# Analyze Experiment

> Use when an experiment has completed and its results need interpretation, comparison, or durable findings. Triggers on "analyze results", "分析实验", "interpret experiment", "what did exp find", "run analysis".

- Skill: `freemty/analyze-experiment` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add freemty/analyze-experiment`
- Raw SKILL.md: https://api.skillmd.com/api/skills/freemty/analyze-experiment/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: freemty (https://skillmd.com/u/freemty)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/freemty/analyze-experiment

---


# Analyze Experiment

Separate measurement from interpretation. Preserve raw outputs; write claims
only at the level supported by the experiment contract.

1. Read pipeline state, the experiment README/config, result files, and any
   existing summary.
2. Run the experiment's analysis script when present. Treat output as
   measurement, not explanation.
3. Follow `../../references/agent-routing.md` for the `domain-expert` role.
   Provide hypothesis, baseline, metrics, seeds, exclusions, and results.
   Require labels for evidence, inference, and unresolved questions.
4. Merge reviewed interpretation into the experiment README and
   `exp/summary.md`; update pipeline state to `analysis`.
5. Report the quantitative result, causal interpretation, limitations, and
   next discriminating experiment.

Generate slides only when requested. Route research talks through
`research-slides` so its default Speculative Decoding reference profile and
rendered QA apply.