wsdm-submission
GitHub用于在WSDM投稿前审计论文，重点检查EasyChair设置、摘要与正稿提交节奏、含附录的9页预算、伦理章节、匿名化及无反驳期的风险。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill wsdm-submission -g -y
SKILL.md
Frontmatter
{
"name": "wsdm-submission",
"description": "Use when auditing a WSDM submission before the August deadlines - EasyChair setup, the abstract-then-paper week, the appendix-inclusive page budget, the required ethical-considerations section, anonymization that survives Associate-Chair metadata visibility, and desk-reject exposure at a no-rebuttal venue."
}
WSDM Submission
Audit a paper headed for WSDM's single annual deadline. The stakes are unusual: WSDM historically runs no rebuttal, so whatever enters EasyChair in August is the entire case reviewers will ever see. A fixable-in-rebuttal weakness at other venues is a terminal weakness here. Every number below is a 2026-edition anchor (19th WSDM, Boise); reopen the current edition's CFP before quoting any of it.
The two-deadline week
WSDM pairs an abstract registration with a full-paper deadline one week later (2026: abstracts August 7, papers August 14, 2025, both 23:59 AoE, EasyChair). Use the gap deliberately:
- The abstract you register steers bidding at a venue with a compact, expert PC. A vague abstract lands the paper with the wrong specialists; at a no-rebuttal venue you cannot later explain what the paper "really" meant.
- Placeholder titles or abstracts ("TBD") were an explicit desk-reject trigger in the 2026 short-paper call - assume the same hygiene standard everywhere.
- Author lists and orderings should be final at abstract time. Confirm every coauthor's EasyChair account and conflict declarations that week, not on deadline night.
Page-budget arithmetic: appendices live inside the cap
The 2026 long-paper rule: 9 pages including diagrams, tables, and appendices, with unrestricted space only for references and the ethical-considerations section. The 2025 rule was 8 pages plus at most 2 shared by references and ethics. Two consequences that surprise people arriving from KDD or NeurIPS:
- There is no unlimited appendix to absorb overflow. A proof, a prompt listing, or a hyperparameter table "moved to the appendix" still spends the same budget.
- The only uncounted prose you control is the ethics section - and chairs read it as ethics, not as a smuggling route for extra experiments. Abusing it is a credibility hit in front of the same SPC who writes your meta-review.
% WSDM 2025-cycle documented setting - verify options against the current CFP
\documentclass[sigconf,anonymous,review]{acmart}
% sigconf : ACM two-column conference layout (WSDM is an ACM conference)
% anonymous : suppresses the author block for double-blind review
% review : line numbers for reviewer reference
% Never patch \baselineskip, margins, or fonts: a tampered acmart is visible
% to any chair who compiles the reference template next to your PDF.
Budget the 9 pages backwards: fix the space for the evaluation section first (WSDM reviewers weigh it heaviest), then the method, then shrink the introduction to what remains. Cutting evaluation to preserve a long method narrative is the classic losing trade at this venue.
Anonymization under hybrid blinding
WSDM's CFP describes "a combination of single-blind reviewing and double-blind reviewing": the PDF must be fully anonymized for PC and senior PC members, while Associate Chairs can see submission metadata (including authorship) to check conflicts. Practical readings:
- Do not treat AC metadata visibility as permission to be sloppy - the people scoring the paper still must not be able to guess the authors.
- Industry papers leak identity through infrastructure nouns: an internal system name, a proprietary metric, a "our platform serves X million queries" figure that uniquely identifies one company. Rewrite to role descriptions ("a large commercial search engine") and round figures that fingerprint you.
- Self-citations go in third person; a cited anonymous repository must have a history-free remote, no institutional paths, and no author handles in configs.
- WSDM's anonymization seriousness is not folklore: the venue hosted the 2017 controlled experiment showing single-blind reviewers favor famous affiliations, and its policy is written in that shadow.
Desk-level exposure table
| Failure | Why it is severe at WSDM | Recoverable? |
|---|---|---|
| Over the 9-page inclusive cap | Explicit CFP bound; appendices count | No, after AoE |
| Missing ethical-considerations section | Required element of the submission | No |
| Identity leak in PDF or linked repo | Hybrid-blind policy with an experimental rationale | No |
| Wrong or "TBD" registered abstract | Bad bidding + placeholder desk-reject rule | Only in the abstract-paper gap |
| Claim answerable only "in rebuttal" | There is no rebuttal to answer it in | Only by rewriting before the deadline |
| Ethics section used as overflow appendix | Read by SPC as bad faith | Only before the deadline |
Vignette: the platform paper that fingerprinted itself
A search-quality team anonymizes diligently - author block gone, citations in third person, repository scrubbed - and still gets flagged. The leak chain: the paper reports "our engine's 2.3 billion daily queries in 41 languages," names an internal experiment platform in a figure caption, and evaluates on a locale mix only one company serves. No single item names the employer; the conjunction does. The repair pass that would have caught it:
- Read the paper asking "how many organizations could have written this sentence?" - any sentence with answer "one" gets generalized or rounded.
- Rename internal systems to functional labels ("the interleaving platform").
- Report locale/scale details in buckets wide enough to cover several plausible platforms, with exact values deferred to the camera-ready.
This is also why the audit below schedules the anonymity sweep after the content freeze: anonymity is a property of the final text, and every late paragraph reopens it.
Final-week order of operations
- Freeze scope: anything not already in the evaluation cannot be promised later. Delete forward-looking claims ("we will additionally verify...").
- Recompile from a clean
acmartcheckout; check the page count with references and ethics excluded from the 9, everything else included. - Run the self-containment pass: for each likely reviewer objection, verify the
answer is physically in the PDF (see
wsdm-author-responsefor the objection inventory method). - Sweep metadata: PDF author fields, figure-embedded usernames, dataset paths, acknowledgements, grant numbers, repository history.
- Submit the PDF with hours of AoE margin. EasyChair accepts resubmissions until the deadline - upload early, replace often.
Output format
[WSDM submission audit] Ready / Needs fixes / Blocked
[Budget] 9-page-inclusive check: pass / over by N (appendices counted?)
[Ethics section] present and substantive: yes / no
[Anonymity] PDF / repo / metadata / infrastructure-noun leaks found: <list>
[Self-containment] objections lacking in-PDF answers: <list>
[Abstract-week actions] <items only fixable before the paper deadline>
Version History
- 9f86f09 Current 2026-07-19 17:54
