What changed because a panel reviewed the plan — manually curated, with every finding linked to the deliberation that produced it.
Old and new credit writers are never eligible together; canonical uniqueness is installed and proven before intake reopens.
A PASS_AFTER_RETRY is always YELLOW and never improves first-attempt metrics; every test must maintain >=99% first-attempt passes over rolling 200; the allowlist expires after 14 days with an owned issue.
Every path that can create another job execution — including crash recovery — passes one retry-policy function with a hard attempt cap; customer-facing bad_input requires proven byte identity, never metadata.
Durable pre-attempt state, per-encounter isolation, actionable clinician status, and fenced publication ship together; active probes do not.
Discount selection and application are two phases bound by a quote-ID + dual-revision CAS — a checkout can never apply a price the customer wasn't shown.