A running server can return HTTP 200 without returning your page. Introduce one reversible fault, follow the response and repair the request that failed.
At a million requests, moving from 99% to 99.9% allows 9,000 fewer missed reports. Choose the reliability work and customer promise behind that number.
Forty exports of margin looks different after a repaired fault than during a growing failure. Connect the count, its pace and the recovery work to release decisions.
A new worker can finish a hundred reports without touching the code you changed. Choose work that answers the release question, then plan how to stop and recover.
Restoring capacity is only the start of recovery. Follow a small queue calculation to see how new arrivals compete with old work, and how to test the prediction safely.
At 14:15, one region has delayed reports. Eight minutes later, you know about another. Keep people informed without promising a fix you cannot yet see.
A deployment looked suspicious until older-version jobs told a different story. Explain decisions using the clues and authority available when people made them.