How It Works Agents FAQ Upload Paper →
20 Specialist AI Agents

Peer review your paper before the reviewers do

Upload an empirical finance paper and get a full methodological audit in minutes — not months. 20 AI agents check statistics, robustness, data integrity, and more. Then chat directly with each agent about any finding.

paperreferee.org/paper.php?id=42&tab=review
Complete
18
Remaining
2
Total Agents
20
Recommendation
Major Revision
Agent Status Findings
P-Hacking Detector Complete 3 View →
Table Consistency Checker Complete 7 View →
Endogeneity Auditor Running
Robustness Evaluator Pending
20
Specialist AI agents
6
Review categories
~10min
Average full review
1
Synthesized report

Three steps to a stronger paper

No setup required. Upload your PDF and let the agents do what human reviewers take weeks to accomplish.

📄
01

Upload your PDF

Drag and drop your empirical finance paper. We extract the text, tables, and figures automatically — no special formatting required.

🤖
02

Run the agents

Choose which specialist agents to deploy — from p-hacking detection to table consistency, endogeneity audits to robustness checks. Run one, a few, or all twenty.

📊
03

Get your report

Synthesize all findings into a single report with a consensus recommendation, revision roadmap, severity rankings, and a draft response letter guide.

Twenty specialists. One thorough review.

Each agent is a focused expert trained on a specific class of methodological concern in empirical finance research.

Statistical Integrity

P-Hacking Detector

Scans for suspicious clustering around significance thresholds and specification searching patterns.

Statistical Integrity

Table Consistency Checker

Cross-references reported numbers across tables, text, and appendices for internal contradictions.

Causal Identification

Endogeneity Auditor

Evaluates instrument validity, selection bias, simultaneity, and omitted variable concerns.

Causal Identification

Diff-in-Diff Validator

Checks parallel trends assumptions, treatment timing, staggered adoption, and TWFE issues.

Robustness

Robustness Evaluator

Assesses whether reported robustness tests are genuine or cosmetic, and flags missing ones.

Robustness

Sample & Data Auditor

Reviews sample construction, survivorship bias, data vintage issues, and look-ahead bias.

Econometrics

Standard Error Reviewer

Verifies clustering choices, heteroskedasticity handling, and appropriate inference methods.

Econometrics

Model Specification Critic

Examines functional form assumptions, variable transformations, and control variable choices.

Presentation

Writing & Clarity Reviewer

Evaluates exposition quality, logical flow, and whether claims are properly hedged.

Presentation

Literature Gap Finder

Identifies missing citations, mischaracterized prior work, and uncredited methodological contributions.

+ 10 More

And counting…

External validity, economic magnitude, replication feasibility, data ethics, and more specialist agents.

Everything synthesized into one actionable report

No more parsing twenty separate reviews. Get a single document with prioritized findings and a clear revision path.

🎯

Consensus Recommendation

Accept, Minor Revision, Major Revision, or Reject — with a rationale drawn from all agents.

🗺️

Revision Roadmap

Findings ranked by severity (Fatal → Major → Minor → Suggestion) with concrete remedies for each.

💪

Strengths Highlights

Not just critique — the report identifies what works well so you can lead with strengths in your response letter.

✉️

Response Letter Guide

A structured template for responding to each concern, saving hours of drafting time.

Consensus Recommendation
Major Revision
Strengths & Highlights
Consensus Summary
Revision Roadmap
Response Letter Guide

Don't just read findings — discuss them

Every finding has a dedicated chat. The agent remembers its full review, your paper, and the specific issue — so you can push back, ask for clarification, or explore fixes.

🧠

Full context, every time

The agent has your paper, its own review, and the specific finding loaded into every response. No re-explaining needed.

🤺

Push back on findings

Disagree with a critique? Explain your reasoning and the agent will honestly reassess — or hold firm with evidence.

🔧

Get concrete fixes

Ask for specific tests, alternative specifications, or exact code snippets to address each finding.

💾

Persistent conversations

Come back later — every chat is saved. Pick up right where you left off, finding by finding.

Major High confidence Table 3, Col 4

Standard errors are clustered at the firm level, but the treatment varies at the industry-year level. This understates standard errors and inflates t-statistics for the main coefficients.

Discuss
Standard Error Reviewer
Econometrics · Major
Finding: Standard errors are clustered at the firm level, but the treatment varies at the industry-year level…
We cluster at firm because our sample has repeated firm-year observations. Wouldn't industry-year clustering lose power?
You're right that firm clustering handles within-firm serial correlation. But the issue here is that your treatment — the regulation shock — is assigned at industry × year. When treatment varies at a coarser level than clustering, standard errors are biased downward (see Abadie et al., 2023). You could two-way cluster on firm and industry-year to address both concerns.
What would two-way clustering look like in Stata?
With reghdfe, it's straightforward: reghdfe y treatment controls, absorb(firm year) cluster(firm ind_year). If the t-stats on your main coefficient drop below conventional thresholds after this change, that confirms the original clustering was insufficient.
Ask about this finding…

Common questions

Paper Referee is optimized for empirical finance research — papers with regression tables, statistical tests, and quantitative methodology. This includes asset pricing, corporate finance, banking, and financial economics papers. It can also provide useful feedback on empirical papers in adjacent fields like economics and accounting.

Individual agents typically complete in 30–90 seconds. Running all 20 agents sequentially takes around 10–15 minutes. The synthesis report generates in about 60 seconds after agents finish. You can also run agents selectively — just the ones most relevant to your paper.

Your paper is only used to generate the review. It is not shared, indexed, or used to train models. Uploaded files are stored securely and can be deleted at any time from your dashboard.

No — and it is not intended to. Paper Referee is a pre-submission diagnostic tool. Think of it as a thorough first pass that catches the issues reviewers will notice, so you can fix them before submission. Human judgment, domain expertise, and editorial assessment remain irreplaceable.

Findings are ranked on a seven-level scale: Fatal-A (paper-invalidating), Fatal-B, Fatal-C, Major (likely R&R concern), Moderate, Minor, and Suggestion. This helps you triage which issues to address first and allocate revision effort efficiently.

Every finding has a "Discuss" button that opens a dedicated chat with the agent that raised it. The agent has full context — your paper, its complete review, and the specific finding — so you can ask follow-up questions, push back on a critique, or request concrete code and specification changes. Conversations are persisted, so you can return to them at any time.

Library B is the studio behind Paper Referee and other research tools. We build focused, well-crafted software for academics and researchers — each product is a standalone site under the Library B family.

Fix the issues before Reviewer 2 finds them

Upload your paper and get a full AI-powered peer review. No account required.