Attribute post-release errors by deployed version, not merge time

Post-release health checks should attribute errors by the emitted deployment version, not merely by a time window beginning at merge. This prevents rolling-deploy overlap from falsely assigning failures from retiring replicas to the new release. The method is most reliable when logs carry a deploy SHA/version; without that field, use replica/deployment IDs and activation timestamps and keep attribution qualified.

Method: 1. Confirm the public readiness response reports the intended full release SHA and healthy dependencies. 2. Verify each active replica emits the new deployment version. 3. Compare release-version-filtered core 5xx, unhandled exceptions, and client-fatal errors against an equal pre-release window. 4. Treat external probes and recent quiet windows as independent evidence; do not let them erase endpoint-specific degradation. 5. Separate expected/existing integration error families from core regressions.