Paper Write Nature Docx

---

name: paper-write-nature-docx

description: "Draft a Nature-style paper as Markdown for Word export. Use when params.output_format == 'docx'；区别于 paper-write-docx：只用于 Nature 风格 docx 模式。"
argument-hint: [venue-or-section]

allowed-tools: Bash(*), Read, Write, Edit, Grep, Glob, Agent, WebSearch, WebFetch

---



# Nature-Style Paper Writing — Markdown for Word (docx mode)



Draft a Nature-quality paper as Markdown: **$ARGUMENTS**



> docx-mode counterpart of `paper-write-nature`. Keeps hourglass structure, claim-evidence-boundary paragraph architecture, reader-first ordering. Produces **`paper/main.md`** only.

>

> ⛔ **NEVER produce `.tex`, run XeLaTeX, or use LaTeX commands.**



## Constants



- **TARGET_VENUE = `Nature`** — Override via Additional Parameters.

- **MAX_PAGES** — Nature Article: ~5 pages main + Methods (override via Additional Parameters). (enforced via word-budget check below)

- **ANONYMOUS = false**

- **CUSTOM_REQUIREMENTS** — highest priority.

- **REVIEWER_SCRIPT** — external reviewer.



## Inputs



1. PAPER_PLAN.md — outline with claims-evidence matrix

2. RESULTS.md / experiment_results.md / figures/all_results.json

3. figures/ — `.png` / `.pdf` (Nature figure aesthetics from `paper-figure-nature` step)



## Core Architecture



### 1. Identify paper type first

Before writing, determine:

- **Research paper**: why phenomenon matters → what was done → what was found → what it means

- **Methods paper**: does method work → reproducible → better under fair comparison

- **Hypothesis-based**: establish or rule out a causal explanation

- **Algorithmic/device**: propose tool/system → show reliable, advantageous performance



Don't use one narrative for all paper types.



### 2. Reader-first writing order

Write for the reader's cognitive sequence:

1. Is this relevant? (Introduction hook)

2. What's new? (Contribution)

3. Do I trust it? (Results + Methods)

4. Can I reuse it? (Methods detail + Data Availability)

5. What does it mean? (Discussion)



### 3. Productive writing order

1. **Results** — anchor in evidence

2. **Introduction** — frame the gap

3. **Title** — crystallize contribution

4. **Discussion** — interpret + bound

5. **Methods** — reproducibility

6. **Abstract** — last



### 4. Hourglass structure

- **Introduction**: open broad → narrow to gap → state question

- **Discussion**: widen again → connect → explain how gap was filled



### 5. Paragraph architecture: Claim-Evidence-Boundary

Every paragraph:

- **Claim** (topic sentence)

- **Evidence** (data, comparison, literature)

- **Boundary** (limitation, scope, transition)



### 6. Boundary language

Express limitations honestly:

- "These findings hold under [conditions]"

- "We do not claim [X] generalizes to [Y]"

- "The generalizability is limited by..."



## Section Responsibilities



### Title

- ≤ 75 characters including spaces (Nature guideline)

- Searchable, specific, restrained, defensible

- Pattern: `[Core entity] in/through/by [mechanism or context]`

- No vague hooks, no unverified "first"



### Abstract (150–200 words)

Mini-paper structure: context/problem → gap → approach → key result with numbers → implication



### Introduction (~600–800 words for Nature)

- Hook: why the topic matters broadly

- Known: what is established

- Gap: what remains unresolved

- Aim: what this study asks/does

- Value: brief indication of approach and significance

- Do NOT summarize Results or Conclusion here

- Short paragraphs OK (Nature style allows 3–4 sentence paragraphs)



### Results

- Past tense: report what was observed

- Orient reader to figure/table → state main observation → quantitative detail → patterns

- **Figure numbers MUST be cited explicitly (every paragraph names its "Fig. N"), but VARY the citation pattern — don't open every figure with the same "Fig. N shows/presents…" skeleton.** Rotate: parenthetical ("…dropped 4.2% (Fig. 3)"), sentence-start, verb-led ("Comparing the curves in Fig. 4…"), figure-as-subject (when it carries a real quantitative finding). The number always appears; only the sentence structure rotates. Adjacent figures must not use the same pattern.

- Results = what happened, NOT what it means

- Each result paragraph tied to a specific figure or table

- Active voice preferred: "We observed..." not "It was observed that..."



### Discussion

- Restate main finding → plausible explanations → compare with earlier work → limitations → implications → future work

- Short rule: Results = what we observed; Discussion = how we understand it and when it may fail

- Three-part close: contribution → key evidence → implication with boundary



### Materials and Methods

- Specific, complete, transparent, reproducible

- Another group must determine: ethical conformity, materials/conditions, key parameters, data processing, statistical tests, software versions

- Never: "under standard conditions", "using routine methods", "data were analyzed statistically"



### Data Availability Statement

Generate using Nature data policy principles:

- Map each dataset to access route: public repository, controlled access, within supplement, reused source, third-party restricted

- Prefer DOI/accession numbers over personal websites

- Pattern: "The [data type] generated in this study have been deposited in [repository] under accession code [XXX]. Source data are provided with this paper."

- Flag "available upon request" as weak unless legally/ethically required



## Failure Mode Diagnosis



Before editing any section, diagnose the main problem in priority order:



1. **Paper type** — wrong narrative logic for this paper type?

2. **Section job** — section not fulfilling its rhetorical responsibility?

3. **Paragraph logic** — claim without evidence? evidence without claim? missing boundary?

4. **Sentence polish** — clutter, passive voice, overclaim?



Fix from top down. Do not polish sentences while reasoning is broken.



## Nature-Specific Style Rules



### Sentence control

- Each sentence ≤ 30 words

- One core subject-verb proposition per sentence

- Split overloaded sentences rather than polishing cosmetically

- Active voice preferred: "We show..." not "It is shown that..."



### Paragraph control

- Short paragraphs OK (3–5 sentences typical for Nature)

- Each paragraph: one controlling idea + support

- Thematic linking, not repetitive "This suggests..." openings



### De-AI Polish Rules



Remove or replace these AI-typical words:

- "delve" → "examine", "investigate"

- "pivotal" → "important", "central"

- "landscape" → "field", "area"

- "multifaceted" → "complex"

- "underscores" → "shows", "highlights"

- "leveraging" → "using"

- "novel" (overused) → "new", or remove if claim is clear from context

- "groundbreaking" → remove or use specific evidence

- "paradigm shift" → describe the actual change

- "in conclusion" → just state the conclusion directly



### Hedging (Academic Phrasebank patterns)

- "These results suggest that..."

- "A possible explanation is that..."

- "This discrepancy may reflect..."

- "To our knowledge, this is the first..."

- "Further work is needed to determine whether..."



### Transitions

- Contrast: "However,", "By contrast,", "Nevertheless,"

- Addition: "Moreover,", "Furthermore,", "In addition,"

- Cause: "Consequently,", "As a result,", "Therefore,"

- Concession: "Although...,", "Despite...,", "Notwithstanding,"



### Limitations acknowledgment

- "These results should be interpreted with caution because..."

- "A limitation of this study is that..."

- "The generalizability of these findings is limited by..."



## ⛔⛔⛔ Output Contract (highest priority)



**Single artifact**: `paper/main.md` (UTF-8, ≥ 5KB)



**Never produce**: `.tex` / `.bib` / `.cls` / `.aux` / any LaTeX command.



**Mandatory verification**:

```bash

PASS=true

[ -f paper/main.md ] && SZ=$(wc -c < paper/main.md) || SZ=0

[ "$SZ" -ge 5120 ] && echo "✅ paper/main.md ($SZ)" || { echo "❌ paper/main.md missing"; PASS=false; }



if grep -qE '\\(begin|end|input|cite|ref|label|includegraphics|section|chapter)\{' paper/main.md; then

    echo "❌ LaTeX residue:"

    grep -nE '\\(begin|end|input|cite|ref|label|includegraphics|section|chapter)\{' paper/main.md | head -5

    PASS=false

fi



ls paper/*.tex paper/sections/*.tex 2>/dev/null | head -1 | grep -q . && { echo "❌ .tex files detected"; PASS=false; } || true



[ "$PASS" != true ] && { echo "⛔ verification FAILED"; exit 1; }

```



## docx-cn-engine markdown conventions



(Same as paper-write-docx — see that SKILL or the brief recap below.)



- `# Title` (unique), `## Section`, `### Subsection`

- `## Abstract` triggers centered abstract style

- `## References` triggers hanging-indent for `[N] ...` lines

- Math: `$inline$`, `$$display$$`. **公式编号一律写进公式内部 `\tag{n}`（如 `$$ ... \tag{1}$$`）；⛔ 禁止在闭合分隔线行尾加编号（`$$ (1)` 这种）**——编号带尾巴时引擎认不出闭合行，会吞掉后续正文与图片（事故记录见 paper-write-docx:209-218）。

- Figures: `![Figure 1: caption](figures/fig.png)`

- Tables: markdown pipe tables (rendered as 3-line academic style)

- Citations: `[1]`, `[1, 2]`, `[1-3]` — never `\cite{}`



## Workflow



### Step 0: Upstream check + resume



```bash

for f in PAPER_PLAN.md RESULTS.md; do

    [ -f "$f" ] && echo "✅ $f" || echo "  $f not found"

done

[ -f figures/all_results.json ] && echo "✅ figures/all_results.json" || true

ls figures/*.png figures/*.pdf 2>/dev/null | head -10



[ -f paper/main.md ] && cp paper/main.md "paper/main-backup-$(date +%s).md.bak"

```



### Step 1: Identify paper type and target venue



Read PAPER_PLAN.md and TARGET_VENUE. Choose:

- **Nature / Science**: 3000-3500 words main, ~4 figures, single-column logic

- **Nature Methods / Communications**: 4000-5000 words, ~5-6 figures

- **Cell**: 7000-8000 words, more figures, longer methods



### Step 1.5: Figure inventory



Before drafting, build inventory of available figures:



```bash

echo "=== Available figures ==="

ls -la figures/*.png figures/*.pdf 2>/dev/null

echo ""

echo "=== Available tables ==="

ls -la figures/TABLE_*.md 2>/dev/null

echo ""

echo "=== latex_includes.tex (caption reference only) ==="

cat figures/latex_includes.tex 2>/dev/null

```



Build mapping: figure ID → file → target section. Only embed figures whose files exist. Each figure needs ≥ 5 lines analysis after it before next visual.



⛔ Nature standard: Fig. 1 (overview/hero), Figs. 2-4 (main findings).



### Step 2: Pre-fetch verified reference pool



⛔ Build verified pool BEFORE writing any citations.



```bash

PYTHON=""; for _c in "$MH_PYTHON" python python3; do [ -z "$_c" ] && continue; if $_c -c "import sys" >/dev/null 2>&1; then PYTHON="$_c"; break; fi; done; [ -z "$PYTHON" ] && PYTHON=python

mkdir -p _tmp



# Use descriptive citation keys: LastName_Year_topic_keywords

# Examples:

#   - vaswani_2017_attention_transformer

#   - lecun_2015_deep_learning_review

#   - TODO__crispr_cas9_off_target  (author/year unclear)



# Search by topic:

#   $PYTHON "$SCHOLAR_SCRIPT" bibtex "transformer attention mechanism" --max 5

```



Save verified entries to `_tmp/_verified_refs.txt`. Use ONLY verified entries while drafting.



### Step 2.5: BibTeX verification (after body draft is done)



```bash

PYTHON=""; for _c in "$MH_PYTHON" python python3; do [ -z "$_c" ] && continue; if $_c -c "import sys" >/dev/null 2>&1; then PYTHON="$_c"; break; fi; done; [ -z "$PYTHON" ] && PYTHON=python

# List descriptive citation keys to _tmp/_topics.txt

while IFS= read -r key; do

    query=$(echo "$key" | sed 's/^TODO__//; s/_/ /g')

    echo "--- Fetching: $key (query: $query) ---"

    $PYTHON "$SCHOLAR_SCRIPT" bibtex "$query" --max 3

    sleep 0.5

done < _tmp/_topics.txt

```



For each result:

1. **`match_label`**: `"good"` → use; `"partial"` → verify; `"low"` → retry or use WebSearch.

2. **`match_score`**: < 0.3 → don't blindly trust.

3. Format as `[N] Author A, Author B. Title. Journal Year, vol(issue): pages.` under `## References`.

4. References ordered by first-appearance in body.



**Fallback**: WebSearch on Google Scholar / PubMed / Semantic Scholar to verify title + authors + year manually.



⛔ References: Nature ≥ 30; Nature Methods/Communications ≥ 50; Cell ≥ 70. **⛔ Authenticity over count**: these are targets for *真实检索到的* references — NEVER fabricate entries to hit the number. If real retrieval falls short, keep the real ones and note the gap; do not invent citations.



### Step 3: Draft Results first



Anchor in evidence. Each subsection:

- One claim

- 2-3 numerical pieces of evidence

- One figure or one table reference

- Boundary statement



**⛔ Cross-section context + figure-data binding (prevents the "two-layers" disconnect):**

- **After finishing each section**, append a 3-5 line card to `_writing_context.md` in the workspace root (core claim / key numbers / newly defined symbols & terms / figures discussed); re-read it before the next section so later sections carry forward prior conclusions, reuse defined terms (don't redefine), and keep every metric's number consistent — see `<chapter_context_card>` in `shared-scripts/writing_rules.md`.

- **Before writing the analysis for any figure**, follow `<figure_data_binding>`: identify *what quantity the figure plots* from `latex_includes.tex`/figure inventory → locate its real values in `RESULTS.md`/`figures/all_results.json` → use only those real numbers. **Never guess numbers from the plot's shape/position, never fabricate coordinates.**



Required figures (Nature standard): Fig. 1 (overview), Fig. 2-4 (main findings). Each figure embedded with full caption.



### Step 4: Draft Introduction



After Results, write Intro with hourglass:

- Para 1: broad relevance (why phenomenon matters)

- Para 2: narrow to specific gap (what's missing in literature)

- Para 3: state hypothesis/question

- Para 4: preview contribution



### Step 5: Title + Discussion



- Title: ≤ 75 characters, contribution-driven (not "A Study of...")

- Discussion: widen back. Connect findings to broader literature. State boundaries.



### Step 6: Methods (reproducibility)



- Materials, conditions, equipment

- Detailed protocol

- Statistical analysis

- Code/data availability statement



### Step 7: Abstract last



150-200 words. Single paragraph. Cover: context → gap → method → finding → implication.



### Step 8: Final structure



```markdown

# [Title]



[Authors and affiliations]



## Abstract



[150-200 words single paragraph]



## Introduction



[Hourglass]



## Results



### [Result 1 subheading]



![Figure 1: ...](figures/fig1.png)



[Claim-Evidence-Boundary paragraph + ≥5 lines analysis]



### [Result 2 subheading]



...



## Discussion



[Widen + connect + bound]



## Methods



### Data and materials



### Analysis



### Statistics



### Data availability



### Code availability



## References



[1] ...

[2] ...

```



### Step 9: Cross-review



```bash

mkdir -p _tmp

cat << 'EOF' > _tmp/_review_prompt.txt

Nature-style paper review. Focus on:

1. Hourglass structure

2. Claim-evidence-boundary in each paragraph

3. Title/abstract clarity

4. Reader-first ordering

5. Score (1-10) + top-3 improvements

## Paper:

EOF

cat paper/main.md >> _tmp/_review_prompt.txt

PYTHON=""; for _c in "$MH_PYTHON" python python3; do [ -z "$_c" ] && continue; if $_c -c "import sys" >/dev/null 2>&1; then PYTHON="$_c"; break; fi; done; [ -z "$PYTHON" ] && PYTHON=python

$PYTHON "$REVIEWER_SCRIPT" --prompt-file _tmp/_review_prompt.txt --thread-file _tmp/_reviewer_thread.json 2>&1 | tee _tmp/_cross_review.txt

```



### Step 9.5: Self-review checklist



Run failure mode diagnosis on each section. Check (every box must be ✅):



- [ ] **Hourglass structure intact** (Intro broad → narrow → gap; Discussion narrow → broad → implication)

- [ ] **Each paragraph has claim-evidence-boundary** (no claim-only or evidence-only paragraphs)

- [ ] **No AI-typical language remaining** (delve / pivotal / landscape / multifaceted / underscores / leveraging / novel / groundbreaking / paradigm shift / "in conclusion")

- [ ] **Active voice dominant** ("We show..." not "It is shown that...")

- [ ] **Sentences ≤ 30 words** (split overloaded sentences)

- [ ] **Title ≤ 75 characters** (Nature guideline)

- [ ] **Abstract 150–200 words**

- [ ] **No fabricated references** (every citation came from scholar_fetch.py / WebSearch verification)

- [ ] **Hedging appropriate** (no overclaim — use "These results suggest...", "A possible explanation is...", "To our knowledge, this is the first...")

- [ ] **Data Availability statement complete** (DOI/accession for each dataset; "available upon request" only when legally/ethically required)

- [ ] **Author Contributions template present** ("X.Y. designed the study, performed analysis. Z.W. collected data. All authors discussed results and edited the manuscript.")

- [ ] **Boundary language present in Discussion** ("These findings hold under...", "We do not claim X generalizes to Y", "The generalizability is limited by...")



If any box is ❌, fix before proceeding.



### Step 10: Final verification



Re-run the Output Contract block. All ✅ before ending.



## Writing Discipline (apply throughout drafting)



**⛔ Style rules:**

- No bullet/enumerated lists for narrative prose. Use "(1) ... (2) ..." inline numbering or transitional phrases ("First, ...; second, ..."). Bullets OK for input checklists, evaluation metrics, software dependencies.

- Each paragraph 3-5 sentences (for Nature, sometimes 3-4 OK; never 1-2 sentence paragraphs).

- Consecutive paragraphs cannot start with the same syntactic pattern.

- Figure numbers MUST be cited explicitly (every paragraph names its "Fig. N"), but VARY the pattern. What's forbidden is the monotonous empty "Fig. X shows… as can be seen…" skeleton repeated for every figure — NOT figure-as-subject per se. Rotate: parenthetical (preferred, "(Fig. X)"), sentence-start, verb-led, figure-as-subject (allowed when it carries a real finding), post-hoc confirmation. Adjacent figures must not use the same pattern.

- Each figure/table needs ≥ 5 lines of analysis (numerical interpretation + comparison + reasoning) before the next visual.



**⛔ Numbers from data only.**



**⛔ NEVER `cat figures/*_results.json`.** These result files often contain full-precision time-series arrays (tens of MB / hundreds of thousands of lines); reading them whole blows up the context — local models fail outright, and GPT-via-transit chokes on protocol translation of the oversized payload and stalls on repeated `api_retry`. **The paper text only uses scalar values; the giant arrays are for figures, not prose.** Before writing any results section, run the `summarize` script below for a KB-level overview (scalars shown verbatim — zero precision loss — only big arrays compressed to "length + range + first 3 samples"):

```bash

PYTHON=""; for _c in "$MH_PYTHON" python python3; do [ -z "$_c" ] && continue; if $_c -c "import sys" >/dev/null 2>&1; then PYTHON="$_c"; break; fi; done; [ -z "$PYTHON" ] && PYTHON=python

[ -f RESULTS.md ] && cat RESULTS.md

$PYTHON - <<'PY'

import json, os, glob

def summarize(v, depth=0):

    if isinstance(v, list):

        n=len(v); nums=[x for x in v if isinstance(x,(int,float))]

        if nums: return f'list[{n}] range=[{min(nums):.4g},{max(nums):.4g}] sample={v[:3]}'

        if v and isinstance(v[0], (list,dict)): return f'list[{n}] of {type(v[0]).__name__}, first_shape={len(v[0]) if hasattr(v[0],"__len__") else "?"}'

        return f'list[{n}] sample={str(v[:3])[:80]}'

    if isinstance(v, dict) and depth<2:

        return 'dict{'+', '.join(f'{k}: {summarize(x,depth+1)}' for k,x in list(v.items())[:6])+'}'

    return f'{type(v).__name__}={str(v)[:60]}'

for f in sorted(glob.glob('figures/*_results.json')):

    sz=os.path.getsize(f); d=json.load(open(f,encoding='utf-8'))

    print(f'\n=== {os.path.basename(f)} ({sz//1024}KB) ===')

    if isinstance(d, dict):

        for k,v in d.items(): print(f'  {k}: {summarize(v)}')

    else: print(f'  {summarize(d)}')

PY

```

Every scalar you need is in `RESULTS.md` or the range/sample above. If one scalar isn't fully shown, fetch just that value with `$PYTHON -c "import json;d=json.load(open('figures/all_results.json'));print(d['key'])"` — still never read the whole file. Copy exact numbers. No memory-based estimation.



## Expansion strategies (substantive, not padding)



- Formula without derivation → add step-by-step derivation with physical/biological meaning

- Result with only "Figure X shows" → add 2-3 paragraphs (numerical interpretation + comparison + reasoning + boundary)

- Methods only listed → add why this method, what alternatives were considered, why rejected

- Algorithm as pseudocode only → add explanation, complexity, convergence, sensitivity to hyperparameters



## Key Rules (docx mode)



- Single artifact: `paper/main.md`

- No LaTeX (no `\begin`, `\input`, `\cite`, `\section`, `\includegraphics`)

- Math: `$...$` / `$$...$$`

- Figures: `![alt](path)`

- Tables: markdown pipe tables

- Citations: `[N]`

- Hourglass + claim-evidence-boundary architecture

- Backup before overwrite





---



## ⛔ Figure embedding verification (MUST pass before finishing — file existence + actual reference in paper/main.md both required)



```bash

echo "=== Figure embedding check (docx mode: file + ![]() / image reference in paper/main.md) ==="

missing=0



# Markdown docx mode: figure files referenced via ![](figures/xxx.png) or relative path

for img in figures/*.png figures/*.pdf figures/*.jpg figures/*.svg; do

    [ -f "$img" ] || continue

    bn=$(basename "$img")

    # Skip placeholder files

    [ "$bn" = "latex_includes.tex" ] && continue

    if [ -f paper/main.md ]; then

        if ! grep -q "$bn" paper/main.md; then

            echo "MISSING: $bn — produced but not embedded in paper/main.md"

            missing=$((missing + 1))

        fi

    fi

done



# Also check TABLE_*.md files are embedded

for tbl in figures/TABLE_*.md; do

    [ -f "$tbl" ] || continue

    bn=$(basename "$tbl")

    if [ -f paper/main.md ]; then

        if ! grep -q "$bn" paper/main.md; then

            echo "MISSING: $bn — table file produced but not referenced in paper/main.md"

            missing=$((missing + 1))

        fi

    fi

done



echo "Total missing embeddings: $missing"

[ "$missing" -gt 0 ] && echo "⛔ DO NOT finish until missing = 0. Embed each missing figure/table into paper/main.md via ![caption](figures/xxx.png) or cat figures/TABLE_xxx.md."

```



---



## ⛔ FIGURE_MANIFEST audit (run before finishing — must produce + embed every planned figure)



```bash

echo "=== FIGURE_MANIFEST audit ==="

PLAN_FILE=""

for f in PROBLEM_ANALYSIS.md PAPER_PLAN.md MODELING_REPORT.md TOPIC_PLAN.md; do

  [ -f "$f" ] && grep -q "<!-- BEGIN FIGURE_MANIFEST -->" "$f" && { PLAN_FILE="$f"; break; }

done

if [ -n "$PLAN_FILE" ]; then

    START=$(grep -n "<!-- BEGIN FIGURE_MANIFEST -->" "$PLAN_FILE" | head -1 | cut -d: -f1)

    END=$(grep -n "<!-- END FIGURE_MANIFEST -->" "$PLAN_FILE" | head -1 | cut -d: -f1)

    EXPECTED_FIGS=$(sed -n "${START},${END}p" "$PLAN_FILE" | grep -oE "^[[:space:]]*-[[:space:]]+(fig_[a-zA-Z0-9_]+|tikz_[a-zA-Z0-9_]+)" | sed "s/^[[:space:]]*-[[:space:]]*//")

    manifest_missing=0

    for name in $EXPECTED_FIGS; do

        if ! ls figures/${name}.png figures/${name}.pdf figures/${name}.drawio 2>/dev/null | head -1 | grep -q .; then

            echo "❌ MANIFEST: $name file missing"

            manifest_missing=$((manifest_missing + 1))

        elif ! grep -qE "${name}\.(png|pdf)" paper/main.md 2>/dev/null; then

            echo "❌ MANIFEST: $name exists but not embedded"

            manifest_missing=$((manifest_missing + 1))

        fi

    done

    if [ "$manifest_missing" -gt 0 ]; then

        echo "⛔ FIGURE_MANIFEST audit failed ($manifest_missing missing)"

    else

        echo "✅ FIGURE_MANIFEST fully embedded"

    fi

fi

```



## ⛔ Universal paper-stage audit (shared across all writing steps)



Before finishing writing / compiling, run the universal audit. Works without `PROBLEM_FACTS.json`:



```bash

# Universal paper audit:

#   [13] Conclusion consistency: paper text ↔ results.json (prevent "optimal=X but paper says Y")

#   [14] Event source attribution (prevent "guessing source from variable name")

# Falls back to simplified mode if no PROBLEM_FACTS.json (general academic / course / humanities).

PYTHON=""; for _c in "$MH_PYTHON" python python3; do [ -z "$_c" ] && continue; if $_c -c "import sys" >/dev/null 2>&1; then PYTHON="$_c"; break; fi; done; [ -z "$PYTHON" ] && PYTHON=python

if [ -f _utils/facts_audit.py ]; then

    $PYTHON _utils/facts_audit.py --stage paper 2>&1 | tee -a AUDIT_REPORT.md

    PRC=$?

    if [ "$PRC" = "1" ]; then

        echo "❌ Universal paper-stage audit failed — fix paper text / results.json before finishing"

    fi

fi

```