nature-statistics

interface:
  display_name: "Nature Statistics"
  short_description: "Audit statistical design, analysis, and reporting"
  default_prompt: "Use $nature-statistics to audit the study design, analysis choices, and statistical reporting in this manuscript."

name: nature-statistics
version: 1.3.0
description: >
  Declarative manifest for the statistics-reporting workflow. SKILL.md uses
  this to keep the always-needed reporting rules small while loading specialized
  statistical risk checklists only when the manuscript section, figure legend,
  or reviewer comment requires them.

# Design note: nature-statistics is a conservative reporting/audit skill. The
# default path always needs source hierarchy and minimum reporting requirements,
# but heavier risk catalogs should stay on demand so a simple Statistical
# analysis rewrite does not load every checklist.

always_load:
  - references/source-basis.md
  - references/statistical-reporting.md

references:
  on_demand:
    - condition: target is the flagship journal Nature, or the user requests an exact Nature Article statistical submission/readiness audit
      path: references/nature-article-requirements.md
    - condition: target is Nature Machine Intelligence, or exact NMI legend statistics, source-data, reporting-summary, data/code, initial-submission, or production requirements affect the audit
      path: ../nature-shared/journal-formats/nature-machine-intelligence.md
    - condition: nested data, pseudoreplication, repeated measures, many comparisons, correlations, regressions, outliers, small samples, or overstrong p-value language
      path: references/common-failure-modes.md
    - condition: checking figure legends, panel-level n, error bars, stars, box/violin plots, source-data notes, or statistical annotations
      path: references/figure-statistics.md
    - condition: final audit QA, severity labeling, reviewer-risk summary, or response-to-reviewer drafting
      path: references/reviewer-checklist.md
    - condition: the same statistic is reported in more than one place, or interval terminology is in question — one metric at two precisions across table and text, SD/Std and similar abbreviation drift, confidence interval used where prediction interval is meant, and overlapping error bars described as outperformance
      path: ../nature-shared/core/consistency-sweep.md

# `nature-statistics` Skill

[中文说明](README.md)

`nature-statistics` audits, rewrites, or drafts statistical reporting for Nature / high-impact journal submissions, focusing on transparency, reproducibility, and alignment with study design rather than only whether p values are significant.

## What To Use It For

- Check whether the Statistical analysis / Methods section is complete.
- Rewrite statistical statements in Results, figure legends, and source-data notes.
- Distinguish biological replicates, technical replicates, repeated measures, fields/cells/subsamples, and independent experimental units.
- Identify pseudoreplication, nested data, multiple comparisons, interaction interpretation, overinterpreted correlations, and significance abuse.
- Cross-check numeric precision, units, and statistical terminology for the same statistic across the abstract, text, tables, and conclusion.
- Draft conservative responses or revision suggestions for reviewer statistics comments.
- Check the flagship `Nature Article` requirements for test tails, exact n, repeat counts, significant and non-significant P values, ANOVA F/df, and t-test t/df.
- Check `Nature Machine Intelligence` (NMI) legend-level n/error/test definitions, source data, applicable Reporting Summaries, and the separate initial versus final requirements.

## Typical Requests

- "Check whether this Statistical analysis paragraph is Nature-style."
- "How should I write the error bars, n, and p values in this figure legend?"
- "The reviewer says the statistics are insufficient; list revision options and draft a response."

## What You Need To Provide

- Sample size, replicate level, experimental unit, groups, statistical tests, and multiple-comparison method.
- Figures, legends, Results text, statistical output, or reviewer comments.
- Which facts are confirmed and which need conservative marking.

## Outputs

- Statistical reporting review with major issues, risk level, and facts needing author confirmation.
- Ready-to-paste Statistical analysis, Results, or figure-legend rewrite.
- A cross-section statistical consistency risk list.
- Point-by-point response draft for statistical reviewer comments.
- `AUTHOR_INPUT_NEEDED` checklist for missing sample size, test, or design facts.

## Boundaries

- The skill does not invent sample sizes, replicate counts, p values, effect sizes, model assumptions, or test names.
- It does not replace a statistician or a full reanalysis; raw-data modeling requires the original data and analysis code.
- Experimental units and independent replicates cannot be inferred from figures alone and need author confirmation.

## Related Skills

- `nature-figure`: put statistical information into figures and source data.
- `nature-response`: respond to statistical reviewer comments.
- `nature-polishing`: polish the English wording of statistical descriptions.

# `nature-statistics` 技能

[English](README_EN.md)

`nature-statistics` 用于审查、改写或起草 Nature / 高影响力期刊投稿中的统计报告文本，重点是透明、可复核、与实验设计一致，而不只是 p 值是否显著。

## 适合用它做什么

- 检查 Statistical analysis / Methods 中统计方法是否完整。
- 改写 Results、figure legends 和 source data 中的统计说明。
- 区分 biological replicates、technical replicates、重复测量、视野/细胞/子样本和独立实验单位。
- 识别伪重复、嵌套数据、多重比较、交互解释、相关性过度解释和显著性滥用。
- 交叉检查摘要、正文、表格和结论中同一统计量的数值精度、单位及统计术语是否一致。
- 根据审稿人统计意见生成保守回应或修改建议。
- 对旗舰 `Nature Article` 检查单尾/双尾、exact n、重复次数、显著与非显著 P 值、ANOVA F/df 和 t 检验 t/df。
- 对 `Nature Machine Intelligence` 检查图注中的 n/误差/检验、source data、适用的 Reporting Summary，以及初投稿与终稿的不同要求。

## 典型请求

- “帮我检查这段 Statistical analysis 是否符合 Nature 风格。”
- “图注里的 error bars、n 和 p 值怎么写更清楚？”
- “审稿人说统计不充分，帮我列修改方案和回复草稿。”

## 你需要提供

- 样本量、重复层级、实验单位、分组、检验方法和多重比较方式。
- 图表、图注、Results 文本、统计输出或审稿意见。
- 哪些事实已确认，哪些需要保守标注。

## 产出

- 统计报告审查：主要问题、风险等级和需要作者确认的信息。
- 可粘贴的 Statistical analysis、Results 或 figure legend 改写。
- 跨章节统计一致性风险清单。
- 审稿统计问题的逐点回应草稿。
- `AUTHOR_INPUT_NEEDED` 清单，用于标记缺失样本量、检验或设计事实。

## 边界

- 不会替作者编造样本量、重复数、p 值、效应量、模型假设或检验名称。
- 不直接替代统计师或完整再分析；如需重新建模，应要求原始数据和分析代码。
- 不能仅凭图像判断实验单位和独立重复，需要作者确认设计。

## 相关技能

- `nature-figure`：把统计信息落实到图件和 source data。
- `nature-response`：回应统计类审稿意见。
- `nature-polishing`：润色统计描述的英文表达。

# Common statistical failure modes

## Contents

- [P0: likely to undermine the result](#p0-likely-to-undermine-the-result)
- [P1: important reporting or interpretation risk](#p1-important-reporting-or-interpretation-risk)
- [P2: clarity and presentation risk](#p2-clarity-and-presentation-risk)


Use this file to identify reviewer-risk patterns. Do not accuse the authors; state the risk and the fix.

## P0: likely to undermine the result

### Pseudoreplication

Signal:
- `n` is reported as cells, images, fields, spectra, droplets, reads, technical wells, or repeated readings.
- The independent experimental unit is likely animal, patient, culture, donor, batch, plot, device, or experiment.

Risk:
- The test may overstate precision and produce artificially small p values.

Fix:
- Analyse independent units, aggregate technical subsamples, or use a hierarchical / mixed-effects model when justified.
- Show subsample-level spread visually without treating every subsample as independent.

### Uncorrected multiple comparisons

Signal:
- Many genes, proteins, time points, panels, groups, pairwise contrasts, or exploratory endpoints are tested.
- The manuscript reports selected significant comparisons only.

Risk:
- False-positive rate is inflated or the comparison family is unclear.

Fix:
- Define the family of tests and use an appropriate correction or distinguish prespecified primary comparisons from exploratory analyses.

### Wrong interaction inference

Signal:
- Authors say two effects differ because one comparison is significant and the other is not.

Risk:
- Difference in significance is not evidence of a significant difference between effects.

Fix:
- Test the interaction or directly compare effect sizes.

### Analysis unit mismatch

Signal:
- Design is paired, matched, blocked, longitudinal, nested, or repeated-measures, but analysis uses independent tests.

Risk:
- Dependence structure is ignored.

Fix:
- Use paired tests, repeated-measures analysis, mixed-effects models, blocking, or cluster-robust methods as appropriate.

## P1: important reporting or interpretation risk

### Significance-only conclusion

Signal:
- Result is described mainly by stars or p thresholds.

Risk:
- Readers cannot judge magnitude, uncertainty, or practical importance.

Fix:
- Add effect estimates, confidence/credible intervals, raw distributions, and exact p values where possible.

### Small-sample overclaim

Signal:
- Very small `n`, unstable estimates, or no uncertainty interval, but strong language.

Risk:
- Effect size and direction may be imprecise.

Fix:
- Use cautious wording and report the limitation directly.

### Normality or equal-variance assumption not supportable

Signal:
- Parametric tests are used on small or skewed samples with no rationale.

Risk:
- Assumptions may be unverifiable or violated.

Fix:
- Add assumption checks, use robust/nonparametric alternatives, or describe the limitation.

### Outlier handling is unclear

Signal:
- Points disappear, exclusions are mentioned vaguely, or outlier tests are named without prespecified rules.

Risk:
- Post-hoc exclusion may bias results.

Fix:
- State exclusion criteria, timing, number removed, and whether conclusions are robust to inclusion.

### Correlation or regression overclaim

Signal:
- Association is written as mechanism, prediction, or causality without design support.

Risk:
- Confounding, non-independence, and model extrapolation may be ignored.

Fix:
- Reword as association unless experimental or causal identification supports stronger claims.

## P2: clarity and presentation risk

### Error bars are undefined

Fix:
- State s.d., s.e.m., confidence interval, interquartile range, or other convention.

### `n` varies across panels but is not panel-specific

Fix:
- Give panel-specific `n` and define what `n` represents.

### Star notation is not defined

Fix:
- Define thresholds, correction status, and exact p values if available.

### Software is missing

Fix:
- Add software/package and version if supplied by the user.

### `ns` hides the result

Fix:
- Provide exact p value or state the reporting threshold and avoid interpreting lack of significance as proof of no effect.

# Figure statistics and legend alignment

Use this file when checking figure panels, legends, star labels, error bars, source data, or statistical annotations.

## Legend information each quantitative panel should provide

For each panel or panel group, check whether the legend states:

- what points, bars, boxes, lines, or shaded regions represent
- the exact definition of `n`
- whether `n` is independent samples, animals, donors, patients, cultures, experiments, simulations, fields, cells, or technical replicates
- summary convention: mean ± s.d., mean ± s.e.m., median with IQR, min-max, confidence interval, or model estimate
- test/model used
- paired/unpaired or repeated-measures status if relevant
- multiple-comparison correction if multiple contrasts are displayed
- exact p values or star thresholds
- whether source data include raw independent values or only summary values

## Plot-type checks

### Bar plots

Risk:
- Bars can hide sample size and distribution.

Fix:
- Prefer showing individual independent data points when feasible.
- If bars remain, require error-bar definition and panel-specific `n`.

### Box plots

Require:
- median line, box bounds, whisker rule, outlier display rule, and `n`.

### Violin plots

Require:
- what points represent, kernel/density caveat if relevant, and independent-unit definition.
- Avoid making dense cell-level violins look like many independent experiments.

### Time courses

Require:
- whether the same units are followed over time.
- Use repeated-measures or mixed models when inference uses all time points from the same unit.

### Heat maps / omics panels

Require:
- normalization, scaling, clustering distance/linkage if used, multiple-testing or FDR treatment for highlighted features, and whether rows/columns are selected post hoc.

### Regression / correlation panels

Require:
- correlation coefficient or model coefficient, uncertainty if available, sample size, independence of points, and whether the fit is descriptive or inferential.

## Star notation

Avoid legends that only say:

```text
*P < 0.05, **P < 0.01, ***P < 0.001
```

Prefer:

```text
Symbols indicate adjusted p values from AUTHOR_INPUT_NEEDED test with AUTHOR_INPUT_NEEDED correction for the comparisons shown: *p < 0.05, **p < 0.01, ***p < 0.001. Exact p values and sample sizes are provided in Source Data. n denotes independent AUTHOR_INPUT_NEEDED.
```

Only use this after the relevant facts are supplied.

## Source-data notes

Ask whether source data should include:

- raw independent-unit values behind summary panels
- exact p values and test statistics
- sample-size table per panel
- excluded-data notes if any
- code or script used to produce the panel, where central to the claims

## Figure audit output mini-format

```text
Figure statistics audit
- Panel:
- Current legend problem:
- Risk:
- Required fix:
- Suggested legend text:
```

# Flagship Nature Article statistical requirements

Use this checklist only when the target is the flagship journal **Nature** or
the user explicitly asks for a Nature Article submission audit. Keep the
general design checks in `statistical-reporting.md`; this file adds Nature's
current journal-specific minimums.

## Stage and scope

- At initial submission, statistical information must already be complete
  enough for editors and referees to assess the work.
- Production formatting is stage-specific, but missing definitions of `n`,
  replication, tests or error bars are scientific-reporting gaps, not cosmetic
  formatting issues.
- Do not claim that another Nature Portfolio title has the same checklist
  without checking its current instructions.

## Required Methods checks

Confirm that the Methods contain a statistics section that states:

- every statistical test used and the comparison or model it addresses
- whether each applicable test was one-tailed or two-tailed
- the independent experimental unit
- the definition of biological, technical and other replicates
- paired/unpaired or repeated-measures structure where relevant
- inclusion/exclusion criteria, randomization and blinding where applicable
- correction strategy for multiple comparisons

## Required numerical reporting checks

For every reported statistic or applicable figure panel, require:

- an exact `n` value; if `n` varies between experiments, report the individual
  values instead of a range
- a definition of every error bar or interval
- the number of times representative measurements or experiments were repeated
- exact values for statistically significant and non-significant P values where
  relevant
- for ANOVA, the F statistic and degrees of freedom
- for t-tests, the t statistic and degrees of freedom
- the exact comparison, test family and tail direction near the reported result

The general skill may additionally request effect sizes, confidence intervals,
assumption checks and multiplicity details. Those strengthen reporting but
should not be mislabeled as a verbatim Nature requirement unless supported by
the target instruction.

## Reporting Summary gate

Check whether the current Nature Portfolio reporting summary applies:

- life sciences
- behavioural and social sciences
- ecology, evolution and environmental sciences
- covered physical-science areas such as solar cells and claims of lasing

Cross-check the completed form against Methods, Results, legends and Source
Data. Flag contradictory sample sizes, exclusions, randomization, blinding or
software details.

## Audit table

Return one row per analysis or panel:

| Analysis/panel | Test and tail | Exact n and replicate unit | Error/interval | Exact P | Test statistic and df | Repeat count | Status |
|---|---|---|---|---|---|---|---|

Use:

- `pass` when all applicable fields are explicit and consistent
- `AUTHOR_INPUT_NEEDED` when a factual value is absent
- `not applicable` only with a short reason
- `blocked` when the missing information prevents interpretation or exposes
  pseudoreplication, undisclosed exclusions or incompatible analyses

## Official sources

Verified 2026-08-08:

- Nature initial submission, Statistical information:
  <https://www.nature.com/nature/for-authors/initial-submission>
- Nature Portfolio reporting standards:
  <https://www.nature.com/nature-portfolio/editorial-policies/reporting-standards>

# Reviewer checklist for statistical reporting

Use this file before final delivery. It converts the audit into reviewer-facing risk and concrete author actions.

## Severity labels

### P0 — must fix before submission or resubmission

Use P0 when the current statistical reporting or analysis logic could invalidate a central claim.

Examples:
- independent unit is wrong or undefined for a central claim
- paired/repeated/nested structure is ignored
- many comparisons are made without any correction or family definition
- interaction is inferred incorrectly
- exclusion/missing-data handling could change the result but is not disclosed
- analysis cannot be understood from Methods and legends

### P1 — important revision strongly recommended

Use P1 when the claim may be defensible but reporting is too weak for review.

Examples:
- exact sample sizes are missing
- error bars or box-plot conventions are undefined
- effect sizes or uncertainty are absent for key results
- software/package/version is missing
- p-value thresholds are used without exact values or clear correction status
- small-sample limitation is not acknowledged

### P2 — clarity or polish improvement

Use P2 when the issue is unlikely to change the conclusion but could frustrate reviewers.

Examples:
- inconsistent terminology for replicates
- figure legends repeat methods but omit panel-specific `n`
- `ns` labels are unclear
- statistical text is scattered between Methods, legends and supplementary notes

## Final QA questions

Before sending the answer, check:

1. Did we define the independent unit?
2. Did we distinguish biological and technical replicates?
3. Did we avoid inventing p values, tests, software, sample size, randomization, or blinding?
4. Did every central result claim map to a test/model or to a clear descriptive statement?
5. Did we identify whether multiple-comparison correction is needed or missing?
6. Did we state which issues are not assessable from the supplied material?
7. Did the proposed wording avoid causal or mechanistic overclaiming?
8. Did we include short `AUTHOR_INPUT_NEEDED` questions rather than broad requests?

## Reviewer-risk phrasing

Use neutral phrasing:

- `A reviewer may challenge whether the analysis treats the correct independent unit as n.`
- `The current legend does not make clear whether the plotted points are biological replicates or technical subsamples.`
- `The claim is stronger than the statistical evidence currently reported.`
- `The comparison family is not defined, so the correction status is unclear.`

Avoid accusatory phrasing:

- `This is fake significance.`
- `The authors manipulated p values.`
- `The analysis is wrong.`

Unless raw data and full design are supplied, prefer:

- `not assessable from the supplied material`
- `potential risk`
- `requires author confirmation`
- `should be clarified before submission`

## Response to reviewer support

When helping draft a response to statistical reviewer comments, use this structure:

```text
Reviewer concern
- [quote or short paraphrase]

Author-side action needed
- [analysis/text/figure/source data change]

Draft response
- We thank the reviewer for raising this point. We have revised the Statistical analysis section to define AUTHOR_INPUT_NEEDED and have updated Fig. AUTHOR_INPUT_NEEDED legend to report AUTHOR_INPUT_NEEDED. The revised text now states: "AUTHOR_INPUT_NEEDED".

Manuscript change
- Section / figure:
- Replacement text:
```

Do not claim that a new analysis was performed unless the user supplies the result or asks you to run it on data.

# Source basis for `nature-statistics`

This file keeps the skill conservative. It is a local summary of sources the agent should use to justify reporting checks; it is not a replacement for the target journal's current author instructions.

## Primary source hierarchy

1. **Target journal instructions and user-supplied protocol**
   - Use these first when the user provides them.
   - If a reviewer comment or statistical analysis plan gives a stricter requirement, follow that local constraint.

2. **Nature Portfolio reporting standards**
   - Nature Portfolio frames reporting transparency around the ability of readers to replicate and build on published claims.
   - Where relevant, manuscripts sent for review may require completed reporting summary documents.
   - For life sciences, behavioural and social sciences, ecology, evolution and environmental sciences, the reporting summary asks authors to provide details of experimental and analytical design that are often poorly reported.
   - Source: https://www.nature.com/nature-portfolio/editorial-policies/reporting-standards

3. **Nature / Nature Methods statistics guidance**
   - The Nature collection `Statistics for Biologists` groups practical guidance on statistical design, P values, power, sample size, error bars, multiple comparisons, nonparametric tests, experimental design, replication, nested designs, regression, outliers, and correlation versus causation.
   - Treat this collection as practical guidance for common failure modes, not as a single mandatory checklist for every field.
   - Source: https://www.nature.com/collections/qghhqm
   - For a flagship Nature Article, load `nature-article-requirements.md` rather
     than treating this general guidance as the journal's exact checklist.

4. **Study-type reporting guidelines**
   - Use these only when the study type clearly applies or the user requests them.
   - Examples: CONSORT for randomized trials, STROBE for observational studies, PRISMA for systematic reviews, ARRIVE for animal research, and field-specific community standards.

5. **Conservative statistical reporting practice**
   - If no field-specific rule is available, require enough information for a reader to identify the design, analysis unit, test/model, assumptions, correction strategy, sample size, uncertainty, and software.

## Implementation boundaries

The skill should not pretend that Nature has one universal statistical recipe. It should instead enforce transparent reporting and identify risks that reviewers commonly challenge.

Use careful language:

- Prefer: `The manuscript should define the independent experimental unit.`
- Avoid: `Nature requires this exact test.` unless the journal instruction actually says so.

When the user supplies a target journal or reporting summary, ask for or use that document before relying on generic guidance.

## Key reporting principles used by this skill

- Replication depends on the independent experimental unit, not merely the number of measurements.
- Statistical tests are interpretable only relative to the design, assumptions, and data structure.
- P values should not carry the full evidential burden; effect estimates, uncertainty, design quality, and reproducibility matter.
- Multiple testing changes interpretation and usually needs a declared correction or a clearly justified family of planned comparisons.
- Figures should expose the data structure: `n`, error-bar meaning, test/model, correction, and whether points represent independent units or subsamples.
- Missing statistical details should be surfaced as author questions, not silently repaired.

# Statistical reporting checklist

Use this file when drafting or auditing a Statistical analysis, Methods, Results, or Supplementary Methods section.

## Minimum information to extract

For each major analysis, identify:

- endpoint or response variable
- experimental groups / conditions
- independent experimental unit
- biological replicates and technical replicates
- repeated measures, paired observations, blocks, batches, sites, donors, animals, patients, plots, or model runs
- inclusion / exclusion criteria
- missing-data handling
- randomization and blinding, if applicable
- transformation or normalization
- test or model name
- assumptions checked or rationale for robust / nonparametric approach
- multiple-comparison correction or planned-comparison rationale
- reported estimate, uncertainty interval, p value, and sample size
- software, package, and version when available

## Statistical analysis paragraph structure

A clean manuscript paragraph usually follows this order:

1. **Software and environment**
   - Name software and packages when supplied.
   - Do not invent versions.

2. **Data summary convention**
   - State whether values are mean ± s.d., mean ± s.e.m., median with interquartile range, box-plot convention, or another summary.

3. **Sample-size and replication definition**
   - Define `n` for each experiment class.
   - Distinguish independent samples from technical readings or submeasurements.

4. **Test/model choice**
   - State which comparisons used which tests or models.
   - Explain paired vs unpaired, parametric vs nonparametric, repeated-measures or mixed-effects models where relevant.

5. **Multiplicity strategy**
   - State the family of comparisons and correction method, or explain that tests were prespecified and limited.

6. **Thresholds and exact reporting**
   - Use exact p values where possible.
   - If thresholds are used, define them and avoid star-only reporting.

7. **Exclusions and robustness**
   - State pre-established exclusion rules or mark them as missing.
   - Do not add post-hoc exclusions unless the user supplied them.

## Results wording rules

Prefer:

- `Treatment A was associated with a higher response than control (mean difference ..., 95% CI ..., p = ...).`
- `The analysis used animals as the independent unit; cell-level measurements are shown to display within-animal variability.`
- `The evidence is consistent with an increase, although the small sample size limits precision.`

Avoid:

- `proved`, `demonstrated conclusively`, `confirmed the mechanism`, based only on a p value.
- `highly significant` without effect size or uncertainty.
- `n = 300 cells` when the experiment actually has three animals or three independent cultures.
- `ns` as the only result.
- `data were normally distributed` without a clear basis, especially for very small samples.

## Missing-information labels

Use short factual labels:

- `AUTHOR_INPUT_NEEDED: define independent unit for Fig. 2c.`
- `AUTHOR_INPUT_NEEDED: state whether comparisons were corrected for multiple testing.`
- `AUTHOR_INPUT_NEEDED: provide exact p values or the thresholding rule used by the journal.`
- `AUTHOR_INPUT_NEEDED: state software/package and version.`

## Ready-to-paste skeleton

```text
Statistical analyses were performed using AUTHOR_INPUT_NEEDED. Data are presented as AUTHOR_INPUT_NEEDED unless otherwise stated. The independent experimental unit was AUTHOR_INPUT_NEEDED; technical replicates were averaged before inferential analysis where applicable. Comparisons between two groups were analysed using AUTHOR_INPUT_NEEDED. Comparisons among more than two groups were analysed using AUTHOR_INPUT_NEEDED, followed by AUTHOR_INPUT_NEEDED correction for multiple comparisons. Exact p values, test statistics and sample sizes are reported in the figure legends or Source Data where available. No data were excluded unless specified in the relevant Methods section.
```

Use the skeleton only after filling supplied facts. Keep placeholders if facts are missing.

---
name: nature-statistics
description: >-
  Audit, revise, or draft manuscript statistical reporting for Nature / high-impact journal submissions. Use when the user asks to check statistical analysis sections, p values, confidence intervals, sample size, biological versus technical replicates, randomization, blinding, multiple-comparison correction, model assumptions, figure legends, Results statistics wording, reviewer comments about statistics, or Chinese academic drafts needing publication-ready Statistical analysis text. Also trigger on general paper-statistics requests such as 统计审查、统计分析小节、统计方法、p值、样本量、重复数、多重比较、置信区间、效应量、图注统计、审稿人统计意见.
---

# Nature Statistics Reporting Skill

Use this skill to make manuscript statistics transparent, reproducible, and appropriately bounded. It is a reporting and review skill, not a substitute for a statistician reanalysing raw data unless the user supplies the data and explicitly asks for computation.

## Default stance

- Prioritize design transparency over decorative statistical language.
- Separate three questions: what was measured, what unit was analysed, and what inference was claimed.
- Treat the independent experimental unit as the default `n`; do not silently treat cells, fields of view, repeated readings, spectra, model runs, or technical replicates as independent biological or experimental samples.
- Prefer effect sizes, uncertainty intervals, sample sizes, and exact test definitions over significance-only phrasing.
- State missing information as `AUTHOR_INPUT_NEEDED` instead of inventing sample sizes, tests, software, corrections, exclusion rules, randomization, or blinding.
- If a journal-specific instruction, study-type guideline, or field standard conflicts with this skill, follow the more specific source and mark the source used.

## Accepted inputs

The skill may receive:

- a Statistical analysis / Methods subsection
- Results paragraphs containing test statistics or p values
- figure panels, legends, captions, or source-data notes
- reviewer comments about statistics
- author notes in Chinese or English
- tables of reported comparisons
- raw or summary data, only when the user wants a concrete reanalysis or figure-statistics check

If the input is partial, run a bounded audit and state which parts cannot be assessed.

## Workflow

1. **Classify the task.** Decide whether the user wants audit, rewrite, draft, reviewer-response support, figure-statistics alignment, or data-backed reanalysis.
2. **Extract the design.** Identify groups, treatments, time points, endpoints, blocking factors, repeated measures, randomization, blinding, exclusions, and missing-data handling.
3. **Define `n` and replication.** Separate independent experimental units, biological replicates, technical replicates, repeated measures, cells/fields/subsamples, simulations, and pooled observations.
4. **Map claims to analyses.** For each result claim, record the comparison/model, test family, assumptions, correction strategy, effect estimate, uncertainty, and exact p-value policy.
5. **Check common failure modes.** Use `references/common-failure-modes.md` when the text involves nested data, many comparisons, cell-level measurements, interaction claims, correlations, regression, outliers, small samples, or significance-only reasoning.
6. **Check reporting completeness.** Use `references/statistical-reporting.md` to verify that Methods and Results give enough information for readers and reviewers to understand the analysis.
   If the target is the flagship journal Nature, also use
   `references/nature-article-requirements.md` for its exact tail, `n`, repeat,
   P-value, test-statistic and degrees-of-freedom requirements.
   If the target is Nature Machine Intelligence, also use
   `../nature-shared/journal-formats/nature-machine-intelligence.md` for its
   legend-statistics, source-data, reporting-summary and stage-specific checks.
7. **Align figure statistics.** Use `references/figure-statistics.md` when figure legends, panel labels, stars, error bars, box plots, violin plots, source data, or supplementary figure notes are involved.
8. **Draft or revise.** Produce conservative, ready-to-paste text. Keep claims within the supplied design and evidence. Do not upgrade statistical association into mechanism or causality.
9. **Run final QA.** Use `references/reviewer-checklist.md` before final delivery for severity labels, unresolved author questions, and reviewer-facing risk.

## Output format

Unless the user asks for another format, return:

```text
Statistics review scope
- Input reviewed:
- Boundary / missing materials:
- Study design readout:
- Independent unit and replication readout:

Major statistical issues
- [P0/P1/P2] Issue:
  Evidence from supplied text:
  Why it matters:
  Fix:

Ready-to-paste revision
[Rewritten Statistical analysis / Results / figure legend text]

AUTHOR_INPUT_NEEDED
- [short factual questions only]

Reviewer-risk note
- What a statistical reviewer may still challenge:
```

For a clean drafting request with enough information, skip the long issue list and return:

```text
Draft Statistical analysis
[ready-to-paste text]

Reporting notes
- n definition:
- tests/models:
- multiple comparisons:
- software/version:
- unresolved fields:
```

## Red lines

- Do not invent p values, sample sizes, degrees of freedom, confidence intervals, software versions, correction methods, preregistration, exclusion rules, or power calculations.
- Do not recommend a statistical test as final when the unit of analysis or design is unclear.
- Do not accept `n = number of cells/images/measurements` as independent replication without checking the experimental hierarchy.
- Do not use “significant” as a synonym for important, large, causal, or biologically meaningful.
- Do not hide non-significant or weak results by rewriting them into stronger claims.
- Do not give medical, regulatory, or clinical-trial statistical advice beyond reporting checks unless the user provides the relevant protocol and asks for bounded manuscript wording.

## Related files

| File | Open when |
|---|---|
| [references/source-basis.md](references/source-basis.md) | You need the source hierarchy or want to justify why the skill emphasizes transparency, reproducibility, and design reporting |
| [references/nature-article-requirements.md](references/nature-article-requirements.md) | The target is the flagship journal Nature or the user requests its exact statistical submission checklist |
| [../nature-shared/journal-formats/nature-machine-intelligence.md](../nature-shared/journal-formats/nature-machine-intelligence.md) | The target is Nature Machine Intelligence or NMI-specific legend, source-data, reporting or stage requirements affect the audit |
| [references/statistical-reporting.md](references/statistical-reporting.md) | You are drafting or auditing Statistical analysis, Methods, Results, or Supplementary Methods text |
| [references/common-failure-modes.md](references/common-failure-modes.md) | You see nested measurements, many comparisons, interaction claims, correlation/regression, outliers, tiny samples, or overstrong p-value language |
| [references/figure-statistics.md](references/figure-statistics.md) | You are checking figure legends, panel statistics, error bars, stars, box/violin plots, source-data notes, or graphical reporting |
| [references/reviewer-checklist.md](references/reviewer-checklist.md) | You are finalizing an audit or preparing a reviewer-facing risk summary |
| [../nature-shared/core/consistency-sweep.md](../nature-shared/core/consistency-sweep.md) | The same statistic appears in more than one place, or interval terminology is in question: one metric at two precisions across table and text, SD/Std abbreviation drift, `confidence interval` used where `prediction interval` is meant, or overlapping error bars described as outperformance |

## Source hierarchy

Use sources in this order:

1. User-supplied manuscript, data, protocol, statistical analysis plan, reviewer comments, and journal instructions.
2. Nature Portfolio reporting standards and reporting-summary requirements.
3. Nature Methods / Nature Portfolio statistics guidance summarized in `references/source-basis.md`.
4. Study-type reporting guidelines where relevant, for example CONSORT, STROBE, PRISMA, ARRIVE, or field-specific community standards.
5. Conservative statistical reporting practice.

If the supplied material is insufficient for a defensible statistical recommendation, ask for the missing design facts or provide a bounded wording option rather than guessing.