cross-model-review §3: pre-existing-vs-regression is a baseline diff; applicability is a runtime-state question

[firaen22]

What

Two adjudication rules added to cross-model-review §3, for reviews of behavior-preserving refactor branches.

1. Pre-existing is a disposition you establish, not assume. §3 already reproduces findings before acting and carries a pre-existing-tracked disposition, but doesn't say how you establish pre-existing. On a refactor branch a reproduced defect can be one the branch faithfully carried over from master — counting it against the branch is a false regression. Reproduce the finding on the merge-base too: present on both → pre-existing-tracked, spun off as its own fix; gone on master, live on the branch → a real regression the branch owns. Reviewers don't see master (§2 packet is your inlined lines), so this classification is yours, not theirs.
2. Applicability is itself a runtime/deployment-state question, not a source one. A migration or identifier case-collision finding that no live data can trigger — empty store, writer never deployed — is an irreproducible false positive (§4's existing disposition), not a regression the branch owns. Reproduce against what actually shipped, not synthetic data the bug reproduces on but the running system never holds.

Why

Class-distilled from a mining pass over the owner's own sessions (no code taken): a moira-web behavior-preserving refactor where codex's "not merge-ready" and agy's "CRITICAL" both dissolved under reproduction as pre-existing master behaviors the branch preserved (0 regressions, 4 pre-existing quirks spun off); and a TG-bot review where codex's migration/case-collision findings were non-applicable because the store was empty and the writer never deployed.

Ships unprobed per the covenant. .github/checks.py green.

Revision note (2026-07-24)

Force-pushed after a same-day adversarial self-audit (independent reader diffing this branch against main) found two things to tighten, no change to point 1's substance:

• Point 2 originally read as new standalone doctrine ("adjudicate against runtime state... source-only adjudication over-counts"). Reworded to explicitly hook onto §4's existing "irreproducible false positive" disposition rather than introducing parallel vocabulary for the same bucket.
• Point 1's "spun off (§4)" was a dangling cross-reference — §4 ("The loop, bounded") is the packet/review-round-cap section and contains no spin-off mechanism. Dropped the pointer; the phrase now just says "spun off as its own fix."

🤖 Generated with Claude Code