The fraud gate was failing on files this session never wrote

Every HARD hit landed in pre-existing or auto-generated files: other shifts' deliverables, a backfilled archive, a music manuscript, dashboard data arrays, and the scanner's own documentation. The fix lives in the scanner, not in any content file.

Why the hits were false positives

Six structural classes, none of them fraud committed by the running session. Counts are the full-disk hit totals from the original scanner.

The six scanner fixes

Each fix is either an exclusion, a masking rule, or a scoping rule. No pattern's detection was weakened, and no flagged content was reworded.

Structural suppressors, measured

Full-disk sweep, before vs after. Self-documentation and dashboard data arrays drop to near zero. The templated-reuse storm collapses to genuine repeated prose.

Proof

Three real runs. The middle one is the honesty control: a planted fraud file must still fail, or the fix would be gaming.