The canary looks clean. Should you promote it?
A change is running on 5% of traffic. After eight minutes: error rate 0.19% on the canary versus 0.21% on the baseline, p50 latency slightly better, p99 marginally worse. CPU is lower. The deploy dashboard is green and the rollout is waiting for a decision.
The fix that looks right
Promoting because the canary error rate is lower than the baseline. That is the most persuasive possible number and it frequently means the canary is not receiving the traffic that produces errors — a load balancer that routes by session affinity, a canary excluded from the internal client that generates the difficult requests, or a health-check-heavy sample. A canary that looks better than production deserves more suspicion than one that looks slightly worse, because "better" usually means "different sample", and the difference is exactly the thing you were trying to test.