Edmans Audit Skill for Claude

Edmans Audit: Pre-Submission Paper Audit

Applies Edmans' (2025) rejection-reasons framework to the current paper.

Produces a dated report at Output/edmans_audit_YYYY-MM-DD.md.

Read references/edmans_framework.md before starting — it contains the

full rubric with sub-checks and severity guidance.

Phase 1: Read the paper

• Read Paper - Github/2025-Strategic-Defaulting/main.tex to identify

which \input{} lines are active (not commented out). This is the

canonical list of sections currently in the paper.

• Read every active .tex file in include order:
• 11_Introduction.tex — research questions + contribution claims
• 12_Institutional.tex — institutional background
• 12_DescriptiveStat.tex — descriptive statistics
• 13_Results_*.tex — all active results sections
• 14_Conclusions.tex — claimed contributions
• Skim 15_Appendix.tex for supporting-evidence inventory only (do not

audit appendix content unless user requests it).

• Read Code/REGISTRY.md to cross-reference paper claims against the

analysis scripts that actually ran.

• Build a mental map: which RQ does each section address, what is the

stated finding, what identification strategy is used, what tables and

figures support it.

Do not skip any active section. The audit must cover the full paper

as currently compiled.

Phase 2: Apply the Edmans framework

Read references/edmans_framework.md now. Apply each check below.

2a — Contribution

For each research question, assess:

• Novelty: Would a reader of the existing literature meaningfully

update their beliefs? Is the finding inferrable from prior work?

• Importance: Are economic stakes quantified (dollar figures, welfare)?
• Describing vs explaining: Does the RQ make a causal claim, or is

it purely descriptive? Descriptive is fine if acknowledged — but the

paper should not overclaim causality.

• Incremental vs fundamental: Is this "X in a new country" or a

genuinely new mechanism?

2b — Execution

Run per-RQ checks. For each, assess:

• Sample: size, representativeness, selection bias, time coverage
• Identification: endogeneity concerns, instrument validity (if any),

diff-in-diff parallel trends, omitted variables

• Alternative explanations: which plausible stories are ruled out,

which are not

• Mechanism: is the channel demonstrated or only the reduced form?
• Robustness: alternative variable definitions, specifications,

controls, subsamples

• Measurement: variable quality, proxy noise

Cross-cutting checks:

• Are claims in the text supported by the regression tables shown?
• Are economic magnitudes interpreted (not just statistical significance)?
• Is the "efficient reallocation" counter-narrative addressed?

2c — Exposition

Assess:

• Introduction real estate: time wasted on well-known motivation?
• Roadmap clarity: does the reader know the paper's structure?
• Caveat acknowledgment: limitations upfront or buried?
• Length efficiency: any sections that could be cut?
• Figure/table payoff: does each exhibit earn its page space?
• Conclusion: does it overstate or appropriately hedge?

Phase 3: Generate the report

Use this exact template:

# Edmans Audit — [Paper Title]
**Date:** YYYY-MM-DD
**Scope:** [Core paper / Full paper including appendix]
**Paper length:** [N pages, as reported or estimated]

## Executive Summary
[3-5 bullet points: the highest-impact findings across all categories.
Lead with the single most important issue.]

## 1. Contribution
### Strengths
- [Specific strength with file reference, e.g. "11_Introduction.tex:L15"]

### Issues
- [CRITICAL/IMPORTANT/MINOR] [Issue] — [Recommendation]

## 2. Execution

### RQ1: [Question as stated in paper]
**Section:** [file.tex] | **Finding:** [one-line summary of result]
- [Severity] [Issue] — [Recommendation]

### RQ2: [Question as stated in paper]
**Section:** [file.tex] | **Finding:** [one-line summary of result]
- [Severity] [Issue] — [Recommendation]

### RQ3: [Question as stated in paper]
**Section:** [file.tex] | **Finding:** [one-line summary of result]
- [Severity] [Issue] — [Recommendation]

### RQ4: [Question as stated in paper]
**Section:** [file.tex] | **Finding:** [one-line summary of result]
- [Severity] [Issue] — [Recommendation]

### Cross-cutting
- [Severity] [Issue] — [Recommendation]

## 3. Exposition
### Strengths
- [Specific strength]

### Issues
- [CRITICAL/IMPORTANT/MINOR] [Issue] — [Recommendation]

## Top 3 Priorities
1. [Most impactful action — the single change that most improves
   the paper's chance of acceptance]
2. [Second priority]
3. [Third priority]

Phase 4: Save and display

• Save report to Output/edmans_audit_YYYY-MM-DD.md.
• Display the Executive Summary and Top 3 Priorities to the user.
• Offer to dive into any section.

Rules

• Read the LaTeX every time. Never rely on memory or prior runs.
• Be specific, not generic. Every finding must reference a file,

section label, or table. "Consider robustness" is useless.

• Acknowledge strengths before critiques. A good paper has real

strengths — call them out.

• Calibrate severity honestly.
• CRITICAL = likely desk rejection or fatal R&R issue
• IMPORTANT = referee will raise; fixable but non-trivial
• MINOR = polish; won't block acceptance
• Do not rewrite the paper. Audit and recommend only.
• Default scope: core RQs only (skip UCC appendix material).

Expand to full paper only if user requests it.

• Cross-reference claims against actual analysis. Use the script

registry and session logs to verify that what the paper says was

done matches what was actually done.

• No more than 15 issues total. Force-rank. If everything is

CRITICAL, the labels are miscalibrated.