Experiment Design

Design the smallest experiment that can meaningfully reduce uncertainty. The goal is not maximum benchmark performance; it is to decide whether the claimed mechanism works.

For detailed checklists, read references/experiment-design-checklist.md.

Workflow

1. State the hypothesis in falsifiable form.
2. Define the minimal experiment that can test it.
3. Define fair baselines.
4. Define primary and secondary metrics.
5. Identify confounders and controls.
6. Add mandatory sanity checks.
7. Define decision criteria before running.
8. Recommend run order from cheapest to most expensive.