PODC Author Response
Write the response only if the cycle has one — and confirm first. PODC does not always run an
author-response/rebuttal phase, and whether the 2026 cycle does is 待核实; check the call before
planning around one. If there is no rebuttal, all of this energy belongs before the deadline in the
submission itself (podc-experiments, podc-reproducibility). If there is a rebuttal, it is short,
high-stakes, and about the proof, not about persuasion.
First: is there a rebuttal at all?
- Confirm on the current call / HotCRP whether a response phase exists and its length limit.
- If none: there is no second chance this cycle — the decision is on the submission as written. Skip
to the "no-rebuttal" note below.
- If one exists: it is typically a brief, one-round text response to the reviews, still under
lightweight double-blind (do not de-anonymize in the response).
What a PODC rebuttal can and cannot do
A rebuttal moves a borderline proofs paper when it:
-
Corrects a factual misreading of the proof — "Reviewer 2 reads Lemma 3 as assuming synchrony;
it assumes only eventual delivery, as stated in the model box (§2)." This is the highest-value
move: a reviewer who misread the argument may become an advocate once corrected.
-
Supplies a missing step a reviewer flagged — a short proof of a lemma, a clarification of a
case the reviewer thought was open, or a pointer to where in the full version it already appears.
-
States a bound or assumption precisely — resolving an apparent gap that was really a notation
ambiguity.
A rebuttal does not move a paper when it:
-
Argues taste or significance — "we believe this is important" changes no mind; the committee
weighs significance among themselves.
-
Adds a new result — you cannot repair a fundamentally incomplete paper in a rebuttal; there is
no revision round to carry new material into the final version safely.
-
Contests a real proof gap — if the gap is genuine, no wording fixes it; concede and plan for a
later venue.
Prioritize by axis
Answer each reviewer on the axis they actually raised (podc-review-process):
[Correctness question] answer first and concretely: point to the model box, or give the short proof
[Model/assumption doubt] clarify exactly which assumption is used and where it is declared
[Novelty/delta doubt] cite the precise prior bound and restate your delta in model+cost terms
[Significance opinion] acknowledge briefly; do not spend the word budget litigating taste
Spend the limited length on the soundness and model questions that can actually flip a vote, not
on the subjective ones that cannot.
Stay inside the anonymity envelope
- The response is still double-blind: do not reveal author names, affiliations, prior-work
authorship ("in our earlier paper"), or grant/institution identifiers.
- If you must reference your own prior work to answer a question, keep it third-person, as in the
submission (
podc-related-work).
- Do not link to a de-anonymizing arXiv page or repository in the response.
Tone and structure
-
Open with the single most consequential correction, not with thanks-boilerplate.
-
Be specific and verifiable: section/lemma numbers, one-line proofs, exact bounds. A reviewer
will re-check what you claim.
-
Concede cleanly where a reviewer is right; a targeted concession plus a fix elsewhere reads as
competence, not weakness.
-
Stay brief. Respect the length limit; a focused response on the two pivotal points beats a
point-by-point wall that buries them.
If there is no rebuttal (the common case)
- The decision is final on the submitted paper; there is no mechanism to add the missing lemma now.
- Channel the lesson into the resubmission: close the flagged gap, prove the matching lower bound, or
reroute to DISC / a focused venue / a journal (
podc-topic-selection).
- Post the corrected full version to arXiv so the community sees the complete, fixed argument.
Common failures
-
Planning a rebuttal that does not exist this cycle — confirm first.
-
Arguing significance instead of answering the correctness question.
-
Promising a fix as if there were a revision round to deliver it in.
-
De-anonymizing in the response text or a linked page.
-
Ignoring the reviewer's actual axis and answering a different objection.
Output format
[Rebuttal exists?] yes (length limit: <n>) / no — confirmed on the call
[Pivotal points] the 1-2 corrections that can actually flip a vote
[Per-review answers] review point -> axis -> concrete, verifiable response (or clean concession)
[Anonymity] response free of author identity and de-anonymizing links?
[If no rebuttal] resubmission/reroute plan; corrected arXiv full version posted?