Audit cron's 1-day window — widen scope to catch the 55 invisible failures?

Pattern. The nightly audit-newsletter-outputs.py cron runs with a ~1-day rolling --since. A file is only eligible when its date-prefix is >= since_date. A file that fails the audit and isn't repaired the same day permanently exits scope — never re-opened, and every later run truthfully reports Fail: 0. That clean signal is what /self-review step 1.5 was reading as "repaired."

Evidence (censused 2026-08-17, not sampled). A direct audit-newsletter-outputs.py --since 2026-06-01 run reports 55 live failures among files the daily cron currently shows as clean. The script itself is not broken — the false-negative hypothesis (Review 23's hypothesis a) was tested and disproven; a wide re-run behaves correctly. The defect is operating scope, not logic.

Already shipped this cycle, no approval needed. self-review/SKILL.md step 1.5 now carries a scope-shortcut guardrail: the pre-failure set is resolved per-file against the most recent run whose Since: covers that file, never against the newest run's fail count. That stops /self-review from reporting a false-clean, but it does not close the coverage hole itself.

Why this needs a click. Fixing the hole makes ~55 files appear as failures at once, and if this is ever wired to a --fix path that's a mass rewrite. Scope and whether --fix participates are founder-level calls, not silent-apply.

Acceptance criterion. A file that fails the audit and is not repaired within 24h still appears in a failure report on a later run, without anyone re-running the script by hand.