Compare landing and follow-up verification windows
In the deployment’s Checks section, match the check time to the baseline it used. Later checks can reach a different conclusion as more traffic arrives.
Read the four comparison windows#
Checks lists the four checks with each result: No regression, Regressed, Incomplete or Superseded, and when each read. A check without a result reads Not scheduled when an earlier check was Superseded, because verification stops at that check. Otherwise it reads Pending while verification goes on, or Not recorded once verification has ended. The checks read consecutive windows after landing:
| Check | Reads | Compares with |
|---|---|---|
| At landing | The first 5 minutes | The hour before the deploy started |
| 20 min | 5 to 20 minutes | The hour before the deploy started |
| 1 hour | 20 minutes to 1 hour | The hour before the deploy started |
| 1 day | 1 hour to 1 day | The same 23 hours one day earlier, ending at landing |
Each check reads a few minutes after its window ends. A landing check that finds no runtime inventory collected since landing reads again about 15 minutes after landing. A later check, or one that runs late, that finds no current inventory waits about 12 minutes for the next collection and reads once more. A check that finds an Amazon ECS service still replacing tasks waits the same way before it records anything. A signal regresses when its value moves past the signal's Tolerance (%) and Minimum change in the direction set by Worse when.