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
- Identify test family and output columns.
- Verify sample size, statistic, df, p-value, effect size, CI, and correction columns.
- Create a reader layer: finding summary and polished prose.
- Create an evidence layer: titled APA table, assumption notes, corrections, effect sizes, and CIs.
- Create a provenance layer: run id, source result, software versions, code path, and S0-S5 audit.
- Add explicit limitation notes and separate them from interpretation.
- 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.
1---2name: pg-reporting3description: Convert Pingouin results into compact, polished APA-style or Chinese psychology result sections with tables, assumption notes, and uncertainty caveats.4---56# PG Reporting78Use after a Pingouin result table exists or when the user asks to beautify/statistically polish output.910## Load1112Read:1314- `../../references/supervision-gates.md`15- `../../references/apa-output-template.md`16- `../../references/pingouin-api-quickref.md`17- `../../references/archive-contract.md` when the result is part of an analysis run.18- `../../references/pingouin-optimization.md` when inspecting result columns or corrections.19- `../../references/report-output-contract.md` for manuscript or publication-ready output.2021## Inputs2223Need one of:2425- Raw Pingouin result table.26- A dataframe/result object available in the workspace.27- Full numeric values from the user.2829If no numeric results are available, generate a fill-in template and refuse to invent numbers.3031## Workflow32331. Identify test family and output columns.342. Verify sample size, statistic, df, p-value, effect size, CI, and correction columns.353. Create a reader layer: finding summary and polished prose.364. Create an evidence layer: titled APA table, assumption notes, corrections, effect sizes, and CIs.375. Create a provenance layer: run id, source result, software versions, code path, and S0-S5 audit.386. Add explicit limitation notes and separate them from interpretation.397. Save the report under the active run's `reports/` directory and record its source result file.4041## Output Style4243For Chinese users, default to Chinese prose with English statistical symbols:4445```markdown46**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]。47```4849For English:5051```markdown52An 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].53```5455## Formatting Rules5657- Round statistics to 2 decimals unless the user asks otherwise.58- Round p-values to 3 decimals; use `p < .001`.59- Include multiple-comparison correction when present.60- Include effect sizes and CIs when present.61- Separate statistical result from interpretation.62- Mention missing assumption checks if they were not provided.6364## Guardrails6566- Never convert a non-significant result into evidence of no effect without equivalence/Bayes justification.67- Never overstate clinical/practical significance from p-values alone.68- Do not infer direction from p-value; use group means or coefficients.69- If the Pingouin table lacks necessary values, ask for or compute them.70- End with one compact S0-S5 audit line unless the user requests publication-ready prose only.71- For manuscript output, use the full section contract instead of returning a chat-style paragraph alone.
Exekiel179