[abruption]

Summary

Add a synthetic Windows identity lab for W1–W3, with a closed case manifest and independent classification of file identity, ownership/security and process instances. Missing, contradictory or changed observations remain unknown/refused; every record keeps safeToUnlock: false.

The lab uses no-follow handles and 128-bit file IDs, records SID classes rather than raw account SIDs, binds PID/FILETIME evidence to held or fresh-open handles per case, and verifies all registered cleanup handles before deletion. Fault cases preserve partial evidence and unconfirmed children. PID reuse that is not observed is partial coverage (classifier exit 4), not verified coverage.

Add separate hosted CI gates for mock tests, PowerShell 5.1 parsing/C# compilation, NTFS synthetic observations, and five fault modes. Raw lab output stays private; the workflow prints classification summaries and explicitly handles exit 4.

Refs #146, #147, #148. Milestone: v0.4.7.

Validation

• Local: npm ci, check, lint, test (263 pass / 3 Linux skips), and package smoke passed.
• Classifier: 39 mock/negative tests passed, including truncated runs, case/handle/PID binding, environment contradictions, exhausted PID-reuse budgets, fault-record identity and cleanup failures.
• Independent read-only CLI review of the frozen handoff: Claude Code 2.1.302, requested and response-metadata model claude-opus-5-5; Codex CLI 0.163.0 isolated default, actual model not separately reported. Both completed with the same frozen input SHA-256 80c5d81e124061b7fd1bfff6a43ccdddac2c32ee7e36a43ad73a991c36de3e05.
• Integrating review corrected PID-reuse handle binding, environment/control consistency, bootstrap error reporting, public CI output and exit-code handling; focused regression tests and repository checks cover the resulting revision. The CLI reviews themselves were source-only and preceded those corrections.
• Windows compile/synthetic results are supplied by this PR's CI. Local macOS tests do not establish Windows API behavior.

Limits

Research-only files are excluded from the npm artifact. Production behavior, supported platforms and unlock policy are unchanged. ReFS, Medium/High pairing, another user's files and native Codex remain unverified. Cleanup counts represent accepted delete dispositions; another handle can delay physical removal. No host policy, existing ACL, account or volume is changed. No merge or release is requested.

[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 2026-10-18T03:59:53.622548Z	374a3d7	Draft marked ready
🔒 Security Review	✅ Completed 2026-10-18T04:00:25.969847Z	374a3d7	Draft marked ready

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: ebe4948a12

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".