The COLT 2026 review form, explained

How COLT reviewers read your paper

COLT asks reviewers to judge whether the work is close to being ready for publication and to help guide it there with constructive feedback. Papers are evaluated primarily on the main text, though all proofs must be included; there is an author response and discussion phase. Theory papers are judged on the significance and novelty of their results and techniques and on technical correctness, not on empirical evaluation.

• Constructive: aim to help the paper reach publication quality.
• Compare results precisely to the best known bounds and cite the prior work.
• Say exactly which proofs were checked; point to specific steps for any error.
• Do not ask for experiments unless the paper's claims are empirical.

Where the bar sits. COLT 2026 accepted about 190 papers; submission counts are not published (historically ~30% acceptance). On this modeled 1–5 scale a mean of ≥ 3.5 is a rough proxy for acceptance.

The reviewer form, field by field

In the order reviewers see it, with every score and its official wording.

1. 01

   Summary Of Results

   Free text

   State the main theorems informally: setting, assumptions, the bounds or characterizations proved, and how they compare to the best known results.

2. 02

   Significance And Novelty

   List of points

   What is new relative to prior bounds and techniques (cite them), whether the techniques are of independent interest, and how the results advance learning theory.

3. 03

   Technical Correctness

   List of points

   Which proofs were checked and how carefully; any gaps, errors or unjustified steps (with locations); whether assumptions are reasonable and stated. Judgment rests primarily on the main text, but proofs must be complete in the appendix.

4. 04

   Clarity And Presentation

   Free text

   Are the setting, results and proof ideas explained clearly in the main text?

5. 05

   Comments And Questions For The Authors

   List of points

   Constructive suggestions and questions for the author response, aimed at bringing the paper closer to publication.

6. 06

   Overall Score

   Score · 1–5

   Overall recommendation.

   1. 5Strong accept
   2. 4Accept
   3. 3Borderline
   4. 2Reject
   5. 1Strong reject
7. 07

   Confidence

   Score · 1–5

   How certain you are, including how carefully the proofs were checked.

   Score	Meaning
   5	Expert checked the proofs carefully and know the related literature well
   4	Confident checked the key proofs
   3	Fairly confident read the proofs at a high level
   2	Somewhat confident did not check the proofs
   1	Educated guess

Area chair meta-review

COLT area chairs lead discussion and recommend; the form is not public.

• Accept
• Reject

COLT submission requirements

Reviewers check these. Gaps are easy points to lose.

1. At most 12 PMLR-format pages, plus unlimited references and appendices
2. All proofs included (main text should convey the key ideas)
3. Anonymized (double-blind)

Policy on LLMs in reviewing