| name | iclr-submission |
| description | Use when auditing an ICLR main-conference submission for OpenReview readiness, double-blind anonymity, page limits, reciprocal reviewing, LLM-use disclosure, dual-submission policy, supplementary material, and desk-reject risk. |
ICLR Submission
Use this for an ICLR main-conference submission audit. Reopen the current CFP, Author Guide,
OpenReview submission form, template, Code of Ethics, Code of Conduct, and LLM policy before giving
deadline-ready advice.
Submission audit
- Confirm the target is the ICLR main conference, not a workshop, Tiny Paper, blog, affinity event,
or later publication track.
- Check that all authors have complete OpenReview profiles and that the author list follows the
current freeze/change windows.
- Verify that the abstract is substantive and non-duplicative before the abstract deadline; do not
rely on a placeholder.
- Confirm use of the current ICLR LaTeX template and current page-limit rules. ICLR 2027 caps the
main text at 9 pages at submission (10 during discussion and for camera-ready), strictly enforced —
over-length main text is desk-rejected. References are unlimited and outside the cap; appendices
after the bibliography are unlimited, and reviewers need not read them.
- Inspect anonymity in the PDF, appendices, supplementary ZIP, code, data, demos, links, repository
metadata, comments, acknowledgments, and file names.
- Confirm reciprocal-reviewing and quota compliance for the author team, both of which desk-reject
in ICLR 2027 and both of which are decided at the abstract deadline, not later:
- every author on 3+ submissions must review at least 6 papers, and weak or missing reviews by the
rebuttal stage can desk-reject their papers;
- every submission needs at least one author registered to review 3 papers, qualified by an
accepted primary-conference paper at a listed venue as of the abstract deadline — workshop,
position, tiny, demo and industry papers do not qualify, and neither does an acceptance that
lands after that date;
- no author on more than 20 papers, and at most one paper per author where nobody qualifies as a
reciprocal reviewer. Over-quota submissions are desk-rejected at random until each author is
under quota, so this is not a risk that can be absorbed by submitting extra.
- Check every author's OpenReview profile early: eligibility is read off the profile, wrong
information is grounds for desk rejection, and a profile created without an institutional email
can sit in moderation for up to two weeks.
- Write the mandatory AI use statement — ICLR 2027 requires it as a section in the paper (it does
not count toward the page limit) as well as on the submission form. Disclosure is required for
synthetic data, theoretical models, mathematical claims and proof assistance, hypotheses,
methodology or experiment design, implementing methods, translation, dataset cleaning, qualitative
analysis and interpreting results, and recommended for figures, code, drafting, literature search,
brainstorming, titles and reference formatting. Hidden prompt injection, hallucinated citations, or
undisclosed LLM-written scientific content can become ethics issues.
Desk-reject and integrity risks
- Late or missing abstract/full paper.
- New author added after the author-list deadline.
- Template manipulation, over-limit main text, or missing required fields.
- Non-anonymous public links, repository ownership, analytics, demo login, or organization slug.
- Dual submission to another archival venue that violates current ICLR policy.
- Unclear LLM disclosure or review-manipulation text.
- Reciprocal-reviewing noncompliance.
Desk-reject triage table
ICLR enforcement is strict and the consequences are public: even a withdrawn submission leaves a
visible OpenReview trace. Triage the highest-probability rejections first.
| Risk | Quick test | Fix before deadline |
|---|
| De-anonymization | Grep PDF and ZIP for names, remotes, buckets | Strip metadata, neutralize links |
| Over-limit main text | Count pages against the current rule | Move material after references |
| Author-list change | Check the freeze window in the CFP | Finalize authors before the cutoff |
| Dual submission | Confirm no archival overlap | Withdraw or rescope the other venue |
| LLM disclosure | Confirm the form field is truthful | Disclose; remove hallucinated citations |
Worked vignette
A team submits an LLM fine-tuning paper whose anonymized repo link points to a GitHub org naming the
lab, with a caption thanking a named cluster. Both leak identity. The fix: replace the link with an
anonymized supplement ZIP, remove the acknowledgment until camera-ready, re-grep the PDF for the
cluster name, and confirm all authors completed OpenReview profiles before the freeze.
Reviewer-pushback patterns
- "Looks like a workshop paper." Reframe for the main track or route elsewhere.
- "Possible dual submission." Document why the other version is non-archival under current policy.
Output format
[ICLR readiness] Ready / Needs fixes / Not ready
[OpenReview state] profiles / abstract / paper / supplement / reviewer nomination
[Blocking risks] <anonymity/page-limit/dual-submission/LLM/reciprocal-reviewing>
[Fix order] <highest-impact fixes before deadline>
[Current-cycle pages to reopen] <CFP, Author Guide, OpenReview group, template, LLM policy>