub-client-main-thread-render-path.md b/claude-notes/plans/2026-10-24-hub-client-main-thread-render-path.md
new file mode 100644
index 0000000000..aff32b3288
--- /dev/null
+++ b/claude-notes/plans/2026-10-24-hub-client-main-thread-render-path.md
@@ -0,0 +1,402 @@
+---
+title: 'hub-client main-thread render path: audit fixes'
+date: 2026-10-24
+description: 'Phased fixes from the 2026-10-24 main-thread render-path audit: render coalescing, parse-check-only classification, single-parse AST transport, binary asset transfer, then measurement-gated worker work.'
+status: draft
+---
+
+## Overview
+
+An audit on 2026-10-24 examined what heavy, general-purpose work runs on
+the browser main thread on the hub-client render path, and what can be
+removed, coalesced, or moved off it. Context: the entire Rust WASM render
+pipeline (`wasm-quarto-hub-client`: VFS, qmd parse, transform pipeline, AST
+production, theme CSS compile, LSP intelligence) runs on the main thread;
+the only workers are pandoc.wasm, typst, and Monaco language workers
+(architecture map: `claude-notes/designs/pandoc-wasm-architecture.md`,
+which documents main-thread placement as an accepted decision). Every bit
+of general work sharing that thread shows up as typing latency and preview
+lag.
+
+Eight findings, ranked by expected impact, are below. This plan sequences
+the fixes cheapest-first, with measurement gating the larger ones per
+`claude-notes/instructions/performance-profiling.md`.
+
+Caveat: findings are from static analysis plus code-path reading. The cost
+claims are hypotheses until Phase 0 instruments them. File:line references
+were as of the audit; re-locate before editing.
+
+Revised 2026-10-27: file:line references re-verified against main; updated
+for bd-73lilznv (2026-10-26: the sandboxed preview's default origin moved
+to `quarto-dev.github.io/q2-sandbox`, so the preview iframe is now
+genuinely cross-origin — this affects findings 3 and 6 and Phases 3-4) and
+for the TS-side perf instrumentation that has landed since the audit
+(`hub-client/src/utils/perfCounters.ts`). Phase 3's per-consumer scoping is
+now decided (manifest moves iframe-side; the other consumers stay, with
+reasons recorded inline).
+
+## Findings
+
+| # | Finding | Primary site | Cost class |
+|---|---------|--------------|------------|
+| 1 | Per-keystroke full parse whose AST is discarded (format classification) | `hub-client/src/components/render/PreviewRouter.tsx:134` | wasted work per keystroke |
+| 2 | Renders are debounce-free; stale queued renders still execute | `hub-client/src/components/render/ReactPreview.tsx:768`, `ts-packages/preview-runtime/src/wasmRenderer.ts:494` | N full pipelines per N-keystroke burst |
+| 3 | AST serialized twice, parsed twice per render; untransformed AST always shipped | `ts-packages/preview-runtime/src/wasmRenderer.ts:611`, `hub-client/quarto-hub-sandboxed-preview/src/entry.tsx:330` | multi-MB JSON parse on main thread, discarded |
+| 4 | Systemic JSON-string envelopes on every WASM call; binaries are base64-in-JSON | `ts-packages/preview-runtime/src/wasmRenderer.ts:331` and ~30 sites | boundary tax on every call |
+| 5 | Whole-document semantic tokenization synchronous on main thread | `hub-client/src/services/intelligenceService.ts:263` | editor-blocking tree-sitter per token request |
+| 6 | Asset proxy: base64 + JSON + string postMessage per asset | `hub-client/src/components/render/q2-sandboxed-preview/Q2SandboxedPreviewIframe.tsx:183` | ~33% bloat + copies per asset |
+| 7 | Attribution: full Automerge history replay + char-to-byte map on main thread | `hub-client/src/hooks/useAttribution.ts`, `hub-client/src/services/attribution-runs.ts` | O(history) behind a toggle |
+| 8 | dart-sass compiles on main thread via sync VFS bridge | `ts-packages/preview-runtime/src/wasmRenderer.ts:264` | ~500 ms per theme change, async so yielding |
+
+Amplifier noted during the audit: per keystroke the full document text
+crosses the JS-to-WASM boundary at least three times (sync callback
+`ts-packages/preview-runtime/src/automergeSync.ts:108`, semantic-tokens
+write `intelligenceService.ts:275`,
+`resolvePandocFormats` contentOverride `PreviewRouter.tsx:147`) before the
+render itself reads it. Folded into Phases 2 and 5 rather than tracked
+separately.
+
+Existing related strands: the closed bd-k9tfhsst covered SW asset-proxy
+namespace plumbing, not the base64 wire format (finding 6 is new work).
+Rust-side render CPU has its own perf strands (AST memmove, artifact
+re-flush, template recompiles) — orthogonal to this plan.
+
+## Conventions
+
+- TDD per repo policy: behavior-preserving perf fixes rely on
+  characterization tests written **before** the change; anything
+  user-visible gets a failing regression test first.
+- Measurement per `claude-notes/instructions/performance-profiling.md`:
+  Phase 0 builds the harness; no phase past 1 starts without baseline
+  numbers for its target. TS-side profiling uses the Node
+  `--cpu-prof` driver workflow (never vitest), analysis via
+  `hub-client/scripts/perf/analyze-cpuprofile.mjs`.
+- After each phase: `npm run test:ci` and `npm run build:all` from
+  `hub-client/`. Phases touching `crates/wasm-quarto-hub-client` (2, 3, 4)
+  also need the WASM rebuild chain (`npm run build:wasm`,
+  `cargo xtask build-q2-preview-spa` if the SPA is affected) and
+  `cargo xtask verify`.
+- The TS-side perf infrastructure Phase 0 calls for already exists:
+  `hub-client/src/utils/perfCounters.ts` provides `QUARTO_PERF_STATS`-gated
+  `perfMark`/`perfSegment`/`perfNow` (free when off), and
+  `useAutomergeSync.ts` is already instrumented with it as a worked
+  example. Extend it with the gauges below rather than building a parallel
+  mechanism.
+- End-to-end verification per AGENTS.md: each phase gets a manual smoke
+  in the dev server (or states explicitly that browser verification was
+  not done). Record the invocation and observed output.
+- File a braid strand per phase when work begins; reference this plan.
+  Snapshot-file changes (Phase 3 likely) get the AGENTS.md snapshot
+  treatment: counts, summary, surprises called out.
+
+## Phase 0 — Measurement harness and baselines
+
+Build the TS-side profiling driver and record the baseline the later
+phases are judged against. No production code changes in this phase
+beyond instrumentation.
+
+**Status: done 2026-10-29 (bd-k8ptcrd9).** Baselines below; later phases
+record their before/after deltas against the "baseline" column.
+
+- [x] Representative fixture: a mid-size qmd project document (order of
+  100 KB source, several images, theme) plus a geometric scaling series
+  (1x/2x/4x/8x body duplication) under a throwaway dir, per the
+  profiling playbook. → generated fixture, ~99 KB at 1x / ~800 KB at 8x.
+- [x] Standalone driver (`/tmp/...`, not committed): overrides
+  `requestIdleCallback`, imports the render-path modules via tsx,
+  simulates a keystroke burst (N edits, each triggering VFS write,
+  format check, render enqueue, token request) against the real WASM.
+  Run with `node --cpu-prof --cpu-prof-interval=200 --import tsx/esm`.
+  Driver notes: a `file:` fetch shim lets the production `initWasm()`
+  run unmodified; an ESM resolve hook maps `wasm-quarto-hub-client`,
+  `/src/wasm-js-bridge/*` and `@quarto/*` workspace src; `?url` asset
+  imports get a string shim; **`fake-indexeddb/auto` is required** —
+  without it the WASM SASS cache bridge has no IndexedDB and every
+  render recompiles the theme, overstating per-keystroke cost ~2x.
+- [x] Gauges on the TS side via the existing
+  `hub-client/src/utils/perfCounters.ts` (already `QUARTO_PERF_STATS`-gated
+  and free when off), following the `perf.<name>` prefix convention:
+  renders started, renders completed, renders discarded as stale, envelope
+  `JSON.parse` bytes, VFS write count and bytes per keystroke, token
+  requests. Implementation note: the module now exists as TWO copies
+  (`hub-client/src/utils/perfCounters.ts` and
+  `ts-packages/preview-runtime/src/perfCounters.ts`, drift-guarded by
+  `perfCounters.sync.test.ts`) sharing one `globalThis` gauge sink —
+  one table per JS realm, immune to the repo's partial
+  `vi.mock('@quarto/preview-runtime')` mocks. New gauges:
+  `render.started` / `render.completed` / `render.wasm-call` (the five
+  leaf render entries), `render.discarded-stale` (ReactPreview
+  lastContentRef guard), `wasm-envelope.parse` (segment + `bytes` extra,
+  all wasmRenderer envelopes + the semantic-tokens envelope),
+  `vfs.write` (segment + `bytes`), `intelligence.semantic-tokens` +
+  `.tokenize`, `parse.qmd-to-ast`, `router.format-check` +
+  `.parse` + `.classify`.
+- [x] Baseline table (below).
+- [x] Characterization tests pinning current behavior for the paths
+  Phases 1-3 will touch:
+  `ts-packages/preview-runtime/src/renderQueue.test.ts` (serial order,
+  no-coalescing pin that Phase 1 will flip, rejection→caller +
+  no-jam); `previewModeClassification.wasm.test.ts` gained three
+  editor-edge pins with DISCOVERED outcomes (an unknown scalar format
+  key is class `neither` — not `dom`; an unterminated front-matter
+  block resolves *successfully* as the default `html`; a broken body
+  still classifies from the front matter); `Q2SandboxedPreviewIframe.test.tsx`
+  pins the exact 13-key UPDATE_AST payload set.
+
+### Phase 0 baseline (2026-10-29, Node driver, release WASM, warm SASS cache)
+
+Per-keystroke main-thread cost by component (phase A: 10 sequential
+simulated keystrokes; numbers are gauge avg_ms):
+
+| Component (gauge) | 1x (~99 KB) | 8x (~800 KB) | scale |
+|---|---|---|---|
+| render pipeline, total (`render.wasm-call`) | 302.1 ms | 2735.0 ms | 9.1x |
+| format-check parse, AST discarded (`parse.qmd-to-ast`) | 79.4 ms | 707.8 ms | 8.9x |
+| semantic tokens, synchronous (`intelligence.semantic-tokens.tokenize`) | 44.1 ms | 385.9 ms | 8.8x |
+| envelope `JSON.parse`, total (`wasm-envelope.parse`) | 14.4 ms / 6 parses / 16.3 MB | 132.3 ms / 6 parses / 134.5 MB | 9.2x / 8.3x bytes |
+| VFS writes (`vfs.write`) | 0.19 ms ×2 writes / 198 KB | 1.18 ms ×2 writes / 1.6 MB | 8.1x bytes |
+| format resolve (`driver.resolve-formats`) | 0.48 ms | 1.5 ms | — |
+| **per-keystroke wall** | **431 ms** | **3410 ms** | 7.9x |
+
+Keystroke-burst render counts (phase B, fire-and-forget, finding 2):
+
+| Scale | Burst | renders started | renders completed | wall |
+|---|---|---|---|---|
+| 1x | 10 | 10 | 10 | 3852 ms |
+| 8x | 5 | 5 | 5 | 17 544 ms |
+
+Every queued render executes — no coalescing. A 10-keystroke burst at
+1x costs ~3.9 s of main-thread work, of which ~9 renders (~2.7 s) are
+pure waste; a coalesced burst would cost ~0.4 s.
+
+Findings read-out:
+
+- **Finding 1 confirmed**: the discarded format-check parse is 18 % of
+  per-keystroke cost at 1x (79 ms of 431 ms), scaling linearly.
+- **Finding 2 confirmed**: burst renders == keystrokes, no coalescing;
+  render CPU dominates every other component combined.
+- **Finding 3/4 confirmed**: 16.3 MB of envelope JSON is parsed on the
+  main thread per keystroke at 1x (134.5 MB at 8x) — the render
+  envelope (transformed + untransformed AST) dwarfs everything else;
+  `parseEnvelope` is ~3.3 % self-time in the 8x CPU profile even
+  against the whole WASM pipeline.
+- **Amplifier confirmed**: exactly 2 full-text VFS writes per keystroke
+  (automerge sync + token alignment), 198 KB / 1.6 MB — plus the third
+  crossing via `resolvePandocFormats` `contentOverride`.
+- **Finding 8 (cold SASS)**: with the cache bridge disabled (no
+  IndexedDB), every render recompiles the theme: render avg 620 ms vs
+  302 ms warm at 1x, plus ~800 small importer envelope reads per render
+  (8040 vs 60 per 10 keystrokes). Warm cache is the right per-keystroke
+  baseline; a theme *change* pays ~320 ms once.
+
+Caveats: Node proxy of the browser path (release WASM, real production
+modules; browser cross-check deferred to phase regressions). Fixture
+body is verbatim-duplicated sections. 8x BURST=10 exceeded a 10-minute
+driver run (memory pressure from 10 concurrent 800-KB renders) — burst
+at 8x recorded with BURST=5. The gauge wiring in the app bundle was
+verified through unit/integration tests and this driver, **not** a live
+browser session.
+
+**Disabled-overhead verification (2026-10-29):** "free when off"
+checked two ways. Microbench (bare vs gauge-wrapped, profiling off,
+min of 5 batches): +0.005 µs/call on 21-B envelopes, +2.1 µs on 130-KB,
++9.4 µs on 16-MB, +2.0 µs on a 100-KB `vfs_add_file` — projected
+≤ ~20 µs per keystroke at 1x, i.e. ~0.005 % of the 431-ms per-keystroke
+wall. End-to-end A/B (same driver, pre-Phase-0 worktree vs instrumented
+checkout, profiling off, interleaved runs): old walls 5580.8–5675.5 ms,
+new 5575.8–5605.2 ms — identical within run-to-run spread.
+
+Follow-up (same day, PR review): the wrappers were restructured to an
+early-return fast path — disabled code is one `perfEnabled()` property
+read, then the exact pre-instrumentation shape (no `try/finally`, no
+gauge calls); the sink object is cached at module scope so
+`perfEnabled()` itself is a single property read. Re-measured: +0.005 µs
+on 21-B, +0.3 µs on 130-KB, and 16-MB/100-KB deltas lost in noise
+(wrapped measures *faster* than bare) — payload-scaled overhead
+eliminated, residual ≈ 5 ns per instrumented call. Both paths are pinned
+by `perfGauges.test.ts` (enabled records, disabled is identical-result
+and records nothing).
+
+## Phase 1 — Latest-wins render coalescing (finding 2)
+
+Smallest change, largest perceived-latency win: stop executing renders
+whose result is already known-stale.
+
+- [ ] Test first: a burst of N content updates while a render is in
+  flight results in at most one additional render executing (the
+  latest); the final preview reflects the last content; a rejection in a
+  superseded render neither jams the queue nor surfaces. Add to the
+  wasmRenderer/ReactPreview test surface at the queue level
+  (`enqueueRender` is the seam, `wasmRenderer.ts:494`).
+- [ ] Implement latest-wins coalescing: when a new render request
+  arrives and earlier requests are still queued (not yet started), drop
+  the superseded ones before they run. Keep first-render-after-idle
+  immediate (no debounce regression — the debounce-free design at
+  `ReactPreview.tsx:768` is deliberate for typing feel; coalescing
+  targets wasted work, not latency of the final render).
+- [ ] Verify the `lastContentRef` stale-result guard
+  (`ReactPreview.tsx:718`) stays as the second line of defense; add a
+  comment noting it now fires only for the in-flight render.
+- [ ] Re-run Phase 0 burst measurement; record renders-executed per
+  burst before/after in the plan.
+
+## Phase 2 — Parse-check-only format classification (finding 1)
+
+The router pays a full parse + AST serialization + main-thread
+`JSON.parse` per keystroke and keeps only `success`.
+
+- [ ] Test first (Rust, `crates/wasm-quarto-hub-client`): a new export
+  (working name `check_document_for_preview`) returns parseability plus
+  the resolved format classification inputs without serializing an AST;
+  assert classification parity with the `parse_qmd_to_ast` +
+  `resolve_pandoc_formats` pair across format fixtures (default html,
+  revealjs, pdf, docx, unknown key, mid-edit syntax error).
+- [ ] Implement the Rust export (TDD: failing test, then implement).
+  Reuse the metadata/front-matter path; no AST writer run.
+- [ ] TS wrapper in `wasmRenderer.ts` + switch
+  `PreviewRouter.tsx:134-168` to it. While here: skip the check entirely
+  when the edit did not touch the front-matter block (cheap string
+  compare of the YAML region), which also removes the
+  `resolvePandocFormats` contentOverride crossing for body-only edits.
+- [ ] Regression: keystroke-type through the format fixtures in the dev
+  server, confirm no mode flicker; record the per-keystroke cost delta
+  from the Phase 0 harness.
+
+## Phase 3 — Single-parse AST transport (finding 3)
+
+Today the AST is a JSON string inside the JSON response envelope: the
+main thread parses the whole envelope to extract a string it reposts to
+the iframe, which parses it again. Since bd-73lilznv (2026-10-26) the
+sandbox iframe is cross-origin by default, so the iframe-side re-parse
+runs in a separate process under site isolation; the main-thread costs
+that remain are the envelope parse itself and the parent-side AST
+consumers enumerated below.
+
+- [ ] Test first: characterization test that the exact `astJson` string
+  posted to the iframe is byte-identical before/after (golden fixtures
+  in `goldenParity.wasm.test.ts` style), and that
+  `untransformed_ast_json` is absent from the envelope unless requested.
+- [ ] Rust side: change the render response so `ast_json` (and
+  `untransformed_ast_json` when requested) travel as separate string
+  out-params or a wasm-bindgen multi-return, not nested inside the
+  parsed envelope. Gate `untransformed_ast_json` behind a flag the TS
+  side sets only during rich-text editing sessions.
+- [ ] TS side: `renderPageInProjectWithAttribution` and siblings stop
+  parsing the AST; `ReactPreview` threads the opaque string through.
+  Parent-side consumers that parse the AST as an object today (found
+  2026-10-27; grep `.ast_json` / `astJson` / `JSON.parse(astJson)`
+  outside iframe plumbing to catch any new ones), with scoping verdicts:
+  - `buildProxyAssetManifest` (`proxyAssetManifest.ts:29`, run by
+    `Q2SandboxedPreviewIframe` on every render) — **move iframe-side**
+    (decided 2026-10-27; sub-items below).
+  - `ReactRenderer.tsx:230` (`extractRenderComponentPaths`) — **stays**.
+    Mounted by `ReactPreview.tsx:896`, the direct (non-sandboxed) React
+    preview, which renders on the main thread by design; the opaque
+    string applies to the sandboxed path only.
+  - `useCursorToSlide.ts:29`, `useSlideThumbnails.tsx:94` — **stay**.
+    Parent-side slide UI (Editor.tsx, OutlinePanel.tsx), independent of
+    preview path. Optional follow-up, out of Phase 3 scope: have these
+    share one parsed object instead of each re-parsing.
+- [ ] Move `proxyAssetManifest.ts` (+ its test) into
+  `hub-client/quarto-hub-sandboxed-preview/src/` and build the manifest
+  in `updateAst`, dropping `assetManifest` from the UPDATE_AST payload
+  and the parent `useMemo`. Rationale: the function is pure in
+  `(astJson, currentFilePath)` — both already cross in UPDATE_AST — does
+  no VFS reads, and its output is consumed only iframe-side via
+  `AssetManifestContext` (`PreviewRoot.tsx:1603`). The move also fixes a
+  reverse-import smell (parent code currently reaches into the iframe
+  package for `pageRelativeUrlForVfsPath` via `../../../../`). The
+  "Image never renders before its manifest entry" guarantee is
+  preserved: the manifest is derived from the same UPDATE_AST string,
+  synchronously before `root.render`. Do NOT unify with q2-preview's
+  `buildAssetManifest` (`Q2PreviewIframe.tsx:314`): that one reads VFS
+  bytes and mints origin-scoped blob URLs, so it must stay parent-side —
+  cross-comment both files.
+- [ ] Optional, only if iframe-side profiling cares: derive the manifest
+  from the iframe's already-parsed AST instead of a second
+  `JSON.parse(astJson)` in `updateAst`. Phase 3's target is the hub main
+  thread; a same-string parse in the iframe process is off that thread
+  and acceptable.
+- [ ] Snapshot audit: this touches render-response shape — expect
+  `.snap` updates; report counts and surprises per AGENTS.md.
+- [ ] Measure: envelope `JSON.parse` bytes and AST parse count per
+  render before/after, recorded in the plan.
+
+## Phase 4 — Binary asset transfer without base64 (finding 6)
+
+Every preview image/font currently does Rust base64 encode, JSON
+envelope, main-thread parse, string postMessage, SW base64 decode.
+
+- [ ] Test first: binary asset round-trip unit test (isBinaryPath +
+  url_response content) and the existing chromium e2e smoke for proxied
+  assets must pass unchanged on byte equality of served content.
+- [ ] Add a binary VFS read that returns `Uint8Array` directly (no
+  base64, no JSON envelope) on the Rust export; keep the text path
+  as-is. (This touches `crates/wasm-quarto-hub-client` — the Conventions
+  WASM rebuild chain applies to this phase.)
+- [ ] Switch `Q2SandboxedPreviewIframe.tsx:182-221` and the SW
+  `requestVFS` response to carry an ArrayBuffer, transferred (not
+  cloned) via the postMessage transfer list. Cross-origin transfer is
+  exercised by default since bd-73lilznv moved the sandbox origin to
+  `quarto-dev.github.io/q2-sandbox` (2026-10-26) — the smoke test now
+  covers the production topology, not a same-origin special case.
+  (ArrayBuffer transfer across origins is spec-supported; verify in the
+  smoke test regardless. Do not reach for SharedArrayBuffer — it
+  requires COOP/COEP headers the sandbox host does not send.)
+- [ ] File the strand as new work (bd-k9tfhsst is closed and did not
+  cover wire format). Measure asset payload bytes and main-thread copies
+  before/after on an image-heavy fixture.
+
+## Phase 5 — Semantic tokens and the envelope tax (findings 4, 5)
+
+Measurement-gated: start only with Phase 0 numbers showing these matter
+relative to Phases 1-3.
+
+- [ ] Baseline first: token-request frequency from Monaco during a
+  keystroke burst, and synchronous tokenize ms per request at fixture
+  scales. Check how Monaco's built-in ~300 ms semantic-tokens debounce
+  interacts — and note that `refreshSemanticTokens()`
+  (`useIntelligenceProviders.ts`, `services/monacoProviders.ts`)
+  deliberately bypasses it via `schedule(0)` on mount and on WASM
+  ready, so forced refreshes pay the synchronous cost immediately.
+- [ ] Candidate fixes, cheapest first: (a) coalesce/debounce token
+  requests in `monacoProviders` so a burst costs one synchronous run;
+  (b) avoid the extra `vfsAddFile` full-text write when the VFS image is
+  already current (dirty-check against the last write); (c) scoped
+  serde-wasm-bindgen or typed returns for the three hottest envelopes
+  (vfs write, semantic tokens, render response) instead of a blanket
+  JSON.parse migration.
+- [ ] Only if (a)-(c) are insufficient: evaluate moving LSP
+  intelligence into a dedicated worker fed by the same sync callbacks
+  (design note first; overlaps Phase 7).
+- [ ] Record what was measured, which option was taken, and why.
+
+## Phase 6 — Attribution off the main thread (finding 7)
+
+Behind a toggle and debounced, so lowest urgency; still O(document
+history) when on.
+
+- [ ] Baseline: cold-build and incremental-update ms at history scales
+  (the 2026-04-23 attribution profiling session is the template;
+  `requestIdleCallback` override mandatory).
+- [ ] If numbers warrant: move run-list replay and char-to-byte mapping
+  into a worker (Automerge handles are transferable to a worker context
+  via doc export/import; design the wire shape first), keeping the
+  hook's payload contract unchanged.
+- [ ] Regression: attribution overlay output identical on fixtures with
+  mixed author/actor history; toggle on/off flicker-free.
+
+## Phase 7 — Render-worker architecture evaluation (design only)
+
+Not an implementation phase: produce a design note weighing the move of
+`wasm-quarto-hub-client` (pipeline + VFS) into a dedicated worker, which
+would take the whole render pipeline off the main thread.
+
+- [ ] Design note in `claude-notes/designs/`: sync client pushes file
+  changes into the worker VFS; render requests/responses as RPC;
+  constraints enumerated (the SASS bridge calls JS synchronously from
+  Rust; Monaco semantic tokens want synchronous answers; the serial
+  queue moves with the VFS; memory duplication of the VFS during any
+  transition).
+- [ ] Decision record: adopt, defer, or reject — with the Phase 0/5
+  numbers as the evidence base. If adopted, spawn an epic and re-plan.
diff --git a/hub-client/changelog.md b/hub-client/changelog.md
index fe417ded9f..1b80f8c335 100644
--- a/hub-client/changelog.md
+++ b/hub-client/changelog.md
@@ -23,6 +23,10 @@ WASM rebuild is needed for a changelog-only edit.
 
 -->
 
+### 2026-10-29
+
+- [`d3b294042`](https://github.com/quarto-dev/q2/commits/d3b294042): Added opt-in performance gauges to the preview render path (renders, envelope parsing, VFS writes, semantic tokens), enabled with QUARTO_PERF_STATS=1, localStorage quartoPerfStats=1, or a perf=1 URL parameter and dumped with window.quartoPerfDump() — the measurement baseline for upcoming typing-latency work.
+
 ### 2026-10-27
 
 - [`6a4a0d214`](https://github.com/quarto-dev/q2/commits/6a4a0d214): Fixed the "Connection lost - working offline" banner never clearing after the connection is restored, and the share dialog once again selects the link text shortly after it opens (the select-all had silently stopped firing).
diff --git a/hub-client/src/components/render/PreviewRouter.tsx b/hub-client/src/components/render/PreviewRouter.tsx
index 1d3b5d271a..81917daf6a 100644
--- a/hub-client/src/components/render/PreviewRouter.tsx
+++ b/hub-client/src/components/render/PreviewRouter.tsx
@@ -13,6 +13,7 @@ import { DownloadOnlyView, NeitherView } from './DownloadOnlyViews';
 import PdfPreviewPane from './PdfPreviewPane';
 import { formatByKey, menuFormats, pdfPreviewAvailable, withOutputExt } from '../../pandoc/downloadService';
 import { useDownloadAs } from '../../pandoc/useDownloadAs';
+import { perfNow, perfSegment } from '../../utils/perfCounters';
 
 interface PreviewRouterProps {
   content: string;
@@ -133,6 +134,7 @@ export default function PreviewRouter(props: PreviewRouterProps) {
   // Check the format whenever content changes
   useEffect(() => {
     async function checkFormat() {
+      const fxT0 = perfNow();
       try {
         // Skip format check if WASM isn't ready yet (will retry when it is)
         if (!isWasmReady()) {
@@ -141,7 +143,11 @@ export default function PreviewRouter(props: PreviewRouterProps) {
 
         // Parse the QMD only to learn whether it parses: a mid-edit syntax error keeps the
         // previous mode. The format itself comes from the Rust resolver.
+        const parseT0 = perfNow();
         const result = await parseQmdToAst(props.content);
+        // Finding 1 of the render-path plan: this per-keystroke full parse's
+        // AST is discarded — only `success` is read.
+        perfSegment('router.format-check.parse', parseT0);
         const deps = {
           // The editor buffer, not the VFS copy: it can lag a keystroke behind.
           resolve: (path: string) => (isWasmReady() ? resolvePandocFormats(path, { contentOverride: props.content }) : null),
@@ -150,7 +156,9 @@ export default function PreviewRouter(props: PreviewRouterProps) {
         };
         // A format the parser's metadata merge does not know (`latex`) fails the parse. The
         // resolver still classifies it; a failed parse of a previewable document changes nothing.
+        const classifyT0 = perfNow();
         const mode = classifyPreviewMode(props.currentFile?.path, deps);
+        perfSegment('router.format-check.classify', classifyT0);
         if (result.success || mode.mode === 'download' || mode.mode === 'pdf' || mode.mode === 'neither') {
           const format = mode.mode === 'react' ? mode.format : null;
           // Keep the same object when nothing changed, so a keystroke does not re-render consumers.
@@ -162,6 +170,7 @@ export default function PreviewRouter(props: PreviewRouterProps) {
       } catch (err) {
         console.error('[PreviewRouter] Error checking format:', err);
       } finally {
+        perfSegment('router.format-check', fxT0);
         setCheckedPath(props.currentFile?.path);
       }
     }
diff --git a/hub-client/src/components/render/ReactPreview.tsx b/hub-client/src/components/render/ReactPreview.tsx
index 85922a39f4..8a3fa4a2ec 100644
--- a/hub-client/src/components/render/ReactPreview.tsx
+++ b/hub-client/src/components/render/ReactPreview.tsx
@@ -21,7 +21,7 @@ import { stripAnsi } from '@quarto/preview-renderer/utils/stripAnsi';
 import { PreviewErrorOverlay } from '@quarto/preview-renderer/overlays/PreviewErrorOverlay';
 import { usePreference } from '../../hooks/usePreference';
 import { useScrollSync } from '../../hooks/useScrollSync';
-import { perfNow, perfSegment } from '../../utils/perfCounters';
+import { perfNow, perfSegment, perfMark } from '../../utils/perfCounters';
 import type { Q2PreviewIframeHandle } from '@quarto/preview-renderer/iframe/Q2PreviewIframe';
 import ReactRenderer from './ReactRenderer';
 
@@ -715,7 +715,13 @@ export default function ReactPreview({
       attributionJson: attributionPayload,
       captureGzJson: captureBytes,
     });
-    if (qmdContent !== lastContentRef.current) return;
+    if (qmdContent !== lastContentRef.current) {
+      // A newer edit superseded this render while it was in flight; drop
+      // the stale result (finding 2's second line of defense — the wasted
+      // work already happened, which `render.started/completed` counts).
+      perfMark('render.discarded-stale');
+      return;
+    }
 
     // Update diagnostics
     onDiagnosticsChange(result.diagnostics);
diff --git a/hub-client/src/components/render/previewModeClassification.wasm.test.ts b/hub-client/src/components/render/previewModeClassification.wasm.test.ts
index 3ab8532f53..05030c0bd7 100644
--- a/hub-client/src/components/render/previewModeClassification.wasm.test.ts
+++ b/hub-client/src/components/render/previewModeClassification.wasm.test.ts
@@ -124,4 +124,44 @@ describe('classifyPreviewMode against the real wasm', () => {
       expect(modeOfBuffer('/project/a.qmd', '# A\n')).toEqual({ mode: 'react', format: 'q2-preview' });
     });
   });
+
+  // Phase 0 characterization for the main-thread render-path plan
+  // (bd-k8ptcrd9): the outcomes PreviewRouter depends on per keystroke,
+  // pinned so Phase 2's parse-check-only classification can assert parity.
+  // Expected values below are DISCOVERED behavior (probed against the real
+  // resolver 2026-10-29), not the plan's guesses.
+  describe('editor-edge inputs (Phase 0 characterization, bd-k8ptcrd9)', () => {
+    it('an unknown format key classifies as neither (not dom)', async () => {
+      // The resolver succeeds with class 'neither' for a scalar format key
+      // it does not know. (Extension formats such as `acm-html` are a
+      // different path — class 'preview' → dom — pinned in
+      // getQ2Format.test.ts.)
+      expect(await modeOf('format: some-unknown-format')).toEqual({
+        mode: 'neither',
+        formatKey: 'some-unknown-format',
+      });
+    });
+
+    it('a mid-edit broken body still classifies from the front matter', async () => {
+      // Unclosed code fence mid-typing: the resolver reads the YAML, so the
+      // mode must not flicker while the body is transiently unparsable.
+      const buffer = '---\nformat: revealjs\n---\n\n# Deck\n\n```{r}\n';
+      expect(modeOfBuffer('/project/doc.qmd', buffer)).toEqual({ mode: 'react', format: 'revealjs' });
+    });
+
+    it('an unterminated front-matter block still resolves — as the default html', async () => {
+      // Mid-edit YAML: no closing fence. The resolver SUCCEEDS but does not
+      // see the unterminated block's `format:` key — it falls back to the
+      // default (html → q2-preview). So an in-progress edit of the format
+      // key itself can transiently reclassify to q2-preview; PreviewRouter's
+      // parse guard is what suppresses the flicker (component logic).
+      const buffer = '---\ntitle: T\nformat: revealjs\n';
+      const resolved = JSON.parse(
+        wasm.resolve_pandoc_formats('/project/doc.qmd', { contentOverride: buffer }),
+      ) as ResolvePandocFormatsResponse;
+      expect(resolved.success).toBe(true);
+      expect(resolved.key).toBe('html');
+      expect(modeOfBuffer('/project/doc.qmd', buffer)).toEqual({ mode: 'react', format: 'q2-preview' });
+    });
+  });
 });
diff --git a/hub-client/src/components/render/q2-sandboxed-preview/Q2SandboxedPreviewIframe.test.tsx b/hub-client/src/components/render/q2-sandboxed-preview/Q2SandboxedPreviewIframe.test.tsx
index fbf770f454..fe17f337f7 100644
--- a/hub-client/src/components/render/q2-sandboxed-preview/Q2SandboxedPreviewIframe.test.tsx
+++ b/hub-client/src/components/render/q2-sandboxed-preview/Q2SandboxedPreviewIframe.test.tsx
@@ -317,6 +317,38 @@ describe('Q2SandboxedPreviewIframe', () => {
     });
   });
 
+  it('pins the exact UPDATE_AST payload key set (Phase 0 characterization, bd-k8ptcrd9)', async () => {
+    // The main-thread render-path plan's Phase 3 drops `assetManifest` from
+    // this payload; the exact key set is the contract it changes, so pin it.
+    const { iframe, postMessage } = renderIframe();
+    signalIframeReady(iframe);
+
+    await waitFor(() => {
+      const updateAst = postMessage.mock.calls.find(
+        ([msg]) => (msg as { type?: string }).type === 'UPDATE_AST',
+      );
+      expect(updateAst).toBeDefined();
+      const payload = (updateAst![0] as { payload: Record<string, unknown> }).payload;
+      expect(Object.keys(payload).sort()).toEqual(
+        [
+          'astJson',
+          'currentFilePath',
+          'assetManifest',
+          'projectFilePaths',
+          'pendingAnchor',
+          'pendingAnchorEpoch',
+          'renderedContent',
+          'untransformedAstJson',
+          'currentActor',
+          'commentsMode',
+          'unlockNestingCursor',
+          'richText',
+          'nestedEditBuffers',
+        ].sort(),
+      );
+    });
+  });
+
   it('ignores messages whose source is not the sandboxed iframe window', async () => {
     // Any window can postMessage the parent; only the embedded frame's
     // own contentWindow may drive this component. A forged SET_AST from
diff --git a/hub-client/src/services/intelligenceService.ts b/hub-client/src/services/intelligenceService.ts
index 0dff0fcdf1..fcb859386b 100644
--- a/hub-client/src/services/intelligenceService.ts
+++ b/hub-client/src/services/intelligenceService.ts
@@ -23,6 +23,7 @@ import type {
 } from '@quarto/preview-renderer/types/intelligence';
 import { initWasm, vfsAddFile } from '@quarto/preview-runtime';
 import { isSourceFile } from '@quarto/preview-renderer/types/project';
+import { perfEnabled, perfSegment, perfMark, perfExtra } from '../utils/perfCounters';
 
 // Re-export types for convenience
 export type { Symbol, Diagnostic, FoldingRange, DocumentAnalysis, SemanticToken } from '@quarto/preview-renderer/types/intelligence';
@@ -237,7 +238,14 @@ export async function getSemanticTokens(path: string): Promise<SemanticToken[]>
 
   try {
     const wasm = await getWasm();
-    return parseSemanticTokensResponse(wasm.lsp_get_semantic_tokens(path));
+    if (!perfEnabled()) {
+      return parseSemanticTokensResponse(wasm.lsp_get_semantic_tokens(path));
+    }
+    perfMark('intelligence.semantic-tokens');
+    const t0 = performance.now();
+    const tokens = parseSemanticTokensResponse(wasm.lsp_get_semantic_tokens(path));
+    perfSegment('intelligence.semantic-tokens.tokenize', t0);
+    return tokens;
   } catch (e) {
     console.warn('Failed to get semantic tokens:', e);
     return [];
@@ -270,10 +278,20 @@ export async function getSemanticTokensForContent(
 
   try {
     const wasm = await getWasm();
+    if (!perfEnabled()) {
+      // Synchronous from here: align the VFS image to the model, then
+      // tokenise it in the same macrotask (see the enabled path below).
+      vfsAddFile(path, content);
+      return parseSemanticTokensResponse(wasm.lsp_get_semantic_tokens(path));
+    }
+    perfMark('intelligence.semantic-tokens');
+    const t0 = performance.now();
     // Synchronous from here: align the VFS image to the model, then tokenise it
     // in the same macrotask so nothing can re-drift the two in between.
     vfsAddFile(path, content);
-    return parseSemanticTokensResponse(wasm.lsp_get_semantic_tokens(path));
+    const tokens = parseSemanticTokensResponse(wasm.lsp_get_semantic_tokens(path));
+    perfSegment('intelligence.semantic-tokens.tokenize', t0);
+    return tokens;
   } catch (e) {
     console.warn('Failed to get semantic tokens:', e);
     return [];
@@ -282,7 +300,20 @@ export async function getSemanticTokensForContent(
 
 /** Decode a `lsp_get_semantic_tokens` JSON envelope, collapsing errors to `[]`. */
 function parseSemanticTokensResponse(json: string): SemanticToken[] {
+  // Finding 5's envelope tax on the token path folds into the shared
+  // `wasm-envelope.parse` gauge. `bytes` is String.length (UTF-16 units).
+  if (!perfEnabled()) {
+    const result: LspSemanticTokensResponse = JSON.parse(json);
+    if (result.success) {
+      return result.tokens ?? [];
+    }
+    console.warn('Failed to get semantic tokens:', result.error);
+    return [];
+  }
+  const t0 = performance.now();
   const result: LspSemanticTokensResponse = JSON.parse(json);
+  perfSegment('wasm-envelope.parse', t0);
+  perfExtra('wasm-envelope.parse', 'bytes', json.length);
   if (result.success) {
     return result.tokens ?? [];
   }
diff --git a/hub-client/src/utils/perfCounters.sync.test.ts b/hub-client/src/utils/perfCounters.sync.test.ts
new file mode 100644
index 0000000000..1f54f72781
--- /dev/null
+++ b/hub-client/src/utils/perfCounters.sync.test.ts
@@ -0,0 +1,29 @@
+/**
+ * Drift guard for the two perfCounters copies (see the files' own header
+ * comments for why two exist, bd-k8ptcrd9). Asserts the implementations
+ * are byte-identical after stripping each file's leading doc comment —
+ * the only intended difference is the header naming each copy's location.
+ */
+
+import { describe, it, expect } from 'vitest';
+import { readFileSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+function stripHeaderComment(source: string): string {
+  const idx = source.indexOf('interface Gauge');
+  if (idx < 0) throw new Error('perfCounters: "interface Gauge" not found');
+  return source.slice(idx);
+}
+
+describe('perfCounters copies', () => {
+  it('hub-client and preview-runtime implementations stay byte-identical', () => {
+    const here = dirname(fileURLToPath(import.meta.url));
+    const hubCopy = readFileSync(join(here, 'perfCounters.ts'), 'utf8');
+    const packageCopy = readFileSync(
+      join(here, '../../../ts-packages/preview-runtime/src/perfCounters.ts'),
+      'utf8',
+    );
+    expect(stripHeaderComment(packageCopy)).toBe(stripHeaderComment(hubCopy));
+  });
+});
diff --git a/hub-client/src/utils/perfCounters.ts b/hub-client/src/utils/perfCounters.ts
index 7c4d4b2569..e02e8e55d1 100644
--- a/hub-client/src/utils/perfCounters.ts
+++ b/hub-client/src/utils/perfCounters.ts
@@ -1,8 +1,21 @@
 /**
- * Browser-side perf counters — the hub-client counterpart of the Rust
+ * Perf counters — the hub-client / preview-runtime counterpart of the Rust
  * `QUARTO_PERF_STATS` gauges (see
  * claude-notes/instructions/performance-profiling.md).
  *
+ * TWO COPIES OF THIS FILE EXIST and must be kept in sync:
+ *   - hub-client/src/utils/perfCounters.ts   (this one)
+ *   - ts-packages/preview-runtime/src/perfCounters.ts
+ * The duplication is deliberate (bd-k8ptcrd9): hub-client app code cannot
+ * import the package copy through the package boundary without breaking
+ * the repo's many PARTIAL `vi.mock('@quarto/preview-runtime', ...)` test
+ * mocks (any export they don't stub goes undefined). Both copies delegate
+ * all state to a `globalThis` sink, so there is exactly ONE gauge table
+ * per JS realm (hub app window, preview iframe, Node driver) no matter
+ * how many bundle copies exist — `perfDump` from either copy prints
+ * everything, and gauges recorded through one copy aggregate with the
+ * other. When editing one copy, edit the other identically.
+ *
  * Enablement is read once at startup, from the first match:
  *
  * 1. `process.env.QUARTO_PERF_STATS === '1'` (Node drivers; same variable
@@ -18,7 +31,7 @@
  * [key=value ...]`, so a session log can be grepped with `^perf.` and
  * parsed column-wise.
  *
- * Gauges are process-global by name; counters are never reset
+ * Gauges are realm-global by name; counters are never reset
  * automatically (dump is cumulative).
  */
 
@@ -33,7 +46,36 @@ interface Gauge {
   extras: Map<string, number>;
 }
 
-const gauges = new Map<string, Gauge>();
+/** Shared state across every copy of this module loaded in one realm. */
+interface PerfSink {
+  gauges: Map<string, Gauge>;
+  /** Test override: true/false forces recording on/off for all copies. */
+  override: boolean | null;
+  /** Guards the 5s browser interval so two copies don't double-dump. */
+  intervalInstalled: boolean;
+}
+
+const SINK_KEY = '__quartoPerfSink';
+
+function sink(): PerfSink {
+  const g = globalThis as { __quartoPerfSink?: PerfSink };
+  let s = g[SINK_KEY];
+  if (!s) {
+    s = { gauges: new Map(), override: null, intervalInstalled: false };
+    g[SINK_KEY] = s;
+  }
+  return s;
+}
+
+// Cache the sink once: the disabled fast path reads this object's fields
+// directly (one property read) instead of re-resolving globalThis per call.
+const S = sink();
+
+// `process` is not declared by every consumer's tsconfig; reach it
+// through globalThis so Node drivers and browsers both compile and run.
+const nodeProcess = (
+  globalThis as { process?: { env?: Record<string, string | undefined> } }
+).process;
 
 /**
  * True when the flag came from the Node env var rather than a browser
@@ -41,8 +83,7 @@ const gauges = new Map<string, Gauge>();
  * (interval dump, window handle) that would pollute driver output and
  * keep the process alive.
  */
-const enabledByNodeEnv =
-  typeof process !== 'undefined' && process.env?.QUARTO_PERF_STATS === '1';
+const enabledByNodeEnv = nodeProcess?.env?.QUARTO_PERF_STATS === '1';
 
 function readEnabledFlag(): boolean {
   // Node driver path (tsx scripts under scripts/perf or throwaway
@@ -69,9 +110,9 @@ function readEnabledFlag(): boolean {
 
 let enabled = readEnabledFlag();
 
-/** Whether perf recording is on. Cheap: one cached boolean. */
+/** Whether perf recording is on. One property read on the cached sink. */
 export function perfEnabled(): boolean {
-  return enabled;
+  return S.override ?? enabled;
 }
 
 /**
@@ -79,14 +120,14 @@ export function perfEnabled(): boolean {
  * the result straight to {@link perfSegment} without branching.
  */
 export function perfNow(): number {
-  return enabled ? performance.now() : 0;
+  return S.override ?? enabled ? performance.now() : 0;
 }
 
 function gauge(name: string): Gauge {
-  let g = gauges.get(name);
+  let g = S.gauges.get(name);
   if (!g) {
     g = { runs: 0, totalMs: 0, maxMs: 0, extras: new Map() };
-    gauges.set(name, g);
+    S.gauges.set(name, g);
   }
   return g;
 }
@@ -96,7 +137,7 @@ function gauge(name: string): Gauge {
  * No-op when disabled. Omit `endMs` to measure up to the call.
  */
 export function perfSegment(name: string, startMs: number, endMs?: number): void {
-  if (!enabled) return;
+  if (!perfEnabled()) return;
   const end = endMs ?? performance.now();
   const ms = end - startMs;
   const g = gauge(name);
@@ -107,7 +148,7 @@ export function perfSegment(name: string, startMs: number, endMs?: number): void
 
 /** Increment the run counter of `name` without a duration (e.g. early-outs). */
 export function perfMark(name: string): void {
-  if (!enabled) return;
+  if (!perfEnabled()) return;
   gauge(name).runs += 1;
 }
 
@@ -116,7 +157,7 @@ export function perfMark(name: string): void {
  * `perfExtra('effect.project-search', 'changed', changed)`).
  */
 export function perfExtra(name: string, key: string, n = 1): void {
-  if (!enabled) return;
+  if (!perfEnabled()) return;
   const g = gauge(name);
   g.extras.set(key, (g.extras.get(key) ?? 0) + n);
 }
@@ -137,28 +178,35 @@ function formatGauge(name: string, g: Gauge): string {
 
 /** Log all recorded gauges, sorted by name. Cumulative; does not reset. */
 export function perfDump(): void {
-  for (const [name, g] of [...gauges.entries()].sort()) {
+  for (const [name, g] of [...S.gauges.entries()].sort()) {
     console.log(formatGauge(name, g));
   }
 }
 
 /** Clear all gauges. Exposed for drivers and tests. */
 export function perfReset(): void {
-  gauges.clear();
+  S.gauges.clear();
 }
 
 /**
  * Test-only override: force recording on (`true`), off (`false`), or
- * back to the startup flag (`null`). Kept out of the flag reader so the
- * production path stays a single cached boolean.
+ * back to the startup flag (`null`) — for EVERY copy of this module in
+ * the realm (the override lives on the shared sink).
  */
 export function setPerfOverrideForTests(value: boolean | null): void {
-  enabled = value ?? readEnabledFlag();
+  S.override = value;
+  if (value === null) {
+    enabled = readEnabledFlag();
+  }
 }
 
-// Browser conveniences: periodic dump + manual dump handle. Skipped in
-// Node drivers (they call perfDump themselves at checkpoints).
-if (enabled && !enabledByNodeEnv && typeof window !== 'undefined') {
-  window.setInterval(perfDump, 5000);
-  (window as unknown as { quartoPerfDump: () => void }).quartoPerfDump = perfDump;
+// Browser conveniences: periodic dump + manual dump handle, installed at
+// most once per realm (guarded via the sink). Skipped in Node drivers
+// (they call perfDump themselves at checkpoints).
+{
+  if (perfEnabled() && !enabledByNodeEnv && !S.intervalInstalled && typeof window !== 'undefined') {
+    S.intervalInstalled = true;
+    window.setInterval(perfDump, 5000);
+    (window as unknown as { quartoPerfDump: () => void }).quartoPerfDump = perfDump;
+  }
 }
diff --git a/ts-packages/preview-runtime/src/index.ts b/ts-packages/preview-runtime/src/index.ts
index 027bb625e3..3e5991702b 100644
--- a/ts-packages/preview-runtime/src/index.ts
+++ b/ts-packages/preview-runtime/src/index.ts
@@ -16,3 +16,4 @@ export * from './wasmRenderer';
 export * from './automergeSync';
 export * from './diffToEditorChanges';
 export * from './pipelineKind';
+export * from './perfCounters';
diff --git a/ts-packages/preview-runtime/src/perfCounters.ts b/ts-packages/preview-runtime/src/perfCounters.ts
new file mode 100644
index 0000000000..b10cabbd6b
--- /dev/null
+++ b/ts-packages/preview-runtime/src/perfCounters.ts
@@ -0,0 +1,212 @@
+/**
+ * Perf counters — the hub-client / preview-runtime counterpart of the Rust
+ * `QUARTO_PERF_STATS` gauges (see
+ * claude-notes/instructions/performance-profiling.md).
+ *
+ * TWO COPIES OF THIS FILE EXIST and must be kept in sync:
+ *   - ts-packages/preview-runtime/src/perfCounters.ts   (this one)
+ *   - hub-client/src/utils/perfCounters.ts
+ * The duplication is deliberate (bd-k8ptcrd9): hub-client app code cannot
+ * import this package's copy through the package boundary without breaking
+ * the repo's many PARTIAL `vi.mock('@quarto/preview-runtime', ...)` test
+ * mocks (any export they don't stub goes undefined). Both copies delegate
+ * all state to a `globalThis` sink, so there is exactly ONE gauge table
+ * per JS realm (hub app window, preview iframe, Node driver) no matter
+ * how many bundle copies exist — `perfDump` from either copy prints
+ * everything, and gauges recorded through one copy aggregate with the
+ * other. When editing one copy, edit the other identically.
+ *
+ * Enablement is read once at startup, from the first match:
+ *
+ * 1. `process.env.QUARTO_PERF_STATS === '1'` (Node drivers; same variable
+ *    the Rust gauges use),
+ * 2. `localStorage['quartoPerfStats'] === '1'`, or
+ * 3. a `perf=1` URL query parameter.
+ *
+ * When disabled, every entry point is a single predictable branch on a
+ * cached boolean — effectively free. When enabled, counters and timings
+ * accumulate in memory and are dumped to the console (a) every 5 seconds
+ * and (b) on demand via `window.quartoPerfDump()`. Output lines are
+ * machine-readable: `perf.<gauge> runs=N total_ms=X avg_ms=Y max_ms=Z
+ * [key=value ...]`, so a session log can be grepped with `^perf.` and
+ * parsed column-wise.
+ *
+ * Gauges are realm-global by name; counters are never reset
+ * automatically (dump is cumulative).
+ */
+
+interface Gauge {
+  /** Number of recorded events (segments measured or marks counted). */
+  runs: number;
+  /** Accumulated milliseconds across all recorded segments. */
+  totalMs: number;
+  /** Largest single segment. */
+  maxMs: number;
+  /** Free-form cumulative counters (e.g. files=, changed=). */
+  extras: Map<string, number>;
+}
+
+/** Shared state across every copy of this module loaded in one realm. */
+interface PerfSink {
+  gauges: Map<string, Gauge>;
+  /** Test override: true/false forces recording on/off for all copies. */
+  override: boolean | null;
+  /** Guards the 5s browser interval so two copies don't double-dump. */
+  intervalInstalled: boolean;
+}
+
+const SINK_KEY = '__quartoPerfSink';
+
+function sink(): PerfSink {
+  const g = globalThis as { __quartoPerfSink?: PerfSink };
+  let s = g[SINK_KEY];
+  if (!s) {
+    s = { gauges: new Map(), override: null, intervalInstalled: false };
+    g[SINK_KEY] = s;
+  }
+  return s;
+}
+
+// Cache the sink once: the disabled fast path reads this object's fields
+// directly (one property read) instead of re-resolving globalThis per call.
+const S = sink();
+
+// `process` is not declared by every consumer's tsconfig; reach it
+// through globalThis so Node drivers and browsers both compile and run.
+const nodeProcess = (
+  globalThis as { process?: { env?: Record<string, string | undefined> } }
+).process;
+
+/**
+ * True when the flag came from the Node env var rather than a browser
+ * source. Node drivers use this to suppress the browser conveniences
+ * (interval dump, window handle) that would pollute driver output and
+ * keep the process alive.
+ */
+const enabledByNodeEnv = nodeProcess?.env?.QUARTO_PERF_STATS === '1';
+
+function readEnabledFlag(): boolean {
+  // Node driver path (tsx scripts under scripts/perf or throwaway
+  // /tmp drivers) — same variable as the Rust gauges.
+  if (enabledByNodeEnv) {
+    return true;
+  }
+  try {
+    if (globalThis.localStorage?.getItem('quartoPerfStats') === '1') {
+      return true;
+    }
+  } catch {
+    // localStorage can throw (privacy modes); fall through to URL param.
+  }
+  try {
+    if (typeof window !== 'undefined') {
+      return new URLSearchParams(window.location.search).get('perf') === '1';
+    }
+  } catch {
+    // Malformed location in exotic embeds; treat as disabled.
+  }
+  return false;
+}
+
+let enabled = readEnabledFlag();
+
+/** Whether perf recording is on. One property read on the cached sink. */
+export function perfEnabled(): boolean {
+  return S.override ?? enabled;
+}
+
+/**
+ * Start a timing sample. Returns 0 when disabled, so callers can pass
+ * the result straight to {@link perfSegment} without branching.
+ */
+export function perfNow(): number {
+  return S.override ?? enabled ? performance.now() : 0;
+}
+
+function gauge(name: string): Gauge {
+  let g = S.gauges.get(name);
+  if (!g) {
+    g = { runs: 0, totalMs: 0, maxMs: 0, extras: new Map() };
+    S.gauges.set(name, g);
+  }
+  return g;
+}
+
+/**
+ * Record one timed segment for `name`: `endMs - startMs` milliseconds.
+ * No-op when disabled. Omit `endMs` to measure up to the call.
+ */
+export function perfSegment(name: string, startMs: number, endMs?: number): void {
+  if (!perfEnabled()) return;
+  const end = endMs ?? performance.now();
+  const ms = end - startMs;
+  const g = gauge(name);
+  g.runs += 1;
+  g.totalMs += ms;
+  if (ms > g.maxMs) g.maxMs = ms;
+}
+
+/** Increment the run counter of `name` without a duration (e.g. early-outs). */
+export function perfMark(name: string): void {
+  if (!perfEnabled()) return;
+  gauge(name).runs += 1;
+}
+
+/**
+ * Add `n` to a named cumulative extra on the gauge (e.g.
+ * `perfExtra('effect.project-search', 'changed', changed)`).
+ */
+export function perfExtra(name: string, key: string, n = 1): void {
+  if (!perfEnabled()) return;
+  const g = gauge(name);
+  g.extras.set(key, (g.extras.get(key) ?? 0) + n);
+}
+
+/** Format one gauge line (also the unit-tested shape). */
+function formatGauge(name: string, g: Gauge): string {
+  const parts = [`perf.${name}`, `runs=${g.runs}`];
+  if (g.totalMs > 0) {
+    parts.push(`total_ms=${g.totalMs.toFixed(3)}`);
+    parts.push(`avg_ms=${(g.totalMs / g.runs).toFixed(3)}`);
+    parts.push(`max_ms=${g.maxMs.toFixed(3)}`);
+  }
+  for (const [k, v] of [...g.extras.entries()].sort()) {
+    parts.push(`${k}=${v}`);
+  }
+  return parts.join(' ');
+}
+
+/** Log all recorded gauges, sorted by name. Cumulative; does not reset. */
+export function perfDump(): void {
+  for (const [name, g] of [...S.gauges.entries()].sort()) {
+    console.log(formatGauge(name, g));
+  }
+}
+
+/** Clear all gauges. Exposed for drivers and tests. */
+export function perfReset(): void {
+  S.gauges.clear();
+}
+
+/**
+ * Test-only override: force recording on (`true`), off (`false`), or
+ * back to the startup flag (`null`) — for EVERY copy of this module in
+ * the realm (the override lives on the shared sink).
+ */
+export function setPerfOverrideForTests(value: boolean | null): void {
+  S.override = value;
+  if (value === null) {
+    enabled = readEnabledFlag();
+  }
+}
+
+// Browser conveniences: periodic dump + manual dump handle, installed at
+// most once per realm (guarded via the sink). Skipped in Node drivers
+// (they call perfDump themselves at checkpoints).
+{
+  if (perfEnabled() && !enabledByNodeEnv && !S.intervalInstalled && typeof window !== 'undefined') {
+    S.intervalInstalled = true;
+    window.setInterval(perfDump, 5000);
+    (window as unknown as { quartoPerfDump: () => void }).quartoPerfDump = perfDump;
+  }
+}
diff --git a/ts-packages/preview-runtime/src/perfGauges.test.ts b/ts-packages/preview-runtime/src/perfGauges.test.ts
new file mode 100644
index 0000000000..e6a5626e00
--- /dev/null
+++ b/ts-packages/preview-runtime/src/perfGauges.test.ts
@@ -0,0 +1,56 @@
+/**
+ * Enabled/disabled parity for the render-path gauges (bd-k8ptcrd9).
+ * With recording ON, the instrumented wrappers record their gauges; with
+ * it OFF they return identical results and record nothing. This guards
+ * the early-return fast paths against two failure modes: a fast path
+ * that diverges from the measured path's results, and a fast path that
+ * still pays for the gauges.
+ */
+
+import { describe, it, expect, vi, beforeAll, afterEach } from 'vitest';
+import { _setWasmModuleForTest, vfsAddFile, renderQmd } from './wasmRenderer';
+import { perfDump, perfReset, setPerfOverrideForTests } from './perfCounters';
+
+const wasmStub = {
+  vfs_add_file: vi.fn(() => JSON.stringify({ success: true })),
+  render_qmd: vi.fn(async (_path: string) => JSON.stringify({ success: true, html: '<p>x</p>' })),
+};
+
+beforeAll(() => {
+  _setWasmModuleForTest(wasmStub);
+});
+
+afterEach(() => {
+  setPerfOverrideForTests(null);
+  perfReset();
+  vi.restoreAllMocks();
+});
+
+function dumpLines(): string[] {
+  const log = vi.spyOn(console, 'log').mockImplementation(() => {});
+  perfDump();
+  return log.mock.calls.map((c) => c[0] as string);
+}
+
+describe('gauge fast paths', () => {
+  it('records gauges when enabled', async () => {
+    setPerfOverrideForTests(true);
+    vfsAddFile('a.qmd', 'hello');
+    await renderQmd('a.qmd');
+    const lines = dumpLines().join('\n');
+    expect(lines).toContain('perf.vfs.write runs=1 total_ms=');
+    expect(lines).toContain('bytes=5');
+    expect(lines).toContain('perf.render.started runs=1');
+    expect(lines).toContain('perf.render.completed runs=1');
+    expect(lines).toContain('perf.wasm-envelope.parse runs=2');
+  });
+
+  it('returns identical results and records nothing when disabled', async () => {
+    setPerfOverrideForTests(false);
+    const write = vfsAddFile('a.qmd', 'hello');
+    const render = await renderQmd('a.qmd');
+    expect(write).toEqual({ success: true });
+    expect(render).toEqual({ success: true, html: '<p>x</p>' });
+    expect(dumpLines()).toEqual([]);
+  });
+});
diff --git a/ts-packages/preview-runtime/src/renderQueue.test.ts b/ts-packages/preview-runtime/src/renderQueue.test.ts
new file mode 100644
index 0000000000..c3b2dd9743
--- /dev/null
+++ b/ts-packages/preview-runtime/src/renderQueue.test.ts
@@ -0,0 +1,138 @@
+/**
+ * Characterization tests for the serial render queue in wasmRenderer.ts
+ * (`enqueueRender`), written for bd-k8ptcrd9 (the 2026-10-24 main-thread
+ * render-path plan, Phase 0). These pin the CURRENT behavior that plan's
+ * phases build on:
+ *
+ * - renders execute strictly one at a time, in call order;
+ * - EVERY enqueued render executes — there is no coalescing of queued
+ *   work (the plan's finding 2: a burst of N updates runs N full
+ *   pipelines). Phase 1 (latest-wins coalescing) will deliberately
+ *   change this pin; update it when that lands;
+ * - a rejected render reaches its own caller as a `{ success: false }`
+ *   result and does not jam the queue for later renders.
+ */
+
+import { describe, it, expect, vi, beforeAll, beforeEach } from 'vitest';
+
+const { state, mkDeferred } = vi.hoisted(() => {
+  interface Deferred<T> {
+    promise: Promise<T>;
+    resolve: (value: T) => void;
+    reject: (reason?: unknown) => void;
+  }
+  function mkDeferred<T>(): Deferred<T> {
+    let resolve!: (value: T) => void;
+    let reject!: (reason?: unknown) => void;
+    const promise = new Promise<T>((res, rej) => {
+      resolve = res;
+      reject = rej;
+    });
+    return { promise, resolve, reject };
+  }
+  const state = {
+    /** Deferreds for renders the mock has STARTED (the WASM call is in flight). */
+    pending: [] as Deferred<string>[],
+    active: 0,
+    maxActive: 0,
+  };
+  return { state, mkDeferred };
+});
+
+import { _setWasmModuleForTest, renderToHtml } from './wasmRenderer';
+
+/** Let the queued microtasks + a macrotask fire so the serial queue advances. */
+const flush = () => new Promise<void>((r) => setTimeout(r, 0));
+
+beforeAll(() => {
+  // Install the stub directly rather than mocking the
+  // 'wasm-quarto-hub-client' import: with the stub in place, initWasm()
+  // early-returns and the dynamic import (whose glue file only exists
+  // after a WASM build, absent in the CI TS-suites job) never fires.
+  _setWasmModuleForTest({
+    render_page_in_project_with_attribution: vi.fn(() => {
+      state.active += 1;
+      state.maxActive = Math.max(state.maxActive, state.active);
+      const d = mkDeferred<string>();
+      state.pending.push(d);
+      return d.promise.finally(() => {
+        state.active -= 1;
+      });
+    }),
+    // The CSS-version read after every successful render: a miss keeps
+    // the 'default' version suffix.
+    vfs_read_file: () => JSON.stringify({ success: false }),
+  });
+});
+
+beforeEach(() => {
+  state.pending.length = 0;
+  state.active = 0;
+  state.maxActive = 0;
+});
+
+describe('render serial queue (enqueueRender) characterization', () => {
+  it('runs renders one at a time, starting them in call order as slots free up', async () => {
+    const p1 = renderToHtml({ documentPath: 'a.qmd' });
+    const p2 = renderToHtml({ documentPath: 'a.qmd' });
+    const p3 = renderToHtml({ documentPath: 'a.qmd' });
+    await flush();
+
+    // Only the first render has started; the other two are queued.
+    expect(state.pending.length).toBe(1);
+
+    state.pending[0].resolve(JSON.stringify({ success: true, html: '<p>1</p>' }));
+    await flush();
+    expect(state.pending.length).toBe(2);
+
+    state.pending[1].resolve(JSON.stringify({ success: true, html: '<p>2</p>' }));
+    await flush();
+    expect(state.pending.length).toBe(3);
+
+    state.pending[2].resolve(JSON.stringify({ success: true, html: '<p>3</p>' }));
+    const results = await Promise.all([p1, p2, p3]);
+
+    // Never concurrent...
+    expect(state.maxActive).toBe(1);
+    // ...and each caller got its own render's output, in order.
+    expect(results.map((r) => r.html)).toEqual([
+      '<p>1</p><!-- css-version: default -->',
+      '<p>2</p><!-- css-version: default -->',
+      '<p>3</p><!-- css-version: default -->',
+    ]);
+  });
+
+  it('executes every enqueued render — no coalescing of queued work (finding 2 pin)', async () => {
+    // A burst of N content updates with a render already in flight: today
+    // every queued render still runs to completion. Phase 1 of the plan
+    // changes this to latest-wins; this test is the baseline it moves.
+    const promises = Array.from({ length: 5 }, () => renderToHtml({ documentPath: 'a.qmd' }));
+    await flush();
+
+    // Drain the queue one slot at a time until all five have run.
+    for (let started = 0; started < 5; started++) {
+      await flush();
+      expect(state.pending.length).toBe(started + 1);
+      state.pending[started].resolve(JSON.stringify({ success: true, html: `<p>${started}</p>` }));
+    }
+    await Promise.all(promises);
+    expect(state.pending.length).toBe(5);
+  });
+
+  it('a rejected render reports failure to its own caller and does not jam the queue', async () => {
+    const p1 = renderToHtml({ documentPath: 'a.qmd' });
+    const p2 = renderToHtml({ documentPath: 'a.qmd' });
+    await flush();
+
+    state.pending[0].reject(new Error('boom'));
+    // The failing caller sees a success:false result (renderToHtmlInner's
+    // catch), not a rejection.
+    await expect(p1).resolves.toMatchObject({ success: false, error: 'boom' });
+
+    // The queue is unjammed: the next render starts and can succeed.
+    await flush();
+    expect(state.pending.length).toBe(2);
+    state.pending[1].resolve(JSON.stringify({ success: true, html: '<p>after</p>' }));
+    await expect(p2).resolves.toMatchObject({ success: true });
+  });
+});
diff --git a/ts-packages/preview-runtime/src/wasmRenderer.ts b/ts-packages/preview-runtime/src/wasmRenderer.ts
index 1a74dfd373..38572fa6aa 100644
--- a/ts-packages/preview-runtime/src/wasmRenderer.ts
+++ b/ts-packages/preview-runtime/src/wasmRenderer.ts
@@ -26,6 +26,7 @@ import type {
 import { discoverUserGrammars } from './userGrammar/Discovery';
 import { UserGrammarCache } from './userGrammar/Cache';
 import { loadUserGrammar } from './userGrammar/Highlight';
+import { perfEnabled, perfNow, perfSegment, perfMark, perfExtra } from './perfCounters';
 
 // Response types from WASM module
 interface VfsResponse {
@@ -311,6 +312,18 @@ export function isWasmReady(): boolean {
   return wasmModule !== null;
 }
 
+/**
+ * Test-only: install a WASM module stub directly, bypassing
+ * `initWasm()`'s dynamic `import('wasm-quarto-hub-client')`. Unit tests
+ * use this so they never resolve the wasm-pack glue, which does not
+ * exist in CI jobs that skip the WASM build (bd-k8ptcrd9; the
+ * vi.mock approach resolves the aliased real file and fails there).
+ * Pass `null` to reset.
+ */
+export function _setWasmModuleForTest(module: unknown): void {
+  wasmModule = module as WasmModuleExtended | null;
+}
+
 /**
  * Get the WASM module, throwing if not initialized
  */
@@ -321,6 +334,30 @@ function getWasm() {
   return wasmModule;
 }
 
+/**
+ * Parse a WASM JSON-string response envelope, recording parse time and
+ * payload size under the `wasm-envelope.parse` gauge (finding 4 of the
+ * 2026-10-24 main-thread render-path plan: every WASM call pays a
+ * JSON-envelope tax). Sizes are `String.length` (UTF-16 code units) —
+ * exact for ASCII payloads, proportionate otherwise.
+ *
+ * The disabled path is an early return with the exact pre-instrumentation
+ * shape — one flag read, no try/finally — so the gauges are free when
+ * off (see the plan's disabled-overhead verification).
+ */
+function parseEnvelope<T>(json: string): T {
+  if (!perfEnabled()) {
+    return JSON.parse(json) as T;
+  }
+  const t0 = performance.now();
+  try {
+    return JSON.parse(json) as T;
+  } finally {
+    perfSegment('wasm-envelope.parse', t0);
+    perfExtra('wasm-envelope.parse', 'bytes', json.length);
+  }
+}
+
 // ============================================================================
 // VFS Operations
 // ============================================================================
@@ -330,7 +367,19 @@ function getWasm() {
  */
 export function vfsAddFile(path: string, content: string): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_add_file(path, content));
+  // Full document text crosses the JS→WASM boundary here on every
+  // keystroke (the render-path plan's "amplifier"); count, time and size
+  // it. `bytes` is String.length (UTF-16 code units).
+  if (!perfEnabled()) {
+    return JSON.parse(wasm.vfs_add_file(path, content));
+  }
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(wasm.vfs_add_file(path, content));
+  } finally {
+    perfSegment('vfs.write', t0);
+    perfExtra('vfs.write', 'bytes', content.length);
+  }
 }
 
 /**
@@ -338,7 +387,16 @@ export function vfsAddFile(path: string, content: string): VfsResponse {
  */
 export function vfsAddBinaryFile(path: string, content: Uint8Array): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_add_binary_file(path, content));
+  if (!perfEnabled()) {
+    return JSON.parse(wasm.vfs_add_binary_file(path, content));
+  }
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(wasm.vfs_add_binary_file(path, content));
+  } finally {
+    perfSegment('vfs.write', t0);
+    perfExtra('vfs.write', 'bytes', content.byteLength);
+  }
 }
 
 /**
@@ -346,7 +404,7 @@ export function vfsAddBinaryFile(path: string, content: Uint8Array): VfsResponse
  */
 export function vfsRemoveFile(path: string): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_remove_file(path));
+  return parseEnvelope(wasm.vfs_remove_file(path));
 }
 
 /**
@@ -354,7 +412,7 @@ export function vfsRemoveFile(path: string): VfsResponse {
  */
 export function vfsListFiles(): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_list_files());
+  return parseEnvelope(wasm.vfs_list_files());
 }
 
 /**
@@ -362,7 +420,7 @@ export function vfsListFiles(): VfsResponse {
  */
 export function vfsClear(): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_clear());
+  return parseEnvelope(wasm.vfs_clear());
 }
 
 /**
@@ -370,7 +428,7 @@ export function vfsClear(): VfsResponse {
  */
 export function vfsReadFile(path: string): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_read_file(path));
+  return parseEnvelope(wasm.vfs_read_file(path));
 }
 
 /**
@@ -379,7 +437,7 @@ export function vfsReadFile(path: string): VfsResponse {
  */
 export function vfsReadBinaryFile(path: string): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_read_binary_file(path));
+  return parseEnvelope(wasm.vfs_read_binary_file(path));
 }
 
 // ============================================================================
@@ -396,7 +454,7 @@ export function vfsReadBinaryFile(path: string): VfsResponse {
  */
 export function setRuntimeMetadata(yaml: string): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_set_runtime_metadata(yaml));
+  return parseEnvelope(wasm.vfs_set_runtime_metadata(yaml));
 }
 
 /**
@@ -404,7 +462,7 @@ export function setRuntimeMetadata(yaml: string): VfsResponse {
  */
 export function getRuntimeMetadata(): VfsResponse {
   const wasm = getWasm();
-  return JSON.parse(wasm.vfs_get_runtime_metadata());
+  return parseEnvelope(wasm.vfs_get_runtime_metadata());
 }
 
 /**
@@ -512,7 +570,17 @@ export async function renderQmd(
   userGrammars?: unknown,
 ): Promise<RenderResponse> {
   const wasm = getWasm();
-  return JSON.parse(await wasm.render_qmd(path, userGrammars));
+  if (!perfEnabled()) {
+    return JSON.parse(await wasm.render_qmd(path, userGrammars));
+  }
+  perfMark('render.started');
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(await wasm.render_qmd(path, userGrammars));
+  } finally {
+    perfSegment('render.wasm-call', t0);
+    perfMark('render.completed');
+  }
 }
 
 /**
@@ -531,7 +599,17 @@ export async function renderPrintable(
   userGrammars?: unknown,
 ): Promise<RenderResponse> {
   const wasm = getWasm();
-  return JSON.parse(await wasm.render_printable(path, userGrammars));
+  if (!perfEnabled()) {
+    return JSON.parse(await wasm.render_printable(path, userGrammars));
+  }
+  perfMark('render.started');
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(await wasm.render_printable(path, userGrammars));
+  } finally {
+    perfSegment('render.wasm-call', t0);
+    perfMark('render.completed');
+  }
 }
 
 /**
@@ -607,15 +685,33 @@ export async function renderPageInProjectWithAttribution(
   preferPreviewFormat?: boolean,
 ): Promise<RenderResponse> {
   const wasm = getWasm();
-  return JSON.parse(
-    await wasm.render_page_in_project_with_attribution(
-      path,
-      userGrammars,
-      attributionJson ?? undefined,
-      captureGzJson,
-      preferPreviewFormat,
-    ),
-  );
+  if (!perfEnabled()) {
+    return JSON.parse(
+      await wasm.render_page_in_project_with_attribution(
+        path,
+        userGrammars,
+        attributionJson ?? undefined,
+        captureGzJson,
+        preferPreviewFormat,
+      ),
+    );
+  }
+  perfMark('render.started');
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(
+      await wasm.render_page_in_project_with_attribution(
+        path,
+        userGrammars,
+        attributionJson ?? undefined,
+        captureGzJson,
+        preferPreviewFormat,
+      ),
+    );
+  } finally {
+    perfSegment('render.wasm-call', t0);
+    perfMark('render.completed');
+  }
 }
 
 /**
@@ -647,9 +743,21 @@ export async function renderPageForPreview(
   options?: RenderPageForPreviewOptions,
 ): Promise<RenderResponse> {
   const wasm = getWasm();
-  return JSON.parse(
-    await wasm.render_page_for_preview(path, userGrammars, captureGzJson, options),
-  );
+  if (!perfEnabled()) {
+    return JSON.parse(
+      await wasm.render_page_for_preview(path, userGrammars, captureGzJson, options),
+    );
+  }
+  perfMark('render.started');
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(
+      await wasm.render_page_for_preview(path, userGrammars, captureGzJson, options),
+    );
+  } finally {
+    perfSegment('render.wasm-call', t0);
+    perfMark('render.completed');
+  }
 }
 
 // ---- pandoc-wasm epic (R2): the request exports --------------------------
@@ -761,14 +869,14 @@ export function classifyPandocCompletion(
   stderr: string,
   jsonPath: string,
 ): { success: boolean; diagnostics: AstDiagnostic[] } {
-  return JSON.parse(
+  return parseEnvelope(
     getWasm().classify_pandoc_completion(stageName, success, status, stderr, jsonPath),
   );
 }
 
 /** The formats pandoc.wasm can produce, in menu order. */
 export function getPandocFormats(): PandocFormatInfo[] {
-  return JSON.parse(getWasm().get_pandoc_formats()).formats;
+  return parseEnvelope<{ formats: PandocFormatInfo[] }>(getWasm().get_pandoc_formats()).formats;
 }
 
 // ============================================================================
@@ -781,7 +889,7 @@ export function getPandocFormats(): PandocFormatInfo[] {
 
 /** The importable formats (the picker's `accept` list, drop interception) and the source size cap. */
 export function getImportFormatTable(): ImportFormatTable {
-  return JSON.parse(getWasm().get_import_formats());
+  return parseEnvelope(getWasm().get_import_formats());
 }
 
 /**
@@ -790,7 +898,7 @@ export function getImportFormatTable(): ImportFormatTable {
  * only: no `request`. `size` is the length of the bytes actually read.
  */
 export function prepareImport(fileName: string, size: number, sha256Hex: string): PrepareImportResponse {
-  return JSON.parse(getWasm().prepare_import(fileName, size, sha256Hex));
+  return parseEnvelope(getWasm().prepare_import(fileName, size, sha256Hex));
 }
 
 /**
@@ -805,7 +913,7 @@ export function finishImport(
   mediaManifestJson: string,
   format?: string,
 ): FinishImportResponse {
-  return JSON.parse(getWasm().finish_import(jsonText, stderr, targetQmdPath, mediaManifestJson, format));
+  return parseEnvelope(getWasm().finish_import(jsonText, stderr, targetQmdPath, mediaManifestJson, format));
 }
 
 /** Turn a failed import run (`kind`: the host's failure kind) into Q-24 diagnostics. */
@@ -814,7 +922,7 @@ export function classifyImportFailure(
   status: number | null,
   stderr: string,
 ): ClassifyImportFailureResponse {
-  return JSON.parse(getWasm().classify_import_failure(kind, status, stderr));
+  return parseEnvelope(getWasm().classify_import_failure(kind, status, stderr));
 }
 
 /** Options of {@link resolvePandocFormats}. */
@@ -836,7 +944,7 @@ export interface ResolvePandocFormatsOptions {
  * document's own format; `formats` lists every declared key.
  */
 export function resolvePandocFormats(path: string, options?: ResolvePandocFormatsOptions): ResolvePandocFormatsResponse {
-  return JSON.parse(getWasm().resolve_pandoc_formats(path, options));
+  return parseEnvelope(getWasm().resolve_pandoc_formats(path, options));
 }
 
 /**
@@ -849,9 +957,21 @@ export async function renderQmdContent(
   userGrammars?: unknown,
 ): Promise<RenderResponse> {
   const wasm = getWasm();
-  return JSON.parse(
-    await wasm.render_qmd_content(content, templateBundle, userGrammars),
-  );
+  if (!perfEnabled()) {
+    return JSON.parse(
+      await wasm.render_qmd_content(content, templateBundle, userGrammars),
+    );
+  }
+  perfMark('render.started');
+  const t0 = performance.now();
+  try {
+    return parseEnvelope(
+      await wasm.render_qmd_content(content, templateBundle, userGrammars),
+    );
+  } finally {
+    perfSegment('render.wasm-call', t0);
+    perfMark('render.completed');
+  }
 }
 
 
@@ -955,12 +1075,16 @@ export async function parseQmdToAstWithAttribution(
   try {
     await initWasm();
     const wasm = getWasm();
+    // Finding 1: PreviewRouter calls this per keystroke only to read
+    // `success`; the AST is serialized, shipped, parsed and discarded.
+    const pT0 = perfNow();
     const responseJson = await wasm.parse_qmd_to_ast_with_attribution(
       qmdContent,
       attributionJson ?? undefined,
     );
+    if (pT0) perfSegment('parse.qmd-to-ast', pT0);
 
-    const response: ParseResult = JSON.parse(responseJson);
+    const response: ParseResult = parseEnvelope(responseJson);
 
     if (response.success) {
       return {
@@ -1014,7 +1138,7 @@ export async function writeQmd(astJson: string): Promise<WriteQmdResult> {
     const responseJson = await wasm.write_qmd(astJson);
 
     // The response reuses AstResponse structure, but with "qmd" in the "ast" field
-    const response: { success: boolean; ast?: string; error?: string } = JSON.parse(responseJson);
+    const response: { success: boolean; ast?: string; error?: string } = parseEnvelope(responseJson);
 
     if (response.success) {
       return {
@@ -1045,7 +1169,7 @@ export function incrementalWriteQmd(originalQmd: string, newAst: RustQmdJson): s
 
   const newAstJson = JSON.stringify(newAst)
   const responseJson = wasmModule.incremental_write_qmd(originalQmd, newAstJson)
-  const response: AstResponse = JSON.parse(responseJson)
+  const response: AstResponse = parseEnvelope(responseJson)
 
   if (!response.success || !response.qmd) {
     throw new Error(`Incremental write failed: ${response.error}`)
@@ -1077,7 +1201,7 @@ export function parseQmdContentSync(qmdText: string): AstResponse {
   if (!wasmModule) {
     throw new Error('WASM not initialized. Call initWasm() first.')
   }
-  return JSON.parse(wasmModule.parse_qmd_content(qmdText)) as AstResponse
+  return parseEnvelope(wasmModule.parse_qmd_content(qmdText)) as AstResponse
 }
 
 export function applyNodeEdit(
@@ -1096,7 +1220,7 @@ export function applyNodeEdit(
     destinationSourceInfoJson,
     modifiedSubtreeJson,
   )
-  const response: AstResponse = JSON.parse(responseJson)
+  const response: AstResponse = parseEnvelope(responseJson)
 
   if (!response.success || !response.qmd) {
     throw new Error(`apply_node_edit failed: ${response.error}`)
@@ -1123,7 +1247,7 @@ export function regenerateNestedBuffers(
     throw new Error('WASM not initialized. Call initWasm() first.')
   }
   try {
-    return JSON.parse(
+    return parseEnvelope(
       wasmModule.regenerate_nested_buffers(content, untransformedAstJson),
     ) as Record<string, string>
   } catch {
@@ -1168,7 +1292,7 @@ export async function convert(
     const responseJson = await wasm.convert(document, inputFormat, outputFormat);
 
     // The response reuses AstResponse structure, but with output in the "ast" field
-    const response: { success: boolean; ast?: string; error?: string } = JSON.parse(responseJson);
+    const response: { success: boolean; ast?: string; error?: string } = parseEnvelope(responseJson);
 
     if (response.success) {
       return {
@@ -1375,7 +1499,7 @@ export interface CreateProjectResponse {
 export async function getProjectChoices(): Promise<ProjectChoice[]> {
   await initWasm();
   const wasm = getWasm();
-  const response: ProjectChoicesResponse = JSON.parse(wasm.get_project_choices());
+  const response: ProjectChoicesResponse = parseEnvelope(wasm.get_project_choices());
   return response.choices;
 }
 
@@ -1386,7 +1510,7 @@ export async function getProjectChoices(): Promise<ProjectChoice[]> {
 export async function getProjectChoiceGroups(): Promise<ProjectChoiceGroup[]> {
   await initWasm();
   const wasm = getWasm();
-  const response: ProjectChoicesResponse = JSON.parse(wasm.get_project_choices());
+  const response: ProjectChoicesResponse = parseEnvelope(wasm.get_project_choices());
   return response.groups ?? [];
 }
 
@@ -1400,7 +1524,7 @@ export async function getProjectChoiceGroups(): Promise<ProjectChoiceGroup[]> {
 export async function createProject(choiceId: string, title: string): Promise<CreateProjectResponse> {
   await initWasm();
   const wasm = getWasm();
-  return JSON.parse(await wasm.create_project(choiceId, title));
+  return parseEnvelope(await wasm.create_project(choiceId, title));
 }
 
 /**
diff --git a/ts-packages/preview-runtime/vitest.config.ts b/ts-packages/preview-runtime/vitest.config.ts
index d8d9d4d8fa..a3a05250c5 100644
--- a/ts-packages/preview-runtime/vitest.config.ts
+++ b/ts-packages/preview-runtime/vitest.config.ts
@@ -12,6 +12,11 @@ export default defineConfig({
         __dirname,
         '../../hub-client/wasm-quarto-hub-client/wasm_quarto_hub_client.js',
       ),
+      // The WASM glue and wasmRenderer.ts import the SASS bridge by its
+      // Vite-root absolute path (the same one the Rust `raw_module`
+      // annotation uses). Map it to the workspace copy so `initWasm()`
+      // works outside the dev server (hub-client's configs do the same).
+      '/src/wasm-js-bridge': path.resolve(__dirname, '../wasm-js-bridge/src'),
       // Workspace-package aliases to source — mirrors the pattern in
       // hub-client/vitest.config.ts. Vitest doesn't honor the `source`
       // export condition on fresh clones, so workspace deps need