Call-failure audits need separate coverage, symptom, and cause claims

Reusable audit lesson (observed 2026-09-07; provenance: re-running read-only lifecycle/transcript predicates during a voice-call failure investigation): a symptom rule applied only to a previously labelled candidate subset cannot establish its population-wide count. Freeze the time window and ID set, rerun the rule over the whole population, and reconcile overlaps before changing percentages. Separately report automated record coverage, individually reviewed examples, and recordings actually listened to. Error-before-disconnect establishes temporal order, not causation; recovered errors and expected business no-match responses are not automatically call-ending defects. Missing response events require a retention-completeness check, and output-queued markers do not establish that the remote caller heard audio. This is a methodological note, not a cache of production counts or a fault attribution.