submission-readiness-audit

Submission Readiness Audit

The last pass before submission is not another writing pass. It is a risk audit: what would make reviewers distrust the paper, misunderstand the contribution, or find a preventable compliance issue?

Use this when

• The user has a draft or near-final paper.
• The user asks what is missing before submission.
• The paper needs a checklist against venue or artifact requirements.
• The user wants to catch overclaims, inconsistent numbers, missing references, or appendix drift.
• The user is preparing camera-ready revisions.

Do not use this when

• The paper argument is still unclear. Use paper-argument-planner first.
• The user needs new related work prose. Use related-work-writer.
• The user needs artifact packaging rather than paper audit. Use artifact-release-packager.

Workflow

1. Identify submission context

Record:

• venue or format if known
• page limit
• anonymity requirements
• required sections or checklists
• artifact/reproducibility expectations
• current draft components available

If the venue is unknown, run a general research-paper audit and mark venue-specific items as unknown.

For ML/NLP/CV/AI submissions, add an OpenReview-era reviewer-risk pass: check for the objections reviewers are explicitly asked to surface at venues such as ICLR, NeurIPS, CVPR, AAAI, and ACL/ARR. In particular, audit unsupported central claims, vague novelty, unnamed missing baselines, weak reproducibility, missing limitations/ethics discussion, data/code/model attribution gaps, hallucinated or misattributed citations, and questions the authors would struggle to answer during rebuttal.

2. Audit claims against evidence

For abstract, introduction, results, discussion, and conclusion:

• list major claims
• identify supporting figure/table/proof/citation
• mark unsupported or overstated claims
• check that conclusion does not exceed results
• check that limitations do not contradict contributions

Unsupported claims should be narrowed, supported, or removed.

3. Audit experimental and numerical consistency

Check:

• numbers match between text, tables, captions, and appendix
• metrics are defined consistently
• higher/lower-is-better is clear
• baselines are named consistently
• datasets and splits are described consistently
• statistical uncertainty is reported where it matters
• ablations correspond to claimed mechanisms

4. Audit paper mechanics

Check:

• every figure/table is referenced in order
• captions are self-contained
• appendix references resolve
• citations support the sentences they are attached to
• notation is introduced before use
• acronyms are defined
• method names are consistent
• limitations, ethics, and broader impact are present if expected
• anonymization is preserved for double-blind submissions

5. Classify blockers

Use severity levels:

• Blocker — likely to cause rejection, desk rejection, or serious reviewer distrust.
• Major — weakens the paper but may be fixable quickly.
• Minor — polish or clarity issue.
• Venue-specific unknown — cannot verify without venue requirements.

Do not bury blockers in a long checklist.

Output format

# Submission Readiness Audit

## Verdict
Ready / Not ready / Conditionally ready

## Blockers

## Major issues

## Minor issues

## Claim-evidence consistency
| Claim | Location | Evidence | Status | Fix |
|---|---|---|---|---|

## Figures, tables, and appendix checks

## Venue/checklist items

## Final action list

Quality bar

A good audit should reduce preventable reviewer objections. It should be specific enough that the user can fix issues directly, not just feel vaguely worried.