[YuzeJ21]

Saved reviews retained historical analyses, but inspecting them required a compatible active comparison. This adds an explicit read-only history selector and historical JSON/Markdown handoff, including with changed unconfirmed criteria and no active analysis.

Selections bind the full saved record, owner, history position and complete analysis fingerprint. Locked reloads and deferred-download leases reject stale choices; reopening clears selection. Reports preserve exact heads and known/unknown revisions, distinguish original producer identity from current validation, and state evidence limitations. Missing provenance blocks full reports; historical status never grants current acceptance.

Missing persisted producer versions remain unknown in saved formats 1–7; explicit values, new-review defaults and strict active/owner equality stay intact. Constant preview-only styles wrap long values and contain horizontally scrollable tables without hiding evidence or enabling report HTML. Browser checks retain strict geometry, verify every table’s last-column access, follow the existing draft-replacement guard through its visible control, and test reset state including hidden DOM nodes.

Validation at reviewed head 5a7f9b724550b8805c439ae5bceadfc14f4c3254:

• Independent agent review of the entire branch and every subsequent correction: no outstanding findings.
• Exact-head CI: all five engineering lanes and required verify passed, including Python 3.11/3.13 and Windows compatibility. Engineering: 3,764 passed / 14 skipped, 95.33% coverage. Required CodeQL and both analysis jobs passed.
• Fresh local installed-browser checks: 17 passed (three installed-wheel journeys, eight constructed selector checks, six helper/capture checks). Hosted browser: 17 passed. Download contents, history-only access, current-state/byte preservation, desktop/phone/CSS-200% geometry and stale/reset boundaries are covered.
• All 83 Python files match commit/workspace/wheel/isolated installation; dependency checks, CLI checks, both deterministic benchmarks (12 + 2 constructed cases), lint/lock/diff checks and seven frozen R-002 hashes passed.

Squash-merged with an expected-head guard to main 9f7a0b86316063e1fa001d9179d1c67920c42d22. The entire main tree equals the reviewed tree; both local checkouts are synchronized and clean, and the same package was rebound to main. Main CI, main CodeQL and base advance passed. Main engineering: 3,764 passed / 14 skipped, 95.32% coverage; browser: 17 passed.

Record v7, current criteria confirmation/gates/comparison behavior and frozen R-002 remain intact. Source stays 0.2.5 unreleased; public v0.2.4 remains the latest release. Fixtures and browser evidence are constructed automation; agent review is not a qualifying practitioner review. Genuine independent use remains unverified and Stage 3 dormant. No personal testing, participant contact, release, deployment or public post was needed.

Design · Audit and limits.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T09:10:52.410218Z	5a7f9b7	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.