diagnosing-bugslisted

# Diagnosing bugs — the loop comes before the theory - **Can:** run commands, read logs, add tagged instrumentation, and build throwaway harnesses to reproduce a failure. - **Must:** have a command that goes red on THIS bug before forming a theory, rank hypotheses before testing any, and remove every probe before declaring done. - **Cannot:** ship a fix without either a regression test or a written statement of why no seam exists for one. Every other skill in this kit assumes you know what is wrong. This one is for when you do not. ## 🔴 The rule the whole skill exists for **No red-capable command, no hypothesis.** If you catch yourself reading code to build a theory before that command exists, stop. Theories are free, plausible and unfalsifiable in that order, and the cost of a wrong one is not the time it takes to have it — it is the work you build on top of it. > **The shape:** a call hangs. Three theories arrive, each sensible: the scope is too wide, the > command is too heavy, the input is too large. The process gets restructured around the third one > and the hang stops, so the theory looks confirmed. It was not: changing the command changed the > timing, and the real cause — a pipe waiting for an EOF that never came — was untouched and still > there. Four long waits, one of them 32 minutes, and a restructuring that fixed nothing. > > The evidence was on screen the whole time: **39 bytes of output, the same number every run.** > "Zero useful output" was read as ab