The instrument is guilty until proven innocent

scope: generic · severity: law · confidence: proven · subsystem: method

Evidence

taimen AGENTS.md §3b; taimen docs/HANDOFF-2026-08-19-CPUFREQ-SUSPEND.md §3

The largest time sink in a bring-up is not bugs. It is an experiment that ran, produced a clean-looking null, and never exercised the code under test.

The shape is always the same: nothing errors, nothing logs, and the null reads as a refuted hypothesis. You cross a theory off the list and move on. The theory was never tested.

A single session on taimen produced a whole page of these. Each one cost the time to design the experiment, run it, and reason about a result that did not exist.

So: before you believe a negative result, prove the path executed. Not the program — the path. A test that ran to completion and touched none of the code under test is indistinguishable, from the outside, from a test that ran and found nothing.

Corollaries with their own notes: every-test-needs-a-positive-control, a-module-parameter-that-does-not-exist-is-ignored, shipped-configuration-is-not-running-configuration.

This is the most portable thing in this brain. It is not about phones.