fix(voice): retain additional evidence correction history

[seonghobae]

Correcting an additional Voice's truth state or evidence currently overwrites the earlier assignment. Historical cutoff reads then lose the state the reader previously saw. Serialize corrections with primary imports, retain the old half-open interval and PROV derivation, and give the new interval its own provenance identity. Exact retries leave the recorded assignment unchanged.

This PR is stacked on #780 and must remain open until that parent is protected-merged; then retarget to main and refresh exact-head checks, reviews, and rules. The parent's Post-read import repair is reused, not duplicated. No new schema, atomic Voice code, model policy, mathematical kernel, or release version is introduced. Sibling pagination/filter work remains in #934/#935.

Validation:

• Protected-main implementation reproduces the PostgreSQL history loss (one row where two are required).
• 14 targeted tests passed with full-schema synthetic PostgreSQL, real demo OIDC/JWKS, RBAC, hidden-evidence denial, Valkey, historical/live API truth, retries, rollback, waiting concurrent corrections, and primary history.
• Frontend lint and all 534 tests passed; production build passed. A further 66 ontology/schema/docstring contracts and 5 final documentation checks passed.
• Existing Storybook components audited at 1440 and 390 CSS pixels for retained Observed and corrected Proposed states; four synthetic screenshots preserve complete labels without document overflow.

The authenticated API test uses the real ASGI application with real backing services. Storybook screenshots are separate component evidence, not a browser-to-API deployment claim. Protected-main delivery and full authenticated browser acceptance remain unverified. The gap baseline and docs/development-loop-20260905-voice-history.json pin 119 PR heads, parent relationships, required rules, ten ADR-number collisions, six parallel release numbers, and route/model signature drift. The implementation evidence is pinned at b8dd36e713ea1cb123de272fe61314145448e818; final documentation head is e6c1d157dad9f0d26ee703de455b2106a77b34d9.

Synthetic authenticated k6 observations (1 and 4 VUs, 10 seconds each) measured 165.18/315.45 HTTP requests per second, 17.03/80.22 ms HTTP p95, and zero HTTP errors. Two Running Ask jobs were observed through the official contextual-orchestrator boundary. These are short diagnostics, not terminal-answer, saturation, SLO, or population-inference claims. No performance setting was changed. Temporary database/backend/Valkey were cleaned up; official data volumes were untouched.

Final-head Full test suite and Frontend checks are queued; CodeRabbit and Devin statuses are successful, with no formal independent approval. No merge has occurred. Parent #780 retains normal auto-merge; child auto-merge is deliberately not armed against its unprotected feature base.

Latest status refresh: head e6c1d157dad9f0d26ee703de455b2106a77b34d9 remains open but is now draft. Its Tests runs are cancelled/skipped, not successful. The draft admission condition explains the skipped run; no manual run cancellation was performed in this cycle. Preserve the concurrent draft state. Parent #780 remains open with normal squash auto-merge enabled. CodeRabbit explicitly skipped review because the target is not the default branch; its status is not independent approval. The committed evidence inventory is a timestamped earlier snapshot, not a claim that these current statuses passed.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 7e900979-406c-4282-89b0-e75aa3fd9b32

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[seonghobae]

Current-stack governance finding: this PR is Ready while still stacked on unmerged #780 (1d8fa267…), so its exact-head Tests run is queued even though normal merge is intentionally impossible before the parent lands. Preserve the Voice-history delta and predecessor evidence boundary, but return this dependent PR to Draft now. The Draft-transition run should cancel/retire the queued repository-local work under the protected lifecycle contract. Re-enter Ready only after #780 is protected-merged and this branch is non-force converged/retargeted to the new protected base; no predecessor GREEN or review transfers.

[seonghobae]

Intervening non-force head movement reviewed and adopted: b8dd36e… → e6c1d157… is one documentation/evidence-only commit adding docs/development-loop-20260905-voice-history.json and 105 baseline lines; it does not change the Voice-history implementation. The parent-first governance finding therefore still applies to the new exact head. The PR is now correctly Draft behind #780, and the Draft-transition Tests run should be treated as cancellation/admission evidence only, not product GREEN.