NeurIPS 2026 Submission: The 14-Day Pre-Deadline Checklist

1. Day 14 — Lock the Title and Claims

Two weeks out, freeze your title and the claim sentence of your abstract. Every

change after this point costs you time that should go into figures and bugs.

The title should contain the single most specific novelty — not the dataset name,

not the framework name. If your title could apply to any of the last five papers

on the topic, rewrite it tonight. Check what your advisor actually wants

(some want benchmark names in the title, some don't) and lock the decision.

2. Day 13-12 — Reviewer Pool Hygiene

NeurIPS matches reviewers by your chosen keywords and by the authors you cite.

Open OpenReview and check your past coauthor list — anyone who's ever been on

a paper with you in the last three years cannot review, so auto-blocking them

is fine, but overly aggressive blocking can flag your submission. Select three

primary keywords that actually describe your contribution (not your area).

Then check your references: do you cite at least two papers from potential

reviewers who work on exactly your sub-problem? If not, add them — weak citation

coverage hurts you at the bidding stage.

3. Day 11-10 — Figures and Tables Freeze

Freeze every figure that appears in the paper. After this point, caption

rewrites are fine but no regenerating figure images from new runs. Every

figure should answer exactly one question a reviewer would ask. If a figure

needs two paragraphs of caption to be understood, it's the wrong figure.

Run every number in every table through a sanity check: does the improvement

exceed the standard deviation? Are you reporting the mean or median? Is your

baseline the right one (not the one easiest to beat)?

4. Day 9-8 — The Reproducibility Section

NeurIPS reviewers specifically check the reproducibility checklist. Fill it out

honestly — "No" answers are fine if you explain why. The worst move is claiming

compute hours you didn't measure or hyperparameter sweeps you didn't run.

Reviewers catch this, and it's treated as a credibility red flag.

Have your code zip ready. NeurIPS allows anonymous code in supplementary.

Strip author names, git config, comments with your name, and any absolute paths

that reveal your cluster. Zip it with a single README that says "python main.py

reproduces Table 2". Reviewers will not run your code, but they will open the

README — make it confidence-inspiring in 30 seconds.

5. Day 7-5 — Full Reread + One External Reader

Print the paper (yes, physically). Read it straight through in one sitting

with a pen. Mark every sentence you stumble on. On screen, your eye skips

awkward transitions; on paper, it doesn't. Fix the top 20 stumbles.

Then hand the PDF to one person outside your immediate group — ideally

a senior PhD student or postdoc from an adjacent subfield. Give them

two days. The feedback you want is: "I couldn't tell what's new here

until section 4." If they say that, the abstract and introduction fail

their job. Rewrite them before day 3.

6. Day 4-3 — The Rebuttal Prep You're Avoiding

Write down, in a separate file, the three weakest things about your paper.

Be honest. For each, draft a 200-word response you would give if a reviewer

pointed it out. You won't send this now, but on rebuttal day you'll have

a head start instead of panicking.

Common weak points to pre-answer: limited dataset coverage, single-seed

evaluation, missing baseline, heuristic hyperparameter choices, ablations

only on small variants. If you can't defend these now, you can't defend

them in the rebuttal period either — and spotting the gap 10 days before

submission means you might still have time to run the missing experiment.

7. Day 2 — Formatting and Supplementary Material

Run the NeurIPS style file check. Overlength papers are desk-rejected

without review. The page limit is strict to the line, including references.

If you're over by one line, the fix is almost always a tighter related

work section — not squeezing figure captions.

Supplementary material has its own rules. Only include what strengthens

the submission. Proofs, extended experiments, dataset details, and

reproducibility code — yes. "Additional thoughts on future work" — no.

Reviewers read supplementary selectively; padding it dilutes signal.

8. Day 1 — Upload Early, Sleep

OpenReview goes down under load in the final hour. Every cycle. Aim to upload

12 hours before the deadline, not 12 minutes. The abstract deadline and full

paper deadline are separate submissions — miss the abstract, and the full paper

upload form won't appear.

Once uploaded, close OpenReview. Don't open the PDF again to find typos.

Typos are fine; the paper is fine. The actual work is done. The next thing

you do should be sleep — reviewer assignments go out within 72 hours and

you want to be fresh for whatever happens next.