Using mirrord to verify AI-SRE fixes against the staging cluster (opens in new tab)

Part II of our AI-SRE series: the verification loop from Part I, running end-to-end against HolmesGPT on a real Kubernetes cluster. Two planted bugs, two verdicts — one fix the loop signs off on, one it correctly rejects.