fix(evals): bind compound handoff claims to native proof

[vriesd]

Why

Qualification rejected a truthful handoff saying that node scripts/verify.mjs passed on Linux and independent review passed. The evaluator treated the platform and review clauses as an unsupported command suffix.

What changed

• Keep command, review, and qualifier facts separate. Preserve quoted metadata and ordinary prose.
• Bind platforms from the canonical host registry to their own native evidence, including matching deferred captures. Passing claims retain zero-exit and non-observe requirements.
• Add live-handoff regression evidence, proof corruption controls, and multi-capture order tests.

Scope

Only the delivery evaluator and its regression evidence change. The packed 9.6.0 plugin, policy, catalog, case instructions, models, sampling, and release plan remain identical.

Tradeoffs

Command-local typed facts preserve claim ownership. A global text rewrite or broad integrity detector would reject unrelated implementation prose. Unsupported status grammar remains a failure.

Blast Radius

Maintainers can grade the observed compound handoff while retaining its native proof requirements. Wrong platform claims on passing checks and non-pass observations now fail against their own evidence.

Verification

• RED regression commits precede corrections. All eight actual saved campaign inputs pass unchanged with the repair.
• 422 focused tests pass. Independent controls cover native proof corruption, three genuine deferred captures, metadata, ordinary prose, and contradictory continuations.
• Full push preflight passes with 3,038 passing tests, 20 existing skips, and zero failures. Independent exact-head review and maintained packing confirm the approved package, policy, catalog, and plan are unchanged.

Original stopped campaigns remain failed. Offline replay supplies repair evidence, not live qualification.

[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 2026-10-09T10:33:14.612174Z	10421ad	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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 10421adc97

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".