Check Writing Quality
Scan the full paper text for writing issues: grammar errors, typos, unclear phrasing, inconsistent terminology, overclaim wording, and logical flow gaps.
Workflow
- [ ] Step 1: Read full paper text
- [ ] Step 2: Scan for writing issues
- [ ] Step 3: Save report
Step 1: Read Full Paper Text
Read {paper_dir}/metadata/{arxiv_id}/mathpix/{arxiv_id}.md
Read the entire document — do not split by section.
Step 2: Scan for Writing Issues
Scan every section for:
Grammar
- Subject-verb disagreement ("this method can effectively balances")
- Wrong tense ("which we illustrated" when describing a figure → "illustrate")
- Missing articles ("available in community" → "in the community")
- Singular/plural mismatch ("three device" → "three devices")
Typos & Formatting
- Misspelled words ("aling" → "allowing", "chocies" → "choices")
- Missing spaces ("inputsor" → "inputs or")
- Double brackets, broken formatting in tables/figures
- Wrong reference numbers
Unclear Phrasing
- Vague language ("comparable results" without threshold)
- Awkward phrasing ("speeds up the performance")
- Incomplete sentences, missing objects ("surpasses in X" — surpasses what?)
Inconsistent Terminology
- Same concept called different names in different sections
- Inconsistent capitalization of terms
Overclaim Wording
- "significantly improves" without statistical test
- "universally effective" tested on only a few benchmarks
- "inherent property" from a single observation
Logical Flow
- Conclusions not supported by the presented evidence
- Missing transitions between sections
- Claims in abstract/conclusion not backed by experiments
Each issue:
{
"page": 3,
"quote": "this method can effectively balances component number",
"status": "warning",
"reason": "Subject-verb disagreement: 'can effectively balances' should be 'can effectively balance'.",
"suggest": "Change 'balances' to 'balance'."
}
status:
warning — grammar error, typo, unclear phrasing, minor inconsistency, or anything that MIGHT be a PDF parsing artifacterror — only for issues you are 100% certain exist in the actual paper, not caused by PDF parsing. Examples: duplicated words ("that that"), wrong citation (citing paper A for paper B's result), clearly wrong terminology used consistently. If there is ANY chance the issue comes from PDF extraction, use warning instead
Step 3: Save Report
Output: {paper_dir}/reports/check_txt.json
{
"summary": {
"total_issues": 25,
"warning_issues": 22,
"error_issues": 3
},
"results": [
{
"page": 3,
"quote": "this method can effectively balances component number",
"status": "warning",
"reason": "Subject-verb disagreement: 'can effectively balances' should be 'can effectively balance'.",
"suggest": "Change 'balances' to 'balance'."
}
]
}
Tips
- Scan every section including abstract, appendix, table/figure captions, and references
- Non-native English writing often has systematic errors (e.g., missing articles throughout) — report the first 2-3 instances, then note "pattern continues throughout"
- Distinguish style preferences from actual errors — only flag clear mistakes or genuinely confusing phrasing
- Overclaim wording is
error only if it misrepresents the actual results; stylistic exaggeration is warning
Ignore PDF Parsing Artifacts
The input text is extracted from PDF via Mathpix/MinerU. Do NOT flag these as writing issues:
- Broken hyphens:
"ex-\nploratory" or "trajec-\ntory" — these are line-break artifacts, not typos
- Missing hyphens from line breaks:
"turnlevel" instead of "turn-level" — if the compound word clearly needs a hyphen but the text was joined across a line break, skip it
- LaTeX math notation:
$14.9 \%$, $\beta$, \mathbb{E} — these are correct LaTeX, not formatting errors
- Image references:
 — skip these
- Footnote markers:
[^0], ${ }^{1}$ — PDF extraction artifacts
- Garbled capitalization:
"AgentFLOW", "AgENTFlow" — PDF extraction often mangles mixed-case words. Skip these entirely unless the same inconsistency appears multiple times in clearly different contexts
- Grammar around LaTeX:
"achieving average accuracy by $14.9 \%$" — the awkwardness may come from LaTeX rendering, not the actual paper. If the sentence structure is otherwise sound, skip it
Only flag issues that exist in the actual paper, not artifacts from the PDF-to-markdown conversion. When in doubt, downgrade to warning or skip entirely.
1---2name: read-txt3description: Check writing quality — grammar, typos, unclear phrasing, inconsistent terminology, logical flow issues. Scans the full paper text and reports issues by page.4license: MIT5---67# Check Writing Quality89Scan the full paper text for writing issues: grammar errors, typos, unclear phrasing, inconsistent terminology, overclaim wording, and logical flow gaps.1011## Workflow1213```text14- [ ] Step 1: Read full paper text15- [ ] Step 2: Scan for writing issues16- [ ] Step 3: Save report17```1819---2021## Step 1: Read Full Paper Text2223```24Read {paper_dir}/metadata/{arxiv_id}/mathpix/{arxiv_id}.md25```2627**Read the entire document — do not split by section.**2829---3031## Step 2: Scan for Writing Issues3233Scan every section for:3435### Grammar36- Subject-verb disagreement ("this method can effectively balances")37- Wrong tense ("which we illustrated" when describing a figure → "illustrate")38- Missing articles ("available in community" → "in the community")39- Singular/plural mismatch ("three device" → "three devices")4041### Typos & Formatting42- Misspelled words ("aling" → "allowing", "chocies" → "choices")43- Missing spaces ("inputsor" → "inputs or")44- Double brackets, broken formatting in tables/figures45- Wrong reference numbers4647### Unclear Phrasing48- Vague language ("comparable results" without threshold)49- Awkward phrasing ("speeds up the performance")50- Incomplete sentences, missing objects ("surpasses in X" — surpasses what?)5152### Inconsistent Terminology53- Same concept called different names in different sections54- Inconsistent capitalization of terms5556### Overclaim Wording57- "significantly improves" without statistical test58- "universally effective" tested on only a few benchmarks59- "inherent property" from a single observation6061### Logical Flow62- Conclusions not supported by the presented evidence63- Missing transitions between sections64- Claims in abstract/conclusion not backed by experiments6566Each issue:6768```json69{70 "page": 3,71 "quote": "this method can effectively balances component number",72 "status": "warning",73 "reason": "Subject-verb disagreement: 'can effectively balances' should be 'can effectively balance'.",74 "suggest": "Change 'balances' to 'balance'."75}76```7778**`status`**:79- `warning` — grammar error, typo, unclear phrasing, minor inconsistency, or anything that MIGHT be a PDF parsing artifact80- `error` — **only for issues you are 100% certain exist in the actual paper**, not caused by PDF parsing. Examples: duplicated words ("that that"), wrong citation (citing paper A for paper B's result), clearly wrong terminology used consistently. If there is ANY chance the issue comes from PDF extraction, use `warning` instead8182---8384## Step 3: Save Report8586Output: `{paper_dir}/reports/check_txt.json`8788```json89{90 "summary": {91 "total_issues": 25,92 "warning_issues": 22,93 "error_issues": 394 },95 "results": [96 {97 "page": 3,98 "quote": "this method can effectively balances component number",99 "status": "warning",100 "reason": "Subject-verb disagreement: 'can effectively balances' should be 'can effectively balance'.",101 "suggest": "Change 'balances' to 'balance'."102 }103 ]104}105```106107## Tips108109- Scan **every section** including abstract, appendix, table/figure captions, and references110- Non-native English writing often has systematic errors (e.g., missing articles throughout) — report the first 2-3 instances, then note "pattern continues throughout"111- Distinguish style preferences from actual errors — only flag clear mistakes or genuinely confusing phrasing112- Overclaim wording is `error` only if it misrepresents the actual results; stylistic exaggeration is `warning`113114## Ignore PDF Parsing Artifacts115116The input text is extracted from PDF via Mathpix/MinerU. **Do NOT flag these as writing issues:**117118- **Broken hyphens**: `"ex-\nploratory"` or `"trajec-\ntory"` — these are line-break artifacts, not typos119- **Missing hyphens from line breaks**: `"turnlevel"` instead of `"turn-level"` — if the compound word clearly needs a hyphen but the text was joined across a line break, skip it120- **LaTeX math notation**: `$14.9 \%$`, `$\beta$`, `\mathbb{E}` — these are correct LaTeX, not formatting errors121- **Image references**: `` — skip these122- **Footnote markers**: `[^0]`, `${ }^{1}$` — PDF extraction artifacts123- **Garbled capitalization**: `"AgentFLOW"`, `"AgENTFlow"` — PDF extraction often mangles mixed-case words. Skip these entirely unless the same inconsistency appears multiple times in clearly different contexts124- **Grammar around LaTeX**: `"achieving average accuracy by $14.9 \%$"` — the awkwardness may come from LaTeX rendering, not the actual paper. If the sentence structure is otherwise sound, skip it125126**Only flag issues that exist in the actual paper**, not artifacts from the PDF-to-markdown conversion. When in doubt, **downgrade to warning or skip entirely**.
QinghongLin