Let the README say the paid half happened, and close the build plan's rows

The README still said the real-model run had deliberately not been spent and
that the fixture planted eight defects; both stopped being true today. Status
now carries both halves of the evidence with their numbers: the free fixture
run at 11/11, and the 2026-08-04 subscription runs, one refusal that was the
coverage law working, one product bug found and fixed the same day, then two
clean 8/8 runs at about a dollar and five minutes each. What it still does
not claim is the one thing that remains open on purpose: whether the findings
are worth a person's time is the maintainer's verdict, unmade.

The screenshot caption stops implying the smoke never happened and says
precisely what the images are: the browser suite's fake-engine review, with
the real-run evidence living uncommitted under review/. No real project has
been reviewed through the app yet, and the caption now says that is the next
gate rather than leaving it to be inferred.

BUILD-PLAN's WP-G through WP-L still carried the 2026-08-03 audit's PARTIAL
notes for work U6 to U13 has since closed; all six rows now read DONE with
dates and what closed them, and PROJECT-STATE's six-of-twelve line follows.
Construction is finished; judgment is what is left.

Author: Poolchaos

README.md

@@ -19,22 +19,29 @@ You can add a project, review one branch against another, watch the run,
 decide every finding, and export a report. A browser test walks exactly that
 path on every CI run, from an empty app to an exported file.
 
-What is not settled is whether the findings are worth a person's time. That
-needs a real model and real usage, and it has deliberately not been spent yet.
-What has been proven costs nothing: against a fixture with eight planted
-defects and two files that are deliberately correct, the pipeline found all
-eight, invented nothing in the clean files, and discarded no quotation as
-unverifiable. Run it yourself with `npm run demo:fixture -- --fake`, which
-spends nothing. That establishes the machinery carries a
-correct review intact, and no more. The judgment is the maintainer's, and it
-is described in [docs/plans/FG2-CHECKLIST.md](docs/plans/FG2-CHECKLIST.md).
-
-Eleven of twelve work packages are complete: the review engine, the data and
-git layers, every screen, the design pass, the accessibility checks, and the
+What is not settled is whether the findings are worth a person's time. Two
+kinds of evidence exist, and they answer different halves of that question.
+The free half: against a fixture with eleven planted defects (measured
+2026-08-04) and two files that are deliberately correct, the pipeline carries
+a correct review intact, finds all eleven, invents nothing in the clean
+files, and discards no quotation as unverifiable. Run it yourself with
+`npm run demo:fixture -- --fake`, which spends nothing. The paid half ran on
+2026-08-04 on the maintainer's subscription: a haiku smoke was correctly
+refused by the coverage law, the first full-strength run exposed a real
+product bug that was fixed the same day, and the two runs after the fix each
+found 8 of the 8 defects the fixture then planted, at the exact lines, with
+nothing invented and nothing discarded, for about a dollar and five minutes
+each. What no run settles is the judgment itself: whether those findings
+read as worth a person's time is the maintainer's call, described in
+[docs/plans/FG2-CHECKLIST.md](docs/plans/FG2-CHECKLIST.md), and it has not
+been made.
+
+All twelve work packages are built: the review engine, the data and git
+layers, every screen, the design pass, the accessibility checks, and the
 interactive engine mode that runs a review from files in your own terminal
-instead of a subprocess. The twelfth is partly done, and the last one is the
-judgment above, which only a real model can settle. What remains is written
-down item by item in
+instead of a subprocess. What remains open is judgment, not construction:
+the finding-quality verdict above and the design acceptance, both the
+maintainer's. The trail is in
 [docs/plans/M4-FINISH-PLAN.md](docs/plans/M4-FINISH-PLAN.md), and
 [docs/PROJECT-STATE.md](docs/PROJECT-STATE.md) is the current-facts cache.
 
@@ -81,9 +88,13 @@ allowlist. There is no code path that writes into a repository under review.
 grouped by severity beside the rule and the code they cite.](docs/images/confirmation-queue.png)
 
 Both screenshots are of the seeded test fixture, whose defects are planted on
-purpose and whose findings say so: no real model has reviewed real code here
-yet, and the Status section above says why. They are produced by the browser
-suite on every run rather than staged for the README.
+purpose and whose findings say so; they are produced by the browser suite on
+every CI run rather than staged for the README, so what they show is the fake
+engine's scripted review. The real-model smoke runs described in Status
+happened locally and their evidence, including every finding's text, lives
+under `review/`, which is deliberately not committed. No real model has
+reviewed a real project through this app yet: that is the next gate, and it
+is the maintainer's to drive.
 
 The panel worth looking at is Coverage. It says how many files, hunks and
 sweep hits were accounted for, which is what separates "nothing is wrong" from

docs/PROJECT-STATE.md

@@ -22,8 +22,10 @@ Last verified: 2026-08-03
 - A full audit on 2026-08-03 found the gate red on this machine, two pipeline
   promises that were decorative rather than enforced, and a records set two
   milestones behind the code. The verified gap inventory and the remaining
-  work to v1 are in `plans/M4-FINISH-PLAN.md` (items U1 to U13). Of twelve
-  build-plan work packages, six are complete and six are partly built.
+  work to v1 are in `plans/M4-FINISH-PLAN.md` (items U1 to U13). As of
+  2026-08-04 all twelve build-plan work packages are DONE; what remains is
+  the two maintainer judgments, the G1 verdict and the FG-4 design
+  acceptance.
 - The two external AI plans never arrived. They fold in as spec amendments if
   they do; they did not block the build.
 

docs/plans/BUILD-PLAN.md

@@ -43,12 +43,12 @@ to bottom; this table stays the summary of record.
 | WP-D | Ruleset domain: rule/directive model, composition function with snapshot tests, protocol importer + fidelity gate (permanent test), markdown export round-trip                                                                                                                         | WP-B | DONE (2026-07-30; import fidelity gate byte-exact, mutation-proven) |
 | WP-E | Engine: EngineAdapter, headless claude runner (CLAUDECODE-scrubbed env, explicit --system-prompt and --tools, full model ids), stream-json parsing, error classes, timeout guard, limit pause, model registry + cheap probing + auth status check per 06; fake claude binary for tests | WP-A | DONE (2026-07-30; prompt-injection isolation proven against the real CLI) |
 | WP-F | Pipeline: S0 prepare + sweeps, S1-S6 orchestration per 03 incl. the linked-review addendum (symbol dispositions enforced), ledger invariants, session strategy, S5 mechanical quote-match, job manager + SSE + resume/cancel/interrupt                                                 | WP-B,C,D,E | DONE (2026-07-31; `src/server/review/` and `src/server/jobs/`. Two gaps carried to M4: S6 runs no AI stage, and S4 never receives the pre-change file contents the bundle writes. U3 and D-52) |
-| WP-G | Seeded fixture repos (seeded-repo + seeded-core companion) + manifests + engine quality gate incl. linked-review gate wired into e2e (fake engine) per 05 section 4                                                                                                                    | WP-F       | PARTIAL (fixture builder, derived manifest and quality gate exist and are green. Missing: the duplicated-merge-helper defect from 05 section 4, the second cross-repo defect in the dedicated linked assertion, and the fixed-variant branch that should clear. U12) |
-| WP-H | UI: projects + add/clone + detail incl. dependency links; new-review setup with pre-flight and linked-project section; live run screen                                                                                                                                                 | WP-F       | PARTIAL (projects list and detail, links, new review with pre-flight and linked toggle all exist. The run screen shows five stages with no S0 or S6, no per-stage duration or tokens, no coverage panel and no retry. U6, U8) |
-| WP-I | UI: confirmation flow (keyboard-first) + report render/export; merged-branch badge + deletion flows                                                                                                                                                                                    | WP-H       | PARTIAL (confirm, dismiss with reason, complete, report, export, merged badge and deletions all work. The queue is one column rather than two panes, is not grouped by severity, and shows neither the rule text nor the diff hunk; the `e` and `g g` keys are unbound. U7) |
-| WP-J | UI: rulesets manager (list/detail/edit/version, import with fidelity report, export); settings                                                                                                                                                                                         | WP-D,WP-H  | PARTIAL (import, list, detail, per-rule enable with version bump, and export all work. The fidelity report is computed and discarded, unmapped lines do not block an import, severity is not editable, and settings lacks the data directory, engine default and danger zone. U9, U10) |
-| WP-K | Design pass to the 04 bar: tokens, themes, states, a11y (axe green), screenshots of every screen both themes into `review/`                                                                                                                                                            | WP-I,WP-J  | PARTIAL (both themes are complete and photographed, focus is visible, nothing scrolls sideways. No axe anywhere, no type or spacing tokens, three screens unphotographed, and the loading, error and long-content states are uneven. U10) |
-| WP-L | Full e2e suite per 05 section 3; real-engine proof run on the seeded fixture with captured report + usage; README (accurate, no aspiration)                                                                                                                                            | WP-G,WP-K  | PARTIAL (the journey and theme passes run in CI under `--build --e2e`; README made accurate 2026-08-03. The failure-path flows are covered only at the integration layer, and the real-engine proof run is unspent by choice. U12, U13) |
+| WP-G | Seeded fixture repos (seeded-repo + seeded-core companion) + manifests + engine quality gate incl. linked-review gate wired into e2e (fake engine) per 05 section 4                                                                                                                    | WP-F       | DONE (2026-08-04; all three missing pieces landed as rules 12-14, the fixed-variant branch, and the both-cross-repo assertion. Fixture plants 11, fake demo scores 11/11. D-60 to D-62) |
+| WP-H | UI: projects + add/clone + detail incl. dependency links; new-review setup with pre-flight and linked-project section; live run screen                                                                                                                                                 | WP-F       | DONE (2026-08-04; U6 rebuilt the run screen: S0-S6, per-stage duration and tokens, coverage panel, notes, errors, SSE fallback, cancel. U8 finished the new-review screen) |
+| WP-I | UI: confirmation flow (keyboard-first) + report render/export; merged-branch badge + deletion flows                                                                                                                                                                                    | WP-H       | DONE (2026-08-04; U7 finished the queue: severity grouping, rule text and diff hunk beside the finding, e and g g bound, aria-live) |
+| WP-J | UI: rulesets manager (list/detail/edit/version, import with fidelity report, export); settings                                                                                                                                                                                         | WP-D,WP-H  | DONE (2026-08-04; U9/U10 landed the fidelity block, unmapped-line refusal (D-48), severity editing, directives, bulk delete, and the settings screen's data directory and danger zone) |
+| WP-K | Design pass to the 04 bar: tokens, themes, states, a11y (axe green), screenshots of every screen both themes into `review/`                                                                                                                                                            | WP-I,WP-J  | DONE (2026-08-04; U10/U12: axe green over every screen in both themes, tokens, states, every screen photographed. FG-4 design acceptance stays the maintainer's) |
+| WP-L | Full e2e suite per 05 section 3; real-engine proof run on the seeded fixture with captured report + usage; README (accurate, no aspiration)                                                                                                                                            | WP-G,WP-K  | DONE (2026-08-04; failure paths walked in a browser, and the real-engine proof ran with the maintainer's go: two clean fable runs at 8/8 on the then-8-defect fixture, evidence and finding texts under review/2026-08-04-fg2/. The G1 verdict on finding quality remains the maintainer's) |
 
 ## 2. Milestones and founder gates