Synthesis Code Audit

A code audit is a systematic quality measurement of a diff. It is not a judgment call about whether a change should merge — that is the reviewer's job (see synthesis-pr-review). An audit produces a scored report across 10 orthogonal dimensions, each graded independently. The report tells you where the quality problems are; the reviewer decides what to do about them.

Binding rules

1. Measure; don't decide. The audit grades the diff; merging is the reviewer's call. When another workflow (preflight, PR review) invoked the audit, return the findings to it.
2. Compare against a fresh base ref, never a stale local one: a stale base causes both false positives and false negatives.
3. Read every changed file in full, not just the diff hunks; the surrounding file shows whether the change makes sense.
4. Give the reviewer the complete requirements (frozen artifact universe, user outcome, acceptance criteria, exclusions, decisions) without priming the verdict. Challenge a mistaken premise with evidence and route amendments to the decision owner.
5. Grade changed behavior and its affected consumers. An outcome-blocking defect stays a blocker even when its lines predate the diff; record unrelated findings separately and never call unreviewed code clean.
6. Apply project conventions explicitly, including stronger fix-while-touching rules, and don't flag style nits in code this diff did not change.
7. Partial coverage is Incomplete, never Clean. Track reviewed and unreviewed paths; an UNKNOWN dimension or an unreviewed path cannot satisfy a readiness gate.
8. Bind reused findings to the actual diff, required consumer and evidence inputs; changed material reopens the affected judgment.

Contents