Use when a test fails, a bug is reported, an error is pasted in, or a fix did not work. Enforces reproduce-before-theorize, reading the actual error, narrowing by bisection, one falsifiable hypothesis at a time, and a stop rule instead of guess-loops.
