Three, all of which cost real time.

A config push that reports success can have applied nothing you care about. A CLI reports on the fields it manages and stays silent about the ones it does not, so "up to date" was true and useless at the same time. The fix is to assert the OUTCOME — does the thing actually answer? — rather than trusting that a step ran.

A trailing slash on a resource identifier looked like a misconfiguration and was library normalisation. Spec-equivalent, upstream, not worth chasing. Check what the URL type does before hunting for the config that set it.

An endpoint that 404s on a bare GET is not down. Check the method the surface actually serves before declaring an outage — reading a 404 as a broken deploy cost a day of looking in the wrong place.