| name | experimental-design |
| description | Patterns and checklists for designing and refining experiments in this research template. |
Experimental Design Skill
When to Use
- The user is setting up or modifying experiments.
- The user is unsure which experiments are sufficient to support a claim.
- You need guardrails to avoid under-specified or irreproducible experiments.
Instructions
- Use
references/design-checklist.md for the minimal information every experiment needs.
- Refer to the existing structure (
scripts/run_experiment.py, configs/, launch/) when mapping designs to code.
When designing experiments:
- Ask what question the experiment should answer and what constraints apply.
- Use the checklist to ensure hypotheses, datasets, metrics, and configs are clear.
- Propose a small set of experiments (including sanity checks and ablations).
- Suggest concrete config changes and run commands, keeping changes small and reviewable.