Measuring How Long a Silent Failure Lasted Requires Bounded Start and End Points

A blog post examines what it takes to turn a 'time broken before detection' figure into a valid duration. The author argues that a duration needs a bounded start, a bounded end, and persistent evidence to read both from. Using a machine whose push attempts failed for forty days, the post shows how timestamps in logs and repository history can date the window. It also warns that counts of retry events are not the same as spans of time, a common source of inflated durations. The author stresses that durable version-control history is the strongest evidence for silent failures.