Pingouin Stat

Pingouin Stat

This is the main entry point. Use it instead of asking users to choose individual sub-skills.

Load

Read in order:

1. ../../references/intake-checklist.md
2. ../../references/workflow-index.md
3. ../../references/supervision-gates.md

Read ../../references/archive-contract.md before generating code or a deliverable that must be reproducible. Read ../../references/workflow-contract.md for a resumable multi-command run. Read ../../references/pingouin-optimization.md before generating concrete Pingouin code.

Read ../../references/pingouin-api-quickref.md only when generating code. Read ../../references/apa-output-template.md only when writing final prose or tables.

Workflow

1. Intake: clarify only missing items that affect the analysis.
2. Route: choose one or more downstream skills from workflow-index.md.
3. Screen: use pg-data-screening when data shape, missingness, or assumptions are unknown.
4. Analyze: call the selected analysis workflow or generate the relevant Pingouin code.
5. Approve: run pg-analysis-approval before final interpretation.
6. Organize: use pg-reporting to produce the requested deliverable shape.
7. Archive: initialize a run under archive/analysis-runs/, save the exact code and numerical outputs, then write the report and S0–S5 audit record.

Intake Contract

If key information is missing, ask a compact question set covering:

• Research hypothesis.
• Data path or available columns.
• Outcome, predictor, subject ID, covariates/mediators/raters/items.
• Between/within/repeated/nested structure.
• Planned or exploratory tests.
• Output format: Markdown, Word/docx, PDF, LaTeX, CSV tables, notebook/script.
• Figure/table requirements.
• Language and reporting standard.

If the user says "你先决定" or provides enough context, proceed with explicit assumptions.

Routing Output

Before analysis, produce this short plan:

Intake: <known + missing/assumed>.
Route: <skill(s)> -> <Pingouin function(s)>.
Approval gates: S0-S5 will be checked before interpretation.
Deliverable: <format/language/tables/figures>.
Next: <ask question / run screening / generate code>.

Result Assembly

Final result should include only the sections that fit the user's requested format:

• Analysis plan.
• Data screening summary.
• Main statistical result.
• Follow-up comparisons.
• Figure/table code or rendered table.
• APA/Chinese prose.
• Approval findings.
• Reproducible Python code.

Guardrails

• Do not skip approval when producing conclusions.
• Do not invent results when no data/output exists.
• Do not force Pingouin onto designs requiring mixed-effects models, SEM, count models, survival analysis, or complex survey models.
• Use short S0-S5 audit codes by default; expand only failed gates or user-requested rationale.
• Do not finish a code-generating run without saving the code, output references, and audit record under one archive run directory.