brycewang-stanford NDSS Review Process
Each NDSS cycle is a funnel with one public narrowing: Round 1 reads everything and ejects the bottom; Round 2 argues about the rest. Knowing which stage you are in — and what each stage can still change — prevents both wasted effort and missed leverage.
The pipeline (2027 cycle structure, verified 2026-07-08)
submit ──▶ ROUND 1 (breadth read)
│
├─ early reject ──▶ Round-1 reviews released; cycle over for this paper;
│ major-overlap resubmission barred within this
│ symposium's other cycle
│
└─ continue ──▶ ROUND 2 (depth read, more reviewers/discussion)
│ rebuttal + interactive discussion with reviewers
▼
Accept │ Minor Revision │ Major Revision │ Reject
Dates for the fall cycle: early notification September 25, rebuttal and discussion October 8-10, final decision November 4, 2026. The summer cycle followed the same shape (June 12 / June 24-28 / July 22).
What each outcome means
| Outcome | Substance | Author obligation |
|---|---|---|
| Accept | Paper in, Seoul slot earned | Camera-ready by Jan 6, 2027; consider artifact evaluation |
| Minor Revision | Accepted contingent on a bounded fix-up judged satisfactory | Execute exactly the requested changes, promptly |
| Major Revision | Promising but incomplete — new experiments, implementations, or proofs required | Resubmit within the stated window against a written list of revision tasks; re-review checks task fulfillment; at most one major revision per paper |
| Early reject / Reject | Out for this edition | Fold reviews in; retarget (see ndss-workflow) |
The Major Revision list is the closest thing conference review offers to a contract: the original reviewers wrote the tasks, and the resubmission is judged against them — not against fresh whims. Treat each task as a deliverable with evidence; treat silence on a task as a decline you must justify explicitly.
Round 1: what the breadth read screens for
Round-1 reviewers triage many papers quickly. They are deciding whether further review is worth the PC's time, which concentrates attention on: an intelligible threat model, a believable evidence type for the claim category, obvious scope or novelty problems, ethics red flags, and format/anonymity violations. The writing-style skill's first-page contract is calibrated to exactly this read. An early reject is not an insult; it is a verdict that the fast read failed — often fixable, but not this edition.
Round 2: where papers are actually won
Round 2 adds closer reads and a live element: authors see reviews, respond, and then
interact with reviewers during a discussion phase. Decisions crystallize here through
reviewer-to-reviewer argument. Your rebuttal's real audience is the undecided reviewer
and the discussion that follows it (tactics in ndss-author-response). Reviews that
misread the paper are, at this stage, evidence about the paper: something on the page let
the misreading happen.
The Ethics Review Board
Any reviewer can flag a submission as ethically fraught; flagged papers get examined by an ERB drawn from the TPC. It evaluates harm handling against community norms (the Menlo Report is the CFP's named reference), and its judgment can override technical enthusiasm. Papers are not penalized for doing dangerous-looking research — they are penalized for doing it without visible reasoning about harm, consent, and disclosure. The ERB reads the Ethics Considerations section as your testimony; write it before submission as if the board were guaranteed to convene.
Confidentiality and conduct
Reviews, scores, and discussion content are confidential; do not quote them publicly, and do not attempt out-of-band contact with suspected reviewers — chairs are the only channel. Public preprints are a separate question (check the current CFP's stance) but engaging with reviewers about the paper anywhere except HotCRP is a bright line at every security flagship.
Reading your reviews diagnostically
- All reviewers stumble at the same section → structural problem there, not reviewer error.
- Scores split wide → the paper supports two readings; the rebuttal must collapse the wrong one with pointable evidence.
- "Threat model unrealistic" → either defend the model's real-world instantiation with named deployments, or the topic-selection test was failed months ago.
- Every weakness listed is fixable-in-weeks → a Major Revision may be within reach; make the fixes look scoped in your response.
Output format
[Stage] round-1 pending / round-2 / decided: <outcome>
[Cycle dates] next milestone + days remaining
[Review diagnosis] convergent objections, split points, misreadings
[ERB exposure] flagged risk areas + where the paper answers them
[Next action] rebuttal / revision-task plan / retarget, with owner