Pg reporting

PG Reporting

Use after a Pingouin result table exists or when the user asks to beautify/statistically polish output.

Load

Read:

• ../../references/supervision-gates.md
• ../../references/apa-output-template.md
• ../../references/pingouin-api-quickref.md
• ../../references/archive-contract.md when the result is part of an analysis run.
• ../../references/pingouin-optimization.md when inspecting result columns or corrections.
• ../../references/report-output-contract.md for manuscript or publication-ready output.

Inputs

Need one of:

• Raw Pingouin result table.
• A dataframe/result object available in the workspace.
• Full numeric values from the user.

If no numeric results are available, generate a fill-in template and refuse to invent numbers.

Workflow

1. Identify test family and output columns.
2. Verify sample size, statistic, df, p-value, effect size, CI, and correction columns.
3. Create a reader layer: finding summary and polished prose.
4. Create an evidence layer: titled APA table, assumption notes, corrections, effect sizes, and CIs.
5. Create a provenance layer: run id, source result, software versions, code path, and S0-S5 audit.
6. Add explicit limitation notes and separate them from interpretation.
7. Save the report under the active run's reports/ directory and record its source result file.

Output Style

For Chinese users, default to Chinese prose with English statistical symbols:

**结果。** 独立样本 t 检验显示，A 组得分显著高于 B 组，*t*(38) = 2.41, p = .021, Cohen's d = 0.76, 95% CI [0.12, 1.38]。

For English:

An independent-samples t test showed that Group A scored higher than Group B, *t*(38) = 2.41, *p* = .021, Cohen's d = 0.76, 95% CI [0.12, 1.38].

Formatting Rules

• Round statistics to 2 decimals unless the user asks otherwise.
• Round p-values to 3 decimals; use p < .001.
• Include multiple-comparison correction when present.
• Include effect sizes and CIs when present.
• Separate statistical result from interpretation.
• Mention missing assumption checks if they were not provided.

Guardrails

• Never convert a non-significant result into evidence of no effect without equivalence/Bayes justification.
• Never overstate clinical/practical significance from p-values alone.
• Do not infer direction from p-value; use group means or coefficients.
• If the Pingouin table lacks necessary values, ask for or compute them.
• End with one compact S0-S5 audit line unless the user requests publication-ready prose only.
• For manuscript output, use the full section contract instead of returning a chat-style paragraph alone.