diagnoselisted

# diagnose — competing explanations, one decisive probe Explains why something behaves as it does. It does not fix it. Reach for it when two explanations both fit the facts and neither has been ruled out. If you can already reproduce and localise the failure, this is overhead — hand it to the `debugger` agent instead. **Spawning lanes is main-session work.** Without the ability to spawn, run the three lanes yourself in sequence: the discipline is the value, the parallelism is only the speed. ## The contract Seven things stay distinct from the first note to the final answer. Collapsing any two is how a trace becomes a guess with a table around it. | | | |---|---| | **Observation** | what was actually observed, restated exactly | | **Hypotheses** | competing explanations, deliberately different | | **Evidence for** | what supports each | | **Evidence against / gaps** | what contradicts it, and what is still missing | | **Best explanation** | the leader right now | | **Critical unknown** | the one missing fact keeping the top two apart | | **Discriminating probe** | the cheapest step that would collapse the uncertainty | Never return a fix-it loop, a generic debugger summary, raw worker output, or confidence the evidence does not carry. ## Run 1. **Restate the observation** precisely and name the tracing target. 2. **Generate three deliberately different hypotheses.** Use this partition unless the problem suggests a better one: - **code path** — the implementation