--- name: cikm-author-response description: Use when preparing author-side communication around CIKM reviews — drafting for a response window if the cycle offers one (unconfirmed for 2026), writing camera-ready revision notes that answer reviewer concerns, handling post-decision chair correspondence, and converting rejection reviews into a resubmission brief. --- # CIKM Author Response CIKM's response mechanics are genuinely cycle-dependent: no official statement confirming or denying a 2026 rebuttal phase could be retrieved (source map, 2026-07-08, 待核实). This skill therefore covers all four author-side writing situations around CIKM reviews, with the no-window case treated as the default — because when there is no rebuttal, the reply to reviewers gets written anyway, just in different documents. ## Situation map | Situation | The document you actually write | Governing constraint | |---|---|---| | A response window opens | Anonymous reply inside EasyChair | Announced word/format rules; anonymity persists | | No window, paper accepted | Camera-ready revision note | Address concerns without exceeding accepted scope | | No window, paper rejected | Resubmission brief for the next venue | Honest triage, no venue-blaming | | Factual process error | Confidential note to PC chairs | Facts only; never argue merit this way | ## If a window opens: the tri-lane reply CIKM reviews arrive from up to three community lanes (`cikm-review-process`), so a good reply is organized by reviewer, not by theme — each lane's reviewer should find their own objection answered in their own vocabulary: 1. Lead each reviewer block with the concern that most threatens *their* score, not the one easiest to rebut. 2. Answer from submitted material — section, table, appendix, or the artifact the paper already cites. New results pasted into a reply carry no weight a reviewer is obliged to credit and can violate window rules; check the announcement. 3. Concede real limits in one sentence and scope them ("holds for static graphs; streaming is future work") — the blended pool punishes evasion harder than limitation, because at least one reviewer is expert enough to see through it. 4. Preserve anonymity end to end: no names, no identifying URLs, no "as our earlier system at [company]" slips. The double-blind regime does not lapse at review. ## Micro-example: the mining-lane scalability objection > Reviewer: "Complexity analysis suggests quadratic blow-up on dense graphs; the > experiments avoid this regime." Reply skeleton: (a) confirm the reading is correct for worst-case density; (b) point to the submitted Table/appendix entry showing observed scaling on the densest tested graph; (c) state the density regime of the target application and that the claim is scoped to it; (d) offer a one-sentence camera-ready scope clarification. Four moves, no new experiments, no evasion. ## No window, accepted: the revision note The August 7 → August 20 camera-ready sprint (2026 dates) is where reviewer concerns get answered in practice. Build a concern → edit ledger: every substantive review point maps to a camera-ready change, a scoped-limitation sentence, or a reasoned no-change. Keep the ledger; if the proceedings process or a chair asks what changed, the answer is ready, and `cikm-camera-ready` consumes it directly. ## No window, rejected: the resubmission brief Write it within a week, while context is fresh: - Sort review points into: fix-with-evidence, fix-with-writing, disagree-with-reason. - Tag each point by reviewer lane — lane-clustered objections identify which community the paper failed, which drives the `cikm-workflow` fallback choice. - Draft the one paragraph a future reviewer at the next venue would need to see changed. If that paragraph cannot be written, the problem is the contribution, not the venue. ## Tone calibration for a three-community room CIKM's communities have different argument cultures — IR discussion leans protocol-precise, mining leans delta-and-numbers, KM/database leans systems-pragmatic — but one register works for all three: technical, specific, and unwounded. Practical rules: - Quote the reviewer's actual words before answering them; paraphrase invites "that is not what I said." - Numbers over adjectives: "Table 3 row 4 shows 2.9 points at k=10" persuades every lane; "we respectfully believe the improvement is substantial" persuades none. - Never grade the reviewer ("the reviewer misunderstands...") — write "the submission states this compactly in §4.2; we will make it explicit" and let the evidence do the accusing. - Budget by leverage: one blocking objection answered completely beats five minor points each answered at 20%. ## Verify-before-writing checklist Response mechanics are exactly the kind of fact this venue changes per cycle, so before drafting anything reviewer-facing: ```text [ ] Reopen the current cycle's author instructions / notification email: does a response channel exist at all this year? [ ] If yes: word/character limit, format, deadline (AoE?), and whether new results or links are permitted or barred [ ] Anonymity rules during the window (assume double-blind persists) [ ] Who may submit: contact author only, or any author? [ ] For chair notes: the stated contact route on the host site, not a guessed email address ``` Every item comes from the current host site or EasyChair instance — the 2026 source map deliberately records the rebuttal question as unresolved (待核实), and this checklist is how it gets resolved for whatever cycle is live. ## The one-page ceiling Whatever the channel — window reply, camera-ready note, or chair correspondence — hold to roughly a page of substance per audience. Beyond that length, the reader is skimming, and a skimmed concession does as much damage as an unmade one. If the honest response needs three pages, the paper had three pages of problems; choose the one whose resolution changes the decision and spend the page there. ## Output format ```text [Situation] response-window / camera-ready note / resubmission brief / chair note [Per-reviewer leads] [Evidence anchors] [Concessions] [Deliverable deadline] ```