--- name: brycewang-stanford/acl-workflow source: https://app.decimal.ai/s/brycewang-stanford-acl-workflow@1/SKILL.md source_sha256: 024195d42a8a --- # ACL Workflow Use this as the project-management layer for an ACL paper. Unlike one-deadline conferences, ACL runs on a **cycle-and-commit** rhythm: pick an ARR cycle, ride its ~10-week review train, then decide at the commitment window whether to send the package to the conference or loop back for another cycle. ACL is a conference with rotating per-edition program committees, not a journal: there is no standing editor and no APC. Costs are registration-based, and Anthology publication is free and open access. ## The 2026 rhythm as a worked anchor ACL 2026 (64th edition, San Diego, July 2-7, 2026) drew from the ARR October 2025 and January 2026 cycles: | Milestone | Date (verified for ACL 2026) | |---|---| | ARR January cycle submission | January 5, 2026 | | Reviews + meta-reviews to authors | by March 9-10, 2026 | | Commitment to ACL 2026 | March 14, 2026 | | Acceptance notification | April 4, 2026 | | Camera-ready | April 19, 2026 | | Conference | July 2-7, 2026 | Later 2026 cycles (March 16, May 25, August 3, October 12 submissions) feed EMNLP/AACL 2026 and EACL 2027 instead. Every future date is cycle-volatile — re-read aclrollingreview.org/dates before committing a team to any of them. ## Choosing a cycle - Work backward from the conference you want: each \*ACL conference announces which cycles it accepts commitments from. The *earlier* eligible cycle buys a free resubmission shot at the *same* conference if reviews go badly. - A cycle deadline you hit with a rushed paper is worse than the next cycle hit with a finished one — reviews persist and follow resubmissions. - Mind the response window inside the cycle (a few days, roughly five weeks after submission): the team must be available then, not just at submission. ## Backward plan from a cycle deadline | Weeks out | Milestone for an experiments-heavy NLP paper | |---|---| | 8 | Claim fixed; datasets, baselines, and evaluation protocol frozen | | 6 | Main results in; error-analysis sampling started | | 4 | Human evaluation fielded; ablations running | | 3 | Full draft in ACL style; internal red-team read | | 2 | Responsible NLP checklist drafted against the real paper | | 1 | Limitations written; anonymity sweep; supplement archive built | | 0 | Submit early on deadline day (AoE) | ## The resubmission loop ```text cycle N reviews weak? -> revise (new experiments, rescoped claims) -> cycle N+1 submission WITH: link to prior submission, change summary, point-by-point response to old weaknesses -> commit the stronger package to the next eligible conference ``` Budget one loop into any plan that matters: the modal strong ACL paper today has seen two cycles. Skipping the revision note is a desk reject; a floor meta-review score forbids quick resubmission without wholesale revision. ## Post-decision branches - **Main conference**: camera-ready sprint (see `acl-camera-ready`), registration, presentation prep, artifact release. - **Findings**: same camera-ready track; decide early on the presentation option the edition offers and whether to present at an affiliated workshop. - **Reject**: hold a post-mortem against the meta-review within a week while context is fresh; choose next cycle + venue deliberately, not reflexively. ## Coordination rules that save cycles - One owner for the checklist, one for anonymity, one for the supplement — the three most common last-day failures are all ownerless tasks. - Freeze a "submitted" tag in the repo: the author response may only cite what that tag contains. - Calendar the response window and the commitment deadline the day you submit; both are short and both are missable. ## Running multiple papers through the machine - Stagger sibling papers across cycles rather than stacking one deadline: the response windows will otherwise collide, and response quality is the highest-leverage hour-for-hour work in the pipeline. - Track each paper as (cycle, state, target venue, fallback venue) — the fallback is a real decision because commitment windows do not wait. - Reuse infrastructure deliberately: one anonymization script, one checklist template pre-filled with lab defaults, one supplement layout. - Reviewer-duty obligations may attach to submissions in current cycles; budget senior-author reviewing time into the same calendar (verify the current cycle's policy — this has changed over the years). ## Signals to re-plan immediately - The dates page announces a cycle-structure change (it has happened multiple times; 10-week cycles arrived only in May 2025). - Your target conference posts its eligible-cycle list and your intended cycle is not on it. - Reviews arrive with a fixable core objection and the next cycle's deadline is within two weeks — a rushed loop beats a stale package only if the fix is genuinely small. - A competing preprint lands mid-cycle: plan the response-window framing and the camera-ready citation now, not at notification. ## Output format ```text [Stage] planning / cycle-N under review / response window / commit window / accepted / revising [Target] [Critical path] [Loop budget] [Owner map] ```