Missing trace-level grouping metadata does not by itself prove that historical images

Missing trace-level grouping metadata does not by itself prove that historical images are ungroupable. Before rejecting a historical evaluation frame, inspect authoritative application records for stable trace references and parent/child entity IDs. Join by exact IDs, propagate exposure exclusions across connected identity components, and keep image-byte/decoded-pixel/perceptual checks as a separate later gate. Runtime deployment provenance, source-record provenance, and statistical representativeness are distinct claims.

This applies when an application preserves an event/trace ID in a submission or result record, even if older trace payloads predate instrumentation. It does not apply when linkage is guessed from timestamps alone. Saved-result cohorts may omit abandoned attempts; current demo/deletion flags may differ from capture-time state; equipment identifiers may be imperfect proxies. Record the resulting estimand and stage attrition, and do not present nominal sampling weights as proof of a production-wide unbiased estimate.