paper-polisher
Polish mathematical modeling paper drafts for grammar, clarity, formula consistency, hedging calibration, overclaim detection, and contest formatting compliance. Use after paper-section-writer has drafted sections.
zhnnky329 · AI review on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add zhnnky329/MathModeling-skills --skill paper-polisherAbout this skill
name: paper-polisher description: Polish mathematical modeling paper drafts for grammar, clarity, formula consistency, hedging calibration, overclaim detection, and contest formatting compliance. Use after paper-section-writer has drafted sections. license: MIT
Purpose
Polish mathematical modeling contest paper sections for language quality, logical clarity, formula consistency, and claim calibration.
This skill operates on already-drafted paper sections. It improves wording, fixes grammar, checks formulas, calibrates hedging to match evidence strength, detects overclaims, and ensures formatting compliance. It does not invent new content, add unsupported claims, or rewrite the paper's scientific argument.
Adapted from nature-polishing design principles: language serves the argument, polish should not hide weak reasoning, and claims must be proportional to evidence.
This skill does not write new paper sections, run experiments, generate figures, or perform final QA.
When to use
Use this skill:
- After
paper-section-writerhas drafted one or more paper sections. - Before
quality-assurance-auditor. - When the user says: "polish the paper", "check the English", "fix the grammar", "improve the writing", "calibrate the claims", "check for overclaims", "proofread Q1 section".
- When Chinese-to-English translation has produced rough drafts that need smoothing.
- When formulas, notation, or terminology are inconsistent across sections.
Preconditions
The following should already exist or be provided:
- Paper section drafts under
paper/sections/. - Final method explanations (for formula and notation verification).
- Final result analyses (for claim verification).
- The global symbol table at
planning/symbol_table.md(if available). - Contest formatting requirements (if available).
If paper sections do not exist, hand back to paper-section-writer.
Inputs
Use or request:
paper/sections/*.mdorpaper/sections/*.tex— the drafted sections.methods/Qx/qx_final_method_explanation.md— for formula and notation verification.results/Qx/reports/qx_final_result_analysis.md— for claim verification.planning/symbol_table.md— for notation consistency.- Contest formatting requirements.
Workflow
-
Identify the paper type and section.
- Mathematical modeling contest papers follow a standard structure: Abstract → Problem Restatement → Problem Analysis → Assumptions → Symbols → Model Construction (per Q) → Model Solution → Results Analysis → Robustness → Strengths & Limitations → Conclusion.
- Each section has different polishing priorities (see section-specific rules below).
-
Run the 12-point polish checklist (see below).
-
Calibrate claims against evidence.
- For each numerical or comparative claim, verify it is supported by the final result analysis or robustness report.
- If a claim overstates the evidence, downgrade the language.
- If a claim is unsupported, flag it as a blocker (do not silently remove — the writer needs to decide).
-
Check formula and notation consistency.
- Every symbol must appear in the global symbol table or be defined locally.
- Same concept must use the same symbol across all sections.
- Subscripts, superscripts, and indices must be consistent.
- Formula numbering must be sequential and match references in text.
-
Check terminology consistency.
- Same concept must use the same term throughout.
- Method names must match the final method explanation.
- "Baseline", "main model", "improved model" must be used consistently.
-
Produce polished sections.
- Show a diff or change summary.
- Mark any claims that were downgraded and why.
- Flag any remaining issues that need author attention.
12-Point Polish Checklist
1. Sentence Length
- Split sentences longer than 30 words.
- Vary sentence length: mix short (8-15 words) and medium (15-25 words).
- The first and last sentences of each paragraph should be the clearest.
2. Paragraph Structure
- Each paragraph should have one main point.
- Topic sentence first, support following, transition at end (or beginning of next).
- Paragraphs longer than 5-6 sentences should be split or tightened.
3. Tense Consistency
- Problem restatement / Assumptions / Symbols: Present tense.
- Model construction: Present tense for model description.
- Model solution / Results analysis: Past tense for what was done and found.
- Conclusion: Present tense for final findings, past tense for what was done.
- Do not mix tenses within a single paragraph without reason.
4. Hedging Calibration
Match claim strength to evidence:
| Evidence Level | Appropriate Hedging | Example |
|---|---|---|
| Robust, multiple checks | Strong claim, no hedge | "The entropy-TOPSIS method produces stable rankings." |
| Single check, moderate perturbation | Moderate hedge | "The rankings appear stable under moderate weight changes." |
| Limited check, narrow range | Weak hedge | "The results suggest that rankings may be stable within the tested range." |
| No check, extrapolation | No claim allowed | Flag as unsupported. Do not write. |
Hedging phrases (strongest to weakest):
demonstrates/shows/establishes→ strongestindicates/suggests/supports→ moderatemay indicate/appears to/is consistent with→ weakcould potentially/might possibly→ weakest (use sparingly)
5. Overclaim Detection
Flag and downgrade or remove:
- Absolute claims: "always", "never", "proves", "guarantees", "optimal" (unless proven).
- Unwarranted causation: "A causes B" when only correlation is shown.
- Scope expansion: "All models benefit from..." when only one model was tested.
- Unverified "first" or "novel" claims.
- "Significantly" without statistical test or defined threshold.
- "Our model outperforms all existing methods" when only 1-2 bas