Strong Inference

Hypothesis-driven debugging generates 2-4 competing, mutually exclusive hypotheses about a software issue and designs killer experiments to decisively confirm or eliminate each one, systematically isolating the root cause.

To investigate performance problems, apply strong inference by generating multiple competing hypotheses for the behavior, designing decisive tests to eliminate incorrect causes, and tracking evidence in a persistent investigation log.

Yes, the methodology supports collaboration modes for both Codex and Claude, providing templates for hypothesis generation and verification patterns to enhance the root-cause investigation workflow.

Describe your software problem and run the /strong-inference command to begin the investigation, which enforces hypothesis generation, experimental design, and continuous evidence tracking.

The best way to find root causes is generating multiple competing hypotheses, designing killer experiments to decisively eliminate invalid causes, and maintaining a persistent task log to track investigation progress and evidence.

Avoid hypothesis-driven debugging for simple, obvious software issues where the root cause is immediately apparent, as generating multiple hypotheses and experiments adds unnecessary overhead to straightforward investigations.