The Bad Deployment Hunt (Binary Search)

P1 INCIDENT: Production Latency Spike SRE Team: "At 02:00 UTC, our API p99 latency jumped from 50ms to 800ms. We know the bug was introduced in one of the last 500 deployments. We have an internal API check_deployment(commit_id) that tells us if a specific commit is 'bad' (has latency) or 'good' (healthy). Constraint: The CI/CD system will rate-limit us if we make more than 20 API calls. Find the exact bad commit without checking them one by one."