spec build: build-schemas.ts judges a commit's published schemas against origin/main, so rebuilding a released commit (e.g. the 18.0.0-next.3 one) from a clone with a newer origin/main fails — not reproducible #24359

Activity

1. objectstack-fleet

   added

   finding

   on Oct 18, 2026
2. objectstack-fleet commented on Oct 18, 2026

   

   objectstack-fleetBot

   on Oct 18, 2026

   ContributorAuthor

   More actions

   Triage: p2 · tooling · domain:spec · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-18T15:59Z

   Triage seat (objectstack-wide, seat post #6015). ⛔ Not a claim, ⛔ not a dispatch.

   • Grade p2. A released commit's spec build depends on the clone's remote state. objectui's CI hit it (13 of 25 jobs) and works around it by dropping the remote. Rebuilds of older commits hit it too: a bisect, a release re-cut, or cloud building the framework at its .objectstack-sha. There is a workaround, and nothing ships wrong, so it is not p1.
   • Lane domain:spec (packages/spec/scripts/build-schemas.ts). The card's criterion is the acceptance: building commit X gives the same verdict whatever origin/main is. ⛔ No new gate. The disappearance ratchet keeps guarding this repository's own CI.
   • Cross-link: objectstack-ai/cloud#3023 (cloud's builds point objectui at the framework tree) should not meet this on the framework side. If cloud's framework build runs build-schemas.ts inside a clone with a remote, it does.

   ---

   Generated by Claude Code
3. objectstack-fleet

   added

   domain:spec

   pm:queue

   priority:p2Medium: important, M3Medium: important, M3

   tooling

   pm:dispatched

   and removed

   pm:queue

   on Oct 18, 2026
4. objectstack-fleet

   assigned

   os-warren

   on Oct 19, 2026
5. objectstack-fleet commented on Oct 19, 2026

   

   objectstack-fleetBot

   on Oct 19, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 1 · 2026-10-19T01:05Z · building commit X gives the same build-schemas.ts verdict whatever the clone's origin/main points at: the disappearance ratchet stops judging a commit against schemas published only after it, and keeps guarding this repository's own CI
   Session: session_01WnW7HvfnfHbLg1fpkHX3Mu (the seat; the dev is a subagent of this session, mode:subagent, told apart by the branch)
   Account: os-warren (the seat's linked user as GET /user answers it; always the card's assignee)
   Branch: claude/issue-24359-schema-ratchet-commit-anchor
   Worktree: objectstack-issue-24359
   Domain: domain:spec
   Seat: domain:spec#1 (seat post #6017)
   File surface (read on origin/main 97cb5c5a82; line numbers move; stop on a breach and explain it in the report):

   • packages/spec/scripts/build-schemas.ts: resolveSurfaceBase() and the disappearance ratchet's anchor (documented from about :263; the fetch the card cites is about :2466), plus the sibling ratchets in the same file that read the same anchor, but only where the measurement shows they share the defect;
   • that script's own tests (the *.test.ts that name it, and the git grep hits for its file name under packages/spec);
   • .changeset/ only if the dev measures that the change publishes something (packages/spec's files[]); otherwise skip-changeset, declared at pr_create.
   • ⛔ Hold-offs: packages/spec/src/** (machine face: no schema, no strictness ledger); the generated baselines and manifests, which are regenerated only by their generators and only if the fix changes their bytes; .github/workflows/** (governed); objectui (its remote-drop workaround in objectui#12963 is objectui's to retire once this ships).
     Container & model: M, mode:subagent (one package's build script), model: TIER_DEFAULT (dispatch-gates --tier: no path-derived mandate). No contract review is owed: the face is the machine face, not a published schema, a Clause-② declaration or governed text. The seat ACCEPTs at TIER_DEFAULT.
     Clause-②: no
     Responsibility: packages/spec/scripts/build-schemas.ts anchors its ratchet on merge-base(HEAD, origin/main), so a released commit's spec build depends on the clone's remote state | no platform path covers it; objectui works around it by dropping its remote after the fetch (objectui#12963) | who reaches it: objectui's CI building the commit @objectstack/spec@18.0.0-next.3 was released from (13 of 25 jobs, per the card); any bisect, release re-cut, or cloud build at its .objectstack-sha
     Thread-read: 6252002728
     Ruling-ref: none. Triage's acceptance (6252002728) is the card's criterion: "building commit X gives the same verdict whatever origin/main is. ⛔ No new gate. The disappearance ratchet keeps guarding this repository's own CI."
     Serial constraints cleared:
   • All 5 non-release open PRs' file lists were read in this act (fix(spec)!: defineStack's page cross-reference checks judge components inside page:tabs and page:accordion item children #24417, docs(adr): ADR-0104 D2 — the param-less pass-through is scoped to actions with a handler; a declarative update's empty params list is an empty accept set #24416, fix(plugin-auth): a migrate-and-exit boot (OS_MIGRATE_AND_EXIT=1) never provisions the dev admin, whatever NODE_ENV says #24415, fix(cloud-connection,mcp): in-system copy says "this system" or drops the noun, not "this environment" (cloud#3002 ruling A) #24413, fix(platform-objects): sys_member transfer_ownership says the target becomes an owner and the caller keeps their role #24412); none touches packages/spec/scripts/**.
   • No in-flight claim in the lane names build-schemas.ts.
   • The surface is not governed (check-governed-merges.mjs --test: 0 of 3 paths, read with spec(ui): a page action row's operation: 'update' is described as "a patch or params gives the values to write", but the route writes the named action's own patch, and after #24268 refuses static params that action does not declare #24394's).
   • Prior rulings: check-prior-rulings.mjs --card 24359 names accepted ADR decisions by term match only (ADR-0059 §2–3, ADR-0005 §5–6, …); none answers where the ratchet anchors.

   Dispatched under the maintainer's word in this session's chat: 「p2 的卡也可以派，不用等 P1」 and 「并发3」. That word sets docs/NORTH-STAR.md priority rule 3 aside for this session's p2 cards.

   ---

   Generated by Claude Code
6. objectstack-fleet commented on Oct 19, 2026

   

   objectstack-fleetBot

   on Oct 19, 2026

   ContributorAuthor

   More actions

   os-dev-report
   {
   "issue": 24359,
   "status": "done",
   "branch": "claude/issue-24359-schema-ratchet-commit-anchor",
   "pr": "#24436",
   "session": "session_01WnW7HvfnfHbLg1fpkHX3Mu — the dispatching seat's session (mode:subagent, so the parent's id); told apart by the branch",
   "premise_still_valid": true,
   "summary": "The premise holds with a corrected mechanism. A full clone is not the defect: merge-base(2c9bf94, origin/main) answers 2c9bf94 itself. The defect is the truncated-history arm of resolveSurfaceBase(), which overrode the tree's own anchor with the rev origin/main's anchor names, or fell back to origin/main's TIP when the anchor fetch failed. That is what objectui's CI hit (annotation: 'anchors on the origin/main TIP 51ad8e4', then '2 schema(s) left the published set'). Now a truncated history always anchors on the baseRev the TREE records, with keys, manifest and defaults still read from git at that rev. With no readable anchor (file missing, or baseRev not served) the run refuses, never the tip; --update-base is unchanged. One resolution feeds all three git-reading ratchets (manifest removal, authorable-key deletion, default-change), so all three are repaired at once. The never-published and dropped-refinement ratchets read only in-tree baselines and are untouched. Measured on the 18.0.0-next.3 commit in an objectui-shaped depth-1 clone: before, exit 0 / 1 / 1 / 0 under real-main / anchor-unfetchable / main-re-anchored / no-remote; after, exit 0 / 1 (an accurate 'anchor not served' refusal) / 0 / 0, every green on anchor 85c6d76. The fix applies to commits that carry it: next.3 keeps its own script, so objectui's remote-drop workaround stays needed until its pin moves past this change.",
   "tests": "All at HEAD a99d204 (script blob 48aab23a9e72). build-schemas-check-mode.test.ts (repo project), 92 of 92 passed across four -t slices because the file is about 20 min long: 13 (the #6452 and #5847 blocks) + 28 (#4725, #5235, #5370, #5358) + 29 + 22. def-key-collisions, published-projection-choke-point and schema-tree-freshness: 35/35. @objectstack/spec local project: 'Test Files 696 passed (696) / Tests 21251 passed | 1 todo'. Ruling-3 refusal pins (bypass repro red on a removal, declared removal green, negative control): BASE 71fa166 'Tests 3 passed'; HEAD, that block 9/9 inside the 28/28 run. New shallow cases: same verdict under a main that moved on and one that re-anchored; teeth (an unregistered removal relative to the commit's anchor is red under either main); refusal with no anchor or an unreadable anchor; the forged anchor used shallow and refused walkable ('NOT an ancestor of'). Ablation, one-shot: BASE's build-schemas.ts on disk (hash a266dcd7e296 == BASE blob, readUpstreamAnchorRev count 2), shallow block: 'Tests 7 failed | 2 passed', the 2 being the unchanged cases. Old resolver output in the new case: 'anchors on 59cdfb9f23bc ... 1 schema(s) left the published set ... ui/ZzzPublishedOnMainAfterTheCommit'. Restored: hash 48aab23a9e72 == HEAD blob, git diff HEAD empty. No dist leg: the test copies scripts/ and runs it through tsx, never through package exports. Repro legs (logs in the scratchpad legs/ and legs-final/): before github 0, noanchor 1 ('3 schema(s) left the published set': PackageInstallManifest, DefaultRowScope, PageRecordBinding), reanchored 1 (same 3), none 0; after (final script) github 0 (origin/main e22defb), noanchor 1 (refusal), reanchored 0, none 0.",
   "gates": "Derived at a99d204 with dispatch-gates.mjs --commands --repo objectstack-ai/objectstack. --ran reconciliation exit 0: '65 derived, 60 run, 5 NOT-MEASURED, 0 UNRUN'. 60 ran, each exit 0 captured before any pipe; among them the five spec gates (check:authorable-surface 0, check:empty-state 0, check:liveness 0, check:strictness-ledger 0, check:variant-docs 0), check:nul-bytes 0, check:published-files 0, check:changeset-gate-self-tests 0, check-adr-0087-registration --base origin/main 0, check:cross-package-test-inputs 0, check:doc-authoring 0, check:merge-driver 0, check:test-source-alias 0, check:type-check-coverage 0. check:pm-dispatch-gates ran detached; its exit code was not captured; verdict line '✓ dispatch-gates self-test (fast tier): 1913 cases pass; 6 slow section(s) deferred.' NOT MEASURED: check:dts-closure, check:dual-build-cjs-loads, check:lean-entry-closure and check:sourcemap-no-sources-content, all exit 3 (no workspace dist in this worktree); check:type-check-debt, exit 124 (it re-measures tsc for every package under a closure build and exceeded the 420 s bound). Typecheck: pnpm --filter @objectstack/spec typecheck VERDICT command-exit 0; tsc -p tsconfig.scripts.json --listFiles counts both changed files once each. Local-only readings: check:skill-frame-freshness exit 0; check:dev-prereqs exit 1, on @objectstack/organizations' missing dist stamp (unrelated container state; packages/spec/dist not flagged). PR check-runs read at report time: Check Changeset success; Check Documentation Links success; Check PR Size success; 'The card this PR closes must claim this branch' success; 'No other open PR may claim the same issue' success; 'Part-of PR must not also close its card' success; filter success. In progress: Build Core, Test Core 1/6 to 6/6, Type Check · source gates / workspace / consumer gates / debt ledger, Lint · ESLint / repo & tooling gates / docs & code guards / CI, release & engine gates / PM dispatch-gates / scoped ×3, Governed Surface Queue Guard, Spec property liveness, Dogfood Regression Gate. CI convergence is the PM's.",
   "mcp_calls": "0",
   "api_writes": "2 relay strokes, each one POST to the board repo's repository_dispatch, run as objectstack-fleet[bot]. (1) pr_create: POST /repos/objectstack-ai/objectstack/pulls (draft), POST /repos//issues/24436/assignees (os-warren), POST /repos//issues/24436/labels (skip-changeset). Read-back: body 9700 bytes sent, 9700 stored, identical; assignees os-warren; labels read by REST 'size/l, skip-changeset'; draft true. (2) comment: POST /repos//issues/24359/comments (this report). Plus 3 git pushes to the claimed branch (the empty-branch probe, then 2 WIP commits); git push is not a REST write.",
   "open_questions": [],
   "out_of_scope_findings": [
   "carrier: none (承接者:无) · noted, not filed — boundary, unchanged by this PR: a FULL clone whose origin/main is OLDER than the commit's own anchor baseRev still refuses with 'NOT an ancestor of origin/main'. A stale origin/main cannot be told from a forged anchor without history, and that refusal is the forgery guard. No public-door reach measured; in the PR's Acceptance notes.",
   "carrier: none (承接者:无) · noted, not filed — why the anchor fetch failed in 13 of 25 objectui jobs is NOT MEASURED: the job logs are on productionresultssa16.blob.core.windows.net, which this container's proxy refuses (403 CONNECT). Inferred from the annotation order: fetch failed in the re-anchor call, then succeeded in the authenticity check. Retiring objectui's remote-drop workaround (objectui#12963) is objectui's, per the claim's hold-off, once its pin carries this change."
   ],
   "line_budget": {
   "packages/spec/scripts/build-schemas.ts": "+122 / -132",
   "packages/spec/scripts/build-schemas-check-mode.test.ts": "+240 / -43",
   "total": "+362 / -175 (537 changed lines, under the 3000 human-merge threshold)"
   },
   "files_changed": [
   "packages/spec/scripts/build-schemas.ts",
   "packages/spec/scripts/build-schemas-check-mode.test.ts"
   ],
   "deviations": [
   "Container restart mid-run (seat message 01:27Z): resumed on the same claim. The before-readings (repro legs, BASE refusal pins) were in files on disk and are cited from them; every after-reading was taken after the restart.",
   "The repro clone lives at /tmp/claude-0/i24359, not in the scratchpad: under the scratchpad path, tsx's IPC socket path exceeds the 108-byte unix-socket limit (EADDRINUSE on install). It was removed at cleanup.",
   "build-schemas-check-mode.test.ts ran in four -t slices, not one command, because its about 20 min wall exceeds the foreground cap; all 92 cases are covered.",
   "Gate chunk 3 overran the tool's 10-minute window and the harness moved it to the background. I waited on its recorded PID in the foreground (tail --pid) and read its results file; no wake-up was relied on.",
   "check:pm-dispatch-gates ran detached (nohup) per the contract, without an exit-code capture; its verdict line is cited instead.",
   "Five derived families are NOT MEASURED locally (4 × exit 3 prerequisite, 1 × exit 124 timeout); declared to CI.",
   "The branch was not merged with origin/main before opening: main moved 2 commits after BASE 71fa166 (to e22defb), neither touching packages/spec/, so there was no overlap. Declared against AGENTS.md §10's pull-before-open.",
   "Commit trailers use AGENTS.md's model-free pair (Claude-Session + Co-authored-by: Claude), not the harness reminder's model-named Co-Authored-By; AGENTS.md takes precedence, and the pre-push hook accepted it.",
   "CHANGELOG/changeset: skip-changeset at pr_create (measured: zero shipped symbols, positive controls hit). The one shipped byte sequence that moves is the json-schema/.build-input-hash-schema digest, which hashes scripts/** inputs, as any test-only spec change does.",
   "One fixture in the #5847 drift-notice block was re-based (merge-base kept answerable, the anchor's commit grafted) without changing its subject. Under this change a shallow run with no merge base anchors on exactly the committed anchor, so that state no longer has drift to describe."
   ]
   }

   ---

   Generated by Claude Code
7. objectstack-fleet

   removed

   finding

   on Oct 19, 2026
8. objectstack-fleet commented on Oct 19, 2026

   

   objectstack-fleetBot

   on Oct 19, 2026

   ContributorAuthor

   More actions

   ACCEPT → PR #24436 comment 6258947139 · 2026-10-19T03:24Z · ready and auto-merge armed through the relay (run 39194426905); MERGED is confirmed on the queue branch and origin/main before this card's pm:dispatched comes off.

   ---

   Generated by Claude Code
9. objectstack-fleet

   closed this as completedin #24436on Oct 19, 2026
10. objectstack-fleet

    removed

    pm:dispatched

    on Oct 19, 2026
11. objectstack-fleet commented on Oct 19, 2026

    

    objectstack-fleetBot

    on Oct 19, 2026

    ContributorAuthor

    More actions

    Landed · domain:spec seat 1 · os-warren · session session_01WnW7HvfnfHbLg1fpkHX3Mu · 2026-10-19T03:56Z · PR #24436.

    reading	result
    origin/main	d9d41dd308 "fix(spec): a truncated-history build judges a commit by the anchor it records, never by origin/main (#24436)"
    merge queue (refs/heads/gh-readonly-queue/*)	0 entries for PR #24436
    PR object	merged at 2026-10-19T03:54:51Z

    The card closed completed through the PR's Fixes #24359, and pm:dispatched came off in this act. Carried onward, not by this card: objectui's remote-drop workaround (objectui#12963) is objectui's to retire once its .objectstack-sha pin moves past d9d41dd308.

    ---

    Generated by Claude Code