Reproducible Case

A reproducible case is a minimal, reliable procedure that produces a fault on demand. It is the fundamental unit of debugging, because everything else — bisecting, diagnosing, verifying the fix — depends on being able to trigger the behaviour at will.

Building one is often most of the work. The steps that matter are usually to fix the environment (a clean database, a restarted server, a known input) and then to remove everything that does not change the outcome. What remains is both the diagnosis and, frequently, the test.

A differential repro is the strongest form: two builds identical except for one variable, each producing a different observable. It converts "something in this upgrade broke fonts" into "this version emits the preload tags and that one does not", which is a fact rather than a theory. See Ground Truth.

Two habits keep repros honest. Write the reproduction before the fix, so you have seen the red — see Falsifiability. And distinguish a genuine repro from an intermittent one: a fault that appears once and clears on retry is a Flaky Test until proven otherwise, and reasoning about the code from a single occurrence usually wastes the afternoon.