Deep Review Criteria

Use this file when running deep-review.

What to check

1. Formula / derivation errors
2. Notation inconsistency
3. Prose vs formal object mismatch
4. Numerical inconsistency
5. Insufficient justification
6. Claim accuracy / overclaim
7. Misleading ambiguity
8. Missing information / reproducibility gap
9. Internal contradiction
10. Self-consistency of standards
11. Table structure violations — missing three-line format (booktabs), vertical lines present, inconsistent number precision within columns, caption placed below instead of above
12. Abstract structural incompleteness — missing elements in the five-element model (Background / Objective / Methods / Results / Conclusion), data-free results section, hollow conclusion that merely restates results
13. Theory contribution deficiency — core concepts undefined or ambiguous, no substantive dialogue with existing theories, theoretical increment not identifiable (see A5-A7 in domain_reviewer_agent.md, reference playbook)
14. Qualitative methodology opacity — sampling rationale missing, coding process undescribed, no saturation discussion, absent reflexivity in sensitive research (see B6-B10 in methodology_reviewer_agent.md, reference playbook)
15. Pseudo-innovation / straw man — fabricated research gap, mischaracterized prior work, selective citation that hides overlap with existing methods (see prior_art_reviewer_agent.md)
16. Paragraph-level argument incoherence — logical jumps between adjacent paragraphs, causal inversions, evidence that does not support the stated claim (see C5 in critical_reviewer_agent.md, reference playbook)

Editor-in-Chief screening (gate mode)

For gate mode, an EIC screening pass runs before the detailed checklist. Read agents/editor_in_chief_agent.md for the full protocol. The EIC evaluates:

• Pitch quality: does the paper hook the reader within the first three paragraphs?
• Venue fit: is the paper appropriate for the target journal/conference?
• Fatal flaws: any issue that guarantees rejection regardless of technical merit?
• Presentation baseline: does the manuscript meet minimum professional standards?

A desk-reject verdict from EIC screening is a gate blocker.

Reasoning style

For each finding:

• state what initially raised concern
• explain what context you checked
• say what remains unresolved
• keep the strongest evidence
• prefer one well-developed finding over several shallow duplicates

What not to do

• do not report copy-editing trivia
• do not report obvious OCR glitches as author errors unless the issue survives the most charitable correction
• do not criticize a section for omitting content that clearly appears later

Reviewer calibration

A strong paper may still have several major issues if they each threaten different conclusions. Do not over-merge distinct arguments.