The source run used fixed Kubernetes and Prometheus reads with deterministic ranking. All four causes matched their development labels at rank 1. This is a four-case development result; the 24-case model evaluation remains open.

Every retained report observation and referenced cleanup artifact was verified before export. The replay contains selected numeric facts and fixed log indicators. Raw logs, free-text summaries, original incident identifiers and local paths are omitted. A false log indicator means the pattern was not found in that bounded capture.

Source digests identify the retained private artifacts. This projection does not include those original bytes, and does not claim that a digest alone proves source truth. Provider cost, model latency and human investigation time have not been measured.