docs: correct corpus and CI claims with evidence

[ChunkyTortoise]

What

Correct README corpus and CI-enforcement claims, and include the linked extraction/retrieval evidence note.

Why

The README counted two metadata rows as cases and described CI checks as enforced merge protection. The correction identifies 200 cases in 202 JSONL lines and attributes the protection observation to the 2026-09-19 audit.

Changes

• Separate 28 scored fixtures, 72 lookup cases, and 44 pending fixtures; identify weighted field-level accuracy and unmeasured retrieval metrics.
• Use the retained replay values for all six document types.
• Remove the stale proof-card embed and include its replacement evidence links.
• Preserve the existing evidence note byte-for-byte, including historical local paths and its later-RA13 wording.

Testing

• Verified both approved SHA-256 hashes, base identity, corpus counts, all 16 local links and CI anchor, and all six score rows.
• README diff: 14 insertions, 18 deletions. Evidence note: 65 added lines.
• Whitespace and README metrics-ledger checks passed.
• Application tests and replay were not rerun locally for this documentation-only transfer; PR CI will report separately.

This PR contains exactly two documentation files. Broader RA12/RA13 dependencies remain open. No polish, workflow changes, deployment, or merge is included.

[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 2026-09-22T03:29:01.314328Z	7967e79	PR opened

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.