ICWM2027

ICWM 2027

Review Process

ICWM uses an AI-first, human-accountable review process on OpenReview. AI assistance is used to allocate review effort efficiently; responsibility for evaluations and decisions remains with human reviewers and Area Chairs.

Why this structure

Large conferences spend substantial human effort applying a nearly uniform review budget to every paper. ICLR 2026 reported 76,139 reviews for 13,763 papers receiving decisions—about 5.5 reviews per decided paper—provided by 18,054 reviewers.

ICWM separates structured initial analysis from full human peer review. Phase 1 is intended to identify clear cases and missing evidence quickly, while mandatory human approval preserves accountability. Phase 2 then concentrates human attention on the papers that continue.

  • More time per review: smaller reviewer portfolios allow closer reading and more substantive feedback.
  • Faster turnaround: clear Phase 1 cases are resolved early and the full review pool is smaller.
  • More scrutiny where needed: disputed or uncertain papers can receive additional human review.
  • Continuous correction: public reviews, responses, and comments allow errors to be identified during the process.

Illustrative capacity model

1,000 submissions

700–800papers advance if Phase 1 removes 20–30%
2,100–2,400Phase 2 assignments at 3 reviews per paper
2.1–2.4assignments per participating reviewer, on average
3.0nominal worst-case average if every paper advances

Assumes 1,000 distinct eligible reviewing authors, three Phase 2 reviews per advancing paper, and no additional assignments. Actual loads will vary with conflicts, expertise, reviewer availability, duplicate authorship across submissions, and additional review requested by Area Chairs.

Scale comparison: ICLR 2026 review-process retrospective ↗

Open and single-blind

Process

00

Automated compliance screening

Submissions are checked for mechanical requirements, including page limits, minimum font size, required sections, file validity, and PDF readability. There is no scientific evaluation in this stage.

01

December 23, 2026 – January 8, 2027

AI-assisted review with human approval

Eligible papers receive structured AI-assisted assessments. Assigned human reviewers inspect, revise as needed, and explicitly approve the official assessments. AI output alone cannot determine a paper's outcome.

Clear negative assessment: the Area Chair must either confirm a Phase 1 rejection or rescue the paper into Phase 2 for full human review.

Mixed or positive assessment: the paper proceeds to Phase 2. Area Chairs may override the initial routing when the assessment does not adequately reflect the paper.

02

January 9 – February 8, 2027

Full human peer review

Papers continuing beyond Phase 1 receive full evaluation by human reviewers. Review effort is allocated adaptively: papers with conflicting or uncertain assessments may receive additional reviewers, and Area Chairs may request further review whenever needed.

03

February 9–16, 2027

Public rebuttal and discussion

Authors respond to official reviews in public. Reviewers may revise or clarify their assessments, Area Chairs may participate in the discussion, and identifiable OpenReview users may add clearly labeled public comments.

04

February 17 – March 2, 2027

Deliberation and final decision

Reviewers and Area Chairs consider the paper, reviews, rebuttal, discussion, and relevant public comments. Final decisions are posted publicly on March 2, 2027.

Reviewer participation

At least one author of each submission must serve as a reviewer. The reviewing author participates under the same anonymity and conflict-of-interest rules as other official reviewers.

This requirement may be refined before submissions open.