[heybeaux]

Novel experiment → demonstrated failure → fix

Linked SwarmLab PR follows immediately. Pre-registration c8bc120 (Spec69/exp63), configuration correction060e6b8 BEFORE first observation. This is new classifier-coverage AUTHORITY testing at mergeLayers, not replay of exp15/16/17 or cache binding/async drift/defaults.

• Fetched-main4be84ed: two identical baseline rows; 10/10 unconsented matcher/scope losses permitted, rejection0/10, exact accuracy14/25.
• Runtimebc879d6: same frozen25 scenarios/harness/seed; unsafe allow0/10, rejection10/10, accuracy25/25 and all secondary metrics1.
• General fix: require allowDowngrade:true on THIS overlay for opaque matcher changes or case-sensitive tool-scope contraction of enabled preceding effective rules. Set expansions, wildcard widening, flags order, descriptions, disabled retunes, reviewed consent and insertion order preserved. No unsound regex language inclusion algorithm.
• Focused tests13 red/6 pass ->19 green. Full release check builds/typechecks and 1144 tests pass, stored evidence50/50. Single evidence-ID expectation omission after adding mapping was corrected; first red log preserved. No runtime/threshold tuning.
• Exact source/dist/trace receipts in SwarmLab, 3 replay-verified37-event traces. Existing stored evidence is explicitly regression-only.

HOLD — deliberate migration review

Do not automatically merge: previously silently accepted matcher retunes (including genuine regex broadening) now require explicit operator consent. Owner must review migration. Full Rule[][] interface remains unchanged; old partial/deep merge prose is clarified, not implemented. No post-merge alias isolation, predictor calibration or production database claim. Exp60/61/62 existing candidates remain separately unmerged; not cherry-picked or certified. No .acr changes, no Territory refresh.

Paired PR: heybeaux/swarmlab#52

Independent bounded review found no correctness/safety regressions. Two extra unit controls added (reviewed disable and mixed wildcard). Public warning suffix now says “keeping stricter effective rule” rather than “keeping stricter builtin”; downstream literal-message matching should migrate. Empty appliesTo remains loadPack-invalid; no partial overlay contract expansion. Frozen experiment and runtimebc879d6 unchanged.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 80cadc10-99a1-4a7f-bd1b-bbb7e4d579d1

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.}

[heybeaux]

Verified final head88548b2: executable release-check SUCCESS run38524236053; full1146 tests and stored50/50. Paired SwarmLab52 executable CI SUCCESS run38524070942: full static corpus49claims105traces983assertions, exact receipt audit, fresh pinned real-Aegis exp63 GREEN (metrics inspected). Both CLEAN/MERGEABLE. Still HELD for owner migration review: opaque matcher retunes (even broadening) now require explicit per-overlay consent. No auto-merge. CodeRabbit success reports review skipped, not independent automated approval. Bounded local independent review found no correctness issue; report preserved in nightly ledger.