brycewang-stanford/proofread

Proofread Manuscript

Run the proofreading protocol on the "Sewage in Our Waters" manuscript. Produces a report — does NOT edit source files.

Input: $ARGUMENTS — a .tex filename, section number (e.g. 01), or all.

---

Project-Specific Context

File Locations

• Manuscript sections: docs/overleaf/*.tex
• Main document: docs/overleaf/_main.tex
• Bibliography: docs/overleaf/refs.bib
• Generated tables: output/tables/*.tex (tabularray format)
• Generated figures: output/figures/
• Quarto book: book/*.qmd (for cross-referencing analysis)
• Analysis scripts: scripts/R/09_analysis/ (for methodology verification)

Key Checks Specific to This Project

• Spill count/hours metrics must match the 12/24-hour counting methodology
• Radius distances (250m-10km) must be consistent across sections
• LSOA vs MSOA fixed effects must be correctly stated
• Upstream/downstream directionality must match river network logic
• Dry spill definition must be consistent with rainfall threshold used

---

Workflow

Step 1: Identify Files

• If $ARGUMENTS is a specific .tex file: review that file
• If $ARGUMENTS is a section number (e.g. 01): review docs/overleaf/0X_*.tex
• If $ARGUMENTS is all: review _main.tex and all section files in docs/overleaf/
• If $ARGUMENTS is a .qmd file: review as advisory (non-blocking)

Step 2: Run 6-Category Review

Category 1: Structure

• Contribution clearly stated within first 2 pages of introduction
• Standard economics paper sequence (intro → background → data → method → results → conclusion)
• Smooth transitions between sections
• Road map in introduction matches actual section ordering
• Appendix sections properly referenced from main text

Category 2: Claims-Evidence Alignment

• Every stated effect size matches a number in output/tables/
• Percentage impacts correctly computed from log coefficients
• Sample sizes and time periods match data pipeline output
• Radius distances in text match those in tables
• "Significant" claims match actual p-values / confidence intervals

Category 3: Identification Fidelity

• Hedonic specification matches what scripts/R/09_analysis/02_hedonic/ actually estimates
• Repeat sales approach correctly described per Palmquist (1982)
• Long difference specification matches grid-level scripts
• DiD/event study timing and treatment definitions consistent
• Instrument (hydraulic capacity) described consistently with 04_hydraulics_instrument.tex
• Dry spill identification matches rainfall threshold in data pipeline

Category 4: Writing Quality

• No banned hedging phrases ("interestingly", "it is worth noting", "arguably", "it is important to note")
• Notation consistent: LSOA, MSOA, EDM used correctly throughout
• Variable names in text match variable names in specifications
• No AI writing patterns (see humanizer skill for full checklist)
• Tone matches existing author voice

Category 5: Grammar & Polish

• Subject-verb agreement
• Article usage (particular attention to UK vs US English conventions)
• Tense consistency (present for methodology, past for results)
• No orphaned text, repeated words, or copy-paste artifacts
• Acronyms defined on first use

Category 6: Compilation & LaTeX

• All \input{} files exist
• All \textcite{} / \parencite{} keys exist in refs.bib
• All \ref{} targets have matching \label{}
• Table/figure floats properly placed
• No overfull hbox warnings (check tabularray table widths)
• KOMA-Script class options used correctly

Step 3: Scoring

Apply deductions on a 0-100 scale:

| Issue | Deduction | |-------|-----------| | Effect size doesn't match table output | -25 | | Identification strategy misrepresented | -20 | | Broken citations (\textcite key missing) | -15 | | Broken cross-references (\ref undefined) | -15 | | Radius/sample inconsistency across sections | -10 | | Overfull hbox > 10pt | -10 per | | Hedging language | -5 per (max -15) | | Notation inconsistency | -5 | | Overfull hbox 1-10pt | -1 per |

Step 4: Format-Aware Severity

| Context | Scoring | |---------|---------| | Paper manuscript (.tex) | Blocking — issues must be fixed | | Quarto book (.qmd) | Advisory — reported but non-blocking |

Step 5: Present Report

Save report to output/log/proofread_report_[SECTION].md and present summary:

## Proofread Report: [filename]
**Score:** XX / 100
**Date:** YYYY-MM-DD

### Issues by Category
| Category | Critical | Major | Minor |
|----------|----------|-------|-------|
| Structure | ... | ... | ... |
| Claims-Evidence | ... | ... | ... |
| Identification | ... | ... | ... |
| Writing Quality | ... | ... | ... |
| Grammar | ... | ... | ... |
| LaTeX | ... | ... | ... |

### Top 3 Critical Issues
1. ...
2. ...
3. ...

### Escalation Flags
- Claims don't match output → verify against analysis scripts
- Strategy misrepresented → review identification approach
- Framing issues → flag to authors

---

Principles

• Proofreader is a CRITIC, not a creator. Never write or revise — only report.
• Be precise. Quote exact text, cite exact line numbers and file paths.
• Cross-reference against actual output. Always verify numbers against output/tables/.
• Proportional severity. A missing comma is Minor. Numbers that don't match regression output is Critical.
• Format-aware. Paper .tex files are blocking; book .qmd files are advisory.