Use original outcomes when cohort metadata changes over time

When sampling historical model failures across a schema change, require the original explicit model outcome rather than a newer redundant outcome field. Missing newer metadata can otherwise silently truncate the eligible time range. If capture identity is absent in old requests, recover it only through authoritative event-to-record links; exclude unresolved or conflicting bindings instead of using nearby timestamps or current successful retakes. Preserve the schema-era distinction in the audit. This supports faithful failure-cohort construction, not claims about production-wide accuracy or deployment environment when those tags were never recorded.

Applies when: Historical evaluation spans additions to derived outcome or identity metadata. Keywords: cohort schema evolution retry outcome traces provenance historical sampling