Scientific research, statistics, data analysis, epidemiology, and market intelligence.
A .skill file (official Claude Skill package). Import it directly into Claude Code via /skills or unzip it into ~/.claude/skills/.
Download research-statistics-engine.skillNo sign-up. No analytics. Just a direct download.
It makes Claude operate at the standard of a strong-journal peer reviewer and a top-consulting lead analyst — rigorous, transparent about uncertainty, useful for real decisions. It triggers automatically on requests like "is this statistically significant?", "analyze these numbers", "how big is this market?", "what does the literature say about X?", "design a cohort study". No need to invoke it manually.
The user hands over data (a file, a pasted table, a list of numbers) and wants analysis. Inspection before modeling (an EDA script included), assumption checks, deliberate choice of method, execution in real Python, interpretation with effect size + confidence interval, honest limitations.
A question with no dataset: "what does the research say about", "how big is this market", "design a study". It searches for evidence in primary sources (PubMed, IBGE, DATASUS, IPEA, OECD, World Bank, WHO, BACEN), validates date/methodology/bias, reconciles conflicts without lazy averages, always cites.
They exist because a confident, wrong answer in this domain leads to genuinely bad decisions.
No invented sources, no guessed numbers, no placeholder citations. If the number doesn't come from the user's data or a real source consulted, it isn't stated.
FACT (comes from data or a source), INTERPRETATION (a reading), HYPOTHESIS (a testable proposition), SPECULATION (an informed guess). The reader never has to guess.
Causal language only when the design allows it (RCT, natural experiment, causal-inference method) — and the design is made explicit.
Confidence intervals, effect sizes, samples, and p-values appear together. Never a p-value alone. Never an effect size without an interval.
Meta-analysis > systematic review > RCT > cohort > case-control > cross-sectional > institutional dataset > industry report > opinion.
Every analysis ends with what could be wrong: confounding, bias, small n, selection, generalization, data quality.
SKILL.md The full specification: triggers, workflows, principles, a statistical-test router, output templates, a rigor checklist. references/methods.md A catalog of statistical and ML methods: when to use them, assumptions, formulas, evaluation metrics, a preprocessing guide. references/data-sources.md Brazilian data sources (IBGE, DATASUS, IPEA, Fiocruz, SciELO, BACEN) and international ones (PubMed, WHO, OECD, World Bank, OWID) mapped with reliability and data type. scripts/eda.py An automated exploratory data analysis script. Runs on CSV/Excel: shape, dtypes, missingness, descriptives, distributions, outliers, correlations. Click "Download research-statistics-engine.skill" above.
Move the file to ~/.claude/skills/ (or unzip it as a folder in there). Claude Code loads it automatically.
The skill triggers on its own when the request involves data or evidence. Ask "is this statistically significant?" and it activates.
We curate components for applied AI. If your skill works in production and you want to publish, talk to us.