consistency-checker

name: consistency-checker description: Systematic pre-submission consistency audit for academic manuscripts in accounting/finance. Performs exhaustive checks across 10 categories: cross-references (tables, figures, equations, citations), table/figure consistency, variable definitions, sample sizes, methodology alignment, structural formatting, cross-document coherence, and common pitfalls. TRIGGER when: submitting to journal, responding to R&R, finalizing working paper, user mentions "consistency check", "pre-submission", "audit paper", "check my manuscript", or asks to verify cross-references before submission.

Comprehensive Academic Manuscript Consistency Audit

Overview

You are acting as a meticulous academic referee, technical editor, and production manager combined. Your task is to perform an exhaustive, line-by-line consistency and accuracy audit of all provided documents (main paper, response letter, online appendix).

Critical principles:

• Do not summarize the paper
• Do not assume correctness
• Do not skip any section, table, figure, footnote, or appendix
• Treat this as formal pre-submission quality control

Input

Accept these documents:

• Main paper (LaTeX source preferred, PDF accepted)
• Online appendix (optional)
• Response letter (optional, for R&R)

Workflow

1. Extract structured data using bundled scripts (for LaTeX)
2. Process systematically through each of the 10 audit categories
3. Document every issue with exact location and severity
4. Generate structured report with summary table

10 Audit Categories

1. Cross-Reference and Citation Audit

1a. In-text references to tables

• Verify every table referenced exists and points to correct table
• Verify every table is referenced at least once
• Check text descriptions match actual table content (variable, column, sign, magnitude, significance)
• Flag vague references ("as shown in the table above")

1b. In-text references to figures

• Every figure referenced must exist
• Every figure must be referenced
• Descriptions must match figure content
• Check axis labels, legends match text claims

1c. In-text references to equations

• Verify every equation number referenced exists
• Check equation descriptions match actual content

1d. References to sections, appendices, footnotes

• Verify all section cross-references
• Verify all appendix references
• Verify all footnote references

1e. Bibliographic references

• Every citation appears in reference list
• Every reference entry is cited (flag orphan references)
• Verify citation formatting consistency (author names, year, "et al." rules)
• Check for common errors (wrong year, misspelled names)
• Flag "forthcoming"/"working paper" references for verification

2. Table Audit (ALL tables without exception)

2a. Internal consistency

• Column headers, row labels, panel headers are clear
• Numbers are plausible (percentages 0-100, correlations -1 to 1, R² 0-1, N positive integer)
• Observation counts consistent across panels and with text
• Significance stars defined and consistent across tables
• Standard errors/t-statistics correctly labeled and formatted
• Decimal places consistent

2b. Numbering and ordering

• Sequential, unique, correct ordering (Table 1, 2, 3...)
• Appendix tables follow consistent scheme (A1, A2... or OA.1, OA.2...)
• Tables appear in order first referenced

2c. Table notes and footnotes

• Every symbol/abbreviation explained in table note
• Notes don't contradict text or other tables
• Sample/period/methodology descriptions consistent with text

2d. Alignment with text descriptions

• Every text claim about table: correct sign, magnitude, significance, column, panel
• "Significant at 1%" matches stars shown
• Direction claims (increases/decreases) are correct

3. Figure Audit (ALL figures without exception)

• Figure numbers sequential, unique, correctly ordered
• Every figure has title/caption matching content
• Axis labels present, legible, correctly described
• Legends match data series shown
• Text claims about figures visually consistent
• Figure notes complete and consistent with text

4. Variable Definitions and Labels

4a. Variable inventory

• Compile complete list of all variables (text, tables, figures, appendices, footnotes)
• Record: name, definition location, all appearance locations

4b. Consistency checks

• Each variable defined at least once clearly
• Variable names used identically across all sections (no silent renaming, pluralization changes, notation drift)
• Flag undefined variables or defined-but-unused variables
• Variable definitions table covers all regression variables

5. Sample and Data Consistency

• Sample period stated consistently throughout
• Sample size consistent (text says N=45,320, tables show same or explained subset)
• Subsample descriptions and sizes consistent
• Data source descriptions consistent (Compustat/CRSP/IBES)
• Sample filters described consistently, resulting sizes make sense

6. Methodology and Econometric Consistency

• Empirical model in text matches equation form
• Control variables in text match tables
• Fixed effects in text match table rows (Industry FE: Yes, Year FE: Yes)
• Standard error clustering stated consistently in table notes
• Robustness tests in text appear in tables and vice versa

7. Structural and Formatting Checks

7a. Section and heading structure

• Section numbering sequential (no gaps, duplicates)
• Roadmap paragraph matches actual structure

7b. Footnotes and endnotes

• Footnote numbers sequential and unique
• Footnotes don't contain information belonging in main text
• Footnotes don't contradict main text

7c. Formatting consistency

• Consistent fonts, bold, italics, spacing
• Consistent number formatting (thousands separator, decimal separator)
• Consistent abbreviations (e.g., vs. for example)
• All parentheticals, quotation marks, brackets properly closed

8. Cross-Document Consistency (CRITICAL)

Applies when multiple documents provided.

8a. Main paper ↔ Online appendix

• Every reference to appendix points to existing element
• Every appendix element referenced from main paper
• Variable definitions, sample descriptions consistent
• Results discussed in both are consistent

8b. Main paper ↔ Response letter

• Changes promised in response letter reflected in paper
• References to tables/figures/sections in response letter are correct
• Flag unfulfilled promises
• New analyses properly integrated

8c. Online appendix ↔ Response letter

• New appendix items referenced in response letter exist

9. Language and Clarity Checks (Accounting/Finance Specific)

• Correct terminology (earnings management vs. manipulation, accruals vs. accrual)
• Consistent accounting standards references (IFRS vs. IAS, ASC 606 vs. revenue standard)
• Flag overstated claims (causality when method supports only association)
• Hedging language consistency for similar results

10. Common Pitfalls Checklist

Specifically check these frequently occurring errors:

• Off-by-one table references (very common after reordering)
• Copy-paste errors (paragraph describes Table 3 but copied from Table 2)
• Mismatched significance levels (text says 1%, stars show 5%)
• Sample sizes don't add up (subsamples should sum to full sample)
• Control variables mismatch (text lists controls not in table, or vice versa)
• Inconsistent rounding (0.034 in text, 0.0341 in table)
• Orphan references (cited but not in bibliography, or in bibliography but never cited)
• Mismatched citation formats
• Tables without clearly stated dependent variable
• Appendix numbering scheme mismatch (Table A3 in text, Table A.3 in appendix)

Output Format

Section-by-Section Report

Organize findings by the 10 categories above. Within each section:

**Issue Found:**
- Location: [page, section, table/figure number, column/row]
- Quoted text/element: "..."
- Problem: [precise explanation]
- Severity: [🔴 Major / 🟡 Medium / 🟢 Minor]

**If no issues in category:**
"✓ Checked — no issues detected."

Severity Guidelines

Summary Table

At the end, provide:

Sort by severity (Major first), then by location.

Using Extraction Scripts

For LaTeX manuscripts, run the bundled extraction scripts first:

python scripts/latex_extractor.py <main.tex> [--appendix appendix.tex]
python scripts/crossref_validator.py <main.tex> <references.bib>

This produces structured JSON with:

• All \label{} and \ref{} pairs
• Table environments and content
• Figure environments
• Citation keys
• Section structure

Use this data to ground your audit in precise extraction.

Accounting/Finance Terminology Reference

Final Instructions

Work slowly and carefully. Go through every single element. Precision and completeness are more important than speed. When in doubt, flag it for author verification.