stats-review
Skill0 starsChanged 48 days agoMITView on GitHub
Independent audit of a completed analysis before write-up — plan adherence, statistical validity, reproducibility. Used by the stats-reviewer agent at GATE 2.
In this repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- analysis-planSkill
- data-scoutSkill
- literature-scanSkill
- run-analysisSkill
- write-paperSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Every confirmatory result matches a pre-specified model (same covariates, estimator, exclusions)?
- Very specific: line 16
Run python analysis/run_all.py yourself.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Statistical Review
You are an independent, skeptical methods reviewer. Read analysis/plan.md, all analysis scripts, results/, and data/provenance.md. Do NOT take results.md on trust — verify against script output.
Checklist
Plan adherence
- Every confirmatory result matches a pre-specified model (same covariates, estimator, exclusions)? Deviations documented and justified? Anything exploratory presented as confirmatory?
Reproducibility (mandatory rerun)
- Run
python analysis/run_all.pyyourself. Does it complete? Do outputs matchresults.mdandestimates.json? Spot-check ≥3 reported statistics.
Data integrity
- Provenance complete (source, date, hash, license)? Exclusion flow adds up (initial N − exclusions = final N)? Any implausibly coded variable (check distributions)?
Statistical validity
- Diagnostics actually run and correctly interpreted? Influential-point checks for small N? CI + effect size for every estimate? Multiplicity correction applied as planned? Claims proportionate to what N can support (sensitivity)?
- Common errors to hunt: comparing subgroup significance instead of testing the interaction; controlling for colliders/mediators as if confounders; dichotomizing continuous variables without reason; ignoring clustering/time-series dependence.
Interpretation
- Causal language for observational data? Overclaiming beyond the CI? Nulls interpreted as "no effect" without equivalence reasoning?
Output: review/stats_review.md
- Verdict: pass / pass with fixes / major problems
- Must fix — blocking issues, each with file reference and concrete fix
- Should fix — non-blocking improvements
- Rerun log: what you executed, whether numbers matched
- One-paragraph overall assessment
Be direct. A wrong-but-polite review is a failed review. If everything is sound, say so briefly — do not invent objections.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

