a4560c3669
stateExtractField tries bold **Field:**, then a plain ^Field: line, first match wins. STATE.md's frontmatter carries a `status:` key that appears before the body's plain `Status:` line, so extracting "Status" from unstripped content returns the frontmatter value and never the body prose the test was written against. Scoping the lookup to stripFrontmatter() fixes it. The tempting fix was to assert the frontmatter enum instead, on the reasoning that a typed value beats matching prose. That would have been wrong and would have weakened the test: normalizeStateStatus maps both "Phase complete — ready for verification" and "Verifying Phase N" onto the same 'verifying' enum, so the enum cannot tell phase-complete from mid-verification, which is precisely the distinction this case exists to prove. Typed is not automatically stronger when the type conflates the cases under test. Case 4 carried the same collision and was passing only because normalizeStateStatus falls through to the raw text when no known pattern matches, so frontmatter and body happened to agree. Fixed alongside it rather than left for the next person to trip over. All fourteen conversions on this branch were audited against the same failure mode. The collision can only arise where the frontmatter key and the body field name are identical case-insensitively — Status is the only such field, since every other frontmatter key is snake_case against a Title Case body label. Refs #3057 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
456 KiB
456 KiB