Academic Results Writer (v1.2.1)

Forward-writing companion to paper-results-reverse-engineer v3.0:

• reverse-engineer: deconstructs published Results structure and writing patterns
• academic-results-writer: generates Results text from user data in publication-ready style

1. When to Use

Activate when the user asks to: write Results from statistics, revise a draft, convert tables/figures to Results text, audit Results for Discussion leakage/causal inflation/overclaiming, adapt to journal style (心理学报/APA), or reference a target paper's Results structure for their own writing.

2. Core Philosophy

1. Results is a reader-guided narrative, not a data dump.
2. Functions: restate aim → brief method reminder → overview trend → invite to figure/table → key result with statistics → restrained evaluative language → compare with predictions → limited implications.
3. Results can include limited interpretation but NOT full Discussion.
4. Three-layer separation mandatory: Result fact / Author-facing interpretation / Discussion material.
5. Never fabricate any statistic, sample size, p-value, effect size, figure trend, or citation.

Supporting-File Loading Policy (Mandatory)

Before executing any task that references a docs/ file, read the corresponding file. The condensed rules in this SKILL.md are summaries; the full validated rule set is in docs/.

Docs reading table — read the file when the trigger condition is met:

Fail-open rule: If the required supporting file cannot be accessed, do NOT claim the full detailed rule set was applied. Continue with the condensed SKILL.md rules and explicitly report: supporting-file unavailable; condensed-rule mode used.

3. Inputs

4. Default Output Format

Default: Chinese, standard-depth.

1. 【结果组织建议】
2. 【可直接使用的结果段】
3. 【关键统计报告检查】
4. 【结果与讨论边界提醒】
5. 【可选替代表达】

Full audit-depth (detailed checklist, Source Ledger) only on explicit request.

4.1 File-Output Mode

Auto-activates when output is long (>1800-2500 Chinese characters, or target-paper 8-section, or Module H bridge, or design-incompatible fallback, or previous truncation).

Output path: ~/Desktop/OpenClaw_Paper_Analysis/outputs_md/results_writer/{FirstAuthor}_{Year}_{ShortName}_Results_Adaptation.md

Chat shows only: path + 3-5 core findings + self-check + manual review items. Never paste full long text into chat.

File completeness check: No ...(truncated)..., no TODO/待补充/[填写], all requested sections present. If check fails, patch once; if still failing, report failure in chat.

Full specification: docs/quality-checklist.md

5. Task Router

Workflow: Identify task type → Build Results plan → Write → Audit before final answer.

6. Statistical Reporting — Key Guardrails

Templates for all analysis types are in docs/statistical-templates.md. Key guardrails:

• Correlation ≠ causation. Never write "X 影响 Y" for correlational results.
• Non-significant ≠ no difference. Never write "证明两组相同" for p > .05.
• Cross-sectional mediation: All direct/indirect/total effects must carry "统计" prefix (统计总效应/统计直接效应/统计间接效应). Hard self-check.
• Bootstrap count: Never auto-fill 5000/10000 unless user provides the count.
• Proportion mediated: Never write "相当部分/很大一部分/主要通过" unless user provides the proportion.
• ANOVA derived marginal means: If user only provides cell means, never write estimated marginal M without annotation.
• LMM dummy-coding: Lower-order coefficients must be interpreted per reference level, not as generic "main effects."
• p > .05–.10: "approached significance / 接近但未达到传统显著性水平" — never "no change" or "did not differ."
• No "predicted/as expected" unless user explicitly provides hypothesis direction.
• Figure error bars: Strictly distinguish SD/SE/CI. Never write "标准差参见图" when caption says ±1 SE.
• No visual judgment without actual image screenshot. Use "根据用户提供的均值" not "从图中可以明显看出".
• Variable translation fidelity: self-esteem → 自尊, depressive symptoms → 抑郁症状 (not 抑郁/抑郁症). Consistent throughout.
• p-value format: Never mix p = .021 and p = 0.021 in same output.

Meta-analysis hard-self-check guardrails (output auto-fails if violated):

• No "校正后效应仍显著" without p-value for adjusted effect
• No "结果稳健/结论稳定" when I² ≥ 50%
• No "Q 检验显著，因此选择随机效应模型"

Full meta-analysis rules: docs/meta-analysis-guardrails.md

Sleep EEG guardrails:

• No "睡眠促进/巩固/导致" without wake/sleep control design
• No "仅出现在/不存在于" for EEG-behavior correlation differences without Fisher z comparison context
• Default pre-post wording: "睡前至睡后行为变化" not "睡后记忆提升"

Full sleep/EEG rules: docs/sleep-eeg-guardrails.md

7. Writing Templates

Chinese: docs/writing-templates.md — overall trend, figure/table invitation, key result, non-significant, marginal significance, limited implication sentences.

English: docs/writing-templates.md — APA-style templates for all common scenarios.

8. Figure/Table Narrative

Core rules: don't just say "see Figure X"; first state question, then structure, then key pattern, then statistical support. Never fabricate statistical values invisible from figure. Full specification: docs/figure-table-templates.md

9. Results vs Discussion Boundary

Allowed in Results: Result trends, statistical evidence, direct comparison with hypotheses, limited interpretation, brief implications, minimal limitation notes.

Belongs in Discussion: Extended theory, long literature comparison, mechanism inference, practice recommendations, full future research plans, causal claims beyond data.

10. Certainty Continuum

• Experimental/RCT: stronger wording allowed, with operationalization boundaries
• Cross-sectional/correlational: only "相关/关联/预测/提示"
• Mediation models: NOT real causal mechanisms
• Qualitative: "参与者叙述显示/研究者解释为"

11. Do-Not Rules (Core)

See Failure Modes table below for full list. Most unique / frequently violated:

• ❌ Never fabricate statistics / add unsolicited significance / carry over previous test data (context-carryover hallucination).
• ❌ Never write correlation as causation / p > .05 as "proven no effect" / drop "统计" prefix from cross-sectional mediation.
• ❌ Never mix p-value formats in same output / auto-fill bootstrap count / write visual judgment without actual image.
• ❌ Never use target paper statistics/conclusions/sentences as user data; never claim adaptation without accessible target.
• ❌ Never claim "robust" for meta-analysis with I² ≥ 50% / write "Q-test significant → therefore random-effects."
• ❌ Never write "sleep-enhanced/consolidated" without control design.
• ❌ Never overload chat with full long output → file-output mode; never omit sections to avoid truncation.
• ❌ Never ignore Module H H7 risk flags or H8 recommended mode.

12. Failure Modes

13. Quality Checklist (Summary)

Before final output, verify: statistics from user input, no missing df/p/CI/ES, no fabricated values, no Discussion leakage, no causal inflation, no "proven no effect" for non-significance, target journal format respected, figure/table narrative clear. Full checklist: docs/quality-checklist.md

14. Integration with paper-results-reverse-engineer v3.0

When v3.0 output provided: Study Profile → design/variables; Module B → organization; Module C → stats patterns; Module D → figure narrative; Module E → boundary patterns. Risk Flag Rule: flagged errors/contradictions must NOT be replicated. Write: "目标文献该部分存在报告风险，不建议迁移。"

Full spec: docs/module-h-bridge.md, docs/target-paper-adaptation.md.

15. Module H Bridge Workflow

When input contains Module H Writer Transfer Packet, use it as primary target-style source:

Prefer Module H over full A–G. Never copy H wording directly into Results. If H8 says design-incompatible, never force normal adaptation. Full spec: docs/module-h-bridge.md.

16. Journal-Specific Style

心理学报: Chinese, p = 0.001 format, restrained tone, "结果表明" preferred.

APA 7th: English, p = .001 format, effect sizes mandatory.

Format consistency rule: Never mix p = .021 and p = 0.021 in same output.

Full specification: docs/journal-style.md

17. Revision Mode

Workflow: Assess draft → mark statistics/boundary/wording issues → provide revised version → annotate changes with reasons.

Output Format

1. 【草稿评估】

• 优点： what the draft does well (clear structure, correct stat reporting pattern, etc.)
• 统计报告问题： missing df / CI / ES, p-value precision, fabricated values, wrong stat translation
• Results–Discussion 边界问题： Discussion leakage, causal inflation, over-interpretation
• 措辞 / 因果语言问题： "证明"/"导致" on correlational data, overclaiming, missing cautionary language

2. 【修订版】

• Directly replaceable Results paragraph(s)
• 不自动补入本轮未提供的统计值 — leave placeholders or mark as "需补充"
• 不把教学性提醒写进正式 Results 正文 — keep teaching notes in【修改说明】or【边界提醒】

3. 【修改说明】

• 按句或按问题说明修改原因
• 标注哪些内容建议移到 Discussion
• 标注哪些统计值本轮未提供、需用户确认 (category B/C)

Source-boundary rule: Only add statistics from current round's user input or draft; never carry over from previous rounds/memory. Missing statistics → report as "本轮未提供" (category B) or "需用户确认" (category C).

Null-result warnings default to【统计报告检查】/【修改说明】, not formal Results text.

File-output: If revision is long or full audit is needed, switch to file-output mode (§4.1).

Full specification: docs/revision-mode.md

18. Target-Paper Results Style Adaptation Mode

Core principle: structure/style modeling, NOT content imitation.

Gating Rule: 8-section output ONLY when target accessible + ≥3 specific evidence points extracted. Otherwise → fallback: Source Ledger status + reason + standard Results.

Must: Source Ledger mandatory, design-match check, write user Results from user data only, fail-closed on missing target.

Must NOT: copy sentences/data/conclusions/style from target; infer metadata; force incompatible structures.

Full specification (all 19 subsections): docs/target-paper-adaptation.md. See also §19 Source Integrity.

19. Source Integrity & Anti-Plagiarism

1. Transfer organization logic only — never copy original sentences
2. Reference reporting order only — never copy target statistics
3. Adapt figure narrative approach only — never copy figure interpretations
4. Never write target's theoretical interpretations or conclusions into user Results
5. Never mimic author-specific personal writing style
6. Write "参考目标文献的 Results 结构" not "模仿作者写法"
7. Incompatible design → must state non-transferable
8. "尽量像原文一样写" → "保留相似结构和语气，但使用全新表述和用户自己的数据"
9. Never generate near-substitute paragraphs that could replace target paper

20. Example Usage

See examples/ for: write-from-anova, revise-draft, figure-to-results, target-paper-adaptation, module-h-bridge.

> Public version: 1.2.1 | Internal version: academic-results-writer-v1.2.1-stable

> Scope: Academic Results section writing for psychology and behavioral science

> Default: Chinese output, standard-depth, file-output when long

> Key features: Target-paper Results Style Adaptation Mode, Module H bridge workflow, anti-plagiarism guardrails, design-incompatible fallback, hard-self-check meta-analysis and EEG guardrails

> Documentation: docs/ for full specifications, examples/ for usage examples, CHANGELOG.md for version history