Iclr workflow

ICLR Workflow

Use this as the project-management skill for an ICLR submission. Always replace dates with the current cycle's official deadlines and work backwards from OpenReview cutoffs.

ICLR is a conference, not a journal: it has no standing editor-in-chief and no article-processing charge. The rotating leadership is the per-edition General Chair, Senior Program Chair, and Program Chairs (2026 General Chair: Carl Vondrick; Senior PC: Bharath Hariharan; PCs: Raffel, Pinto, Yang, Faust, verified 2026-06-22), and the cost model is registration fees, not APCs — proceedings are open-access on OpenReview. Conference organizers rotate yearly, so re-check the current CFP and committees page rather than carrying a name forward.

Milestones

• Fit decision: confirm the paper has a learning-representation contribution and a credible ICLR audience.
• Evidence lock: freeze the main claim, core baselines, ablations, and reproducibility path.
• Abstract deadline: submit a real abstract and verify author list, conflicts, subject areas, and OpenReview profiles.
• Paper deadline: upload anonymized PDF, supplement, code/data package, disclosures, and required fields.
• Review release: triage reviewer concerns and assign owners for evidence, wording, and revision.
• Discussion: respond, upload allowed revisions, maintain a concise changelog, and escalate only through official channels.
• Decision: archive reviews, meta-review, author discussion, and final submitted version.
• Acceptance: complete camera-ready, poster, slides, video, project page, public artifacts, and registration.

Operating rules

• Maintain one source-of-truth checklist for OpenReview fields and file names.
• Run an anonymity audit after every PDF, supplement, code, or demo change.
• Keep a "reviewer can verify this in five minutes" path for each major claim.
• Do not add new experiments during discussion unless they clarify an already submitted claim and current rules allow the revision.

Phase ownership and risk

ICLR's timeline differs from venues with a short rebuttal: the public discussion phase is long, and revisions you upload remain visible. Plan for an extended, public exchange, not a one-shot rebuttal.

Phase	Owner focus	ICLR-specific risk
Abstract deadline	Profiles, conflicts, real abstract	Placeholder abstract or incomplete profile
Paper deadline	Anonymized PDF, supplement, disclosures	Identity leak in code or links
Review release	Triage, assign evidence owners	Reacting to every score instead of the AC's concern
Discussion	Replies, allowed revisions, changelog	Promising runs the cycle's rules disallow
Camera-ready	De-anonymize, fulfill promises	Final PDF drifts from the meta-review

Worked vignette

A two-person team plans backward from the OpenReview paper deadline for an interpretability paper. They lock the main claim and core ablation early, reserve the final days for an anonymity audit of the PDF and code ZIP, and pre-assign discussion roles: one owns evidence replies, one owns wording and the changelog. When reviews land they already have a "verify in five minutes" probe script, so the public thread stays concrete.

Reviewer-pushback patterns to schedule around

• "We need a new experiment fast." Pre-build a probe or ablation harness so small results are cheap.
• "Anonymity slipped after an edit." Re-run the audit after every PDF, supplement, or code change.
• "Deadline confusion." Treat the current cycle's OpenReview dates as the single source of truth.

Output format

[Current stage] idea / experiments / writing / submission / discussion / accepted
[Next deadline] <official date and source, or unknown>
[Critical path] <three tasks that determine readiness>
[Anonymity checkpoint] pass / needs audit
[Owner map] <task -> person or role>