brycewang-stanford NSDI Review Process
NSDI's review machinery differs from single-deadline conferences in one structural way: decisions are moves in a multi-deadline game, not endpoints. Reading a decision letter correctly requires knowing which moves it opens and which it closes. Mechanics below are the NSDI '27 CFP and the NSDI '26 multiple-deadlines page as rendered on 2026-07-08.
Pipeline shape
Each deadline (spring, fall) runs its own double-blind PC review on its own HotCRP site, producing its own decision cohort — NSDI even publishes per-cohort accepted-paper lists before the symposium. Within a cohort:
- Reviews are written against the paper as submitted; the rendered '27 CFP describes
no author-response phase between reviews and decisions (absence 待核实 — check
the current cycle's pages before relying on it). Plan as if the submission must
argue for itself (
nsdi-author-response). - Review loads are bounded upstream by the eight-submissions-per-author cap, which exists to protect PC bandwidth — expect enforcement, not flexibility.
- Notifications land on published dates: July 23, 2026 for '27 spring; December 8, 2026 for '27 fall.
The three-way outcome space
| Decision | What it means | What it opens / closes |
|---|---|---|
| Accept | into the cohort's program | final paper, artifact evaluation, presenter duties |
| One-shot revision | conditional path: reviewers issue a list of required issues | resubmit at a subsequent deadline (possibly next year's NSDI); same reviewers where possible; one attempt |
| Reject | out, with a ban attached | cannot enter the next deadline; spring reject skips same-edition fall, fall reject skips next edition's spring |
The one-shot revision is neither a rebuttal nor journal-style R&R-until-done. Its grammar: the reviewers' issue list is exhaustive and binding; the revision is judged on whether those issues are fixed; a revision that dodges them is rejected without another revision option. Institutional continuity is built in — the same reviewers re-review where possible, and PC members who issue one-shot revisions at a fall deadline are required to serve as external reviewers the following year, so the issue list is written by people expecting to check it.
What reviewers reward and punish
Reading NSDI reviews (and pre-empting them) is easier with the venue's evaluative axes explicit:
- Design contribution to networked systems or the stack — the scope clause
functions as review criterion; "solid work, wrong venue" is a real NSDI outcome
(
nsdi-topic-selection). - Practical evaluation — trace realism, incumbent baselines, tail visibility
(
nsdi-experiments). - Honesty about limits — named failure regimes read as strength.
- Track-contract fulfillment — operational papers judged on depth of deployed experience; frontiers papers on boldness and framing rather than completeness.
Decoding common letters
"One-shot revision. Required: (1) compare against <deployed system>;
(2) evaluate under churn; (3) clarify lease-expiry semantics."
-> A funded offer. Cost it (~2 machine-months + rewrite), reserve the
testbed now, target the named next deadline. Success odds are good
*because* the same reviewers check their own list.
"Reject. The contribution is an application of known techniques; the
networking stack is unchanged."
-> Scope verdict, not effort verdict. Re-route (SIGCOMM? OSDI?) or
re-center the design contribution. Ban: skip one NSDI deadline.
"Reject. Evaluation limited to microbenchmarks on a two-node setup."
-> Evidence-rung verdict. The ban window is exactly the time to climb
a rung (trace replay, scale-out) before the after-next deadline.
Ban arithmetic, worked
A paper rejected at fall '26 (notified December 9, 2025 — the '26-cycle dates)
could not enter spring '27 (April 2026) and became eligible again at fall '27
(September 2026). Generalize: a plain rejection costs one deadline ≈ 5 months of
eligibility, which is precisely why nsdi-workflow treats speculative submissions
as spending a real option, and why the strongest response to a borderline draft is
often holding it one gate.
Signals to mine from the reviews themselves
Whatever the outcome, NSDI reviews carry venue-calibrated information worth extracting systematically:
- Which track bar was applied — reviews that fault a research-track paper for missing deployment experience (or vice versa) reveal a track mismatch to fix before any resubmission.
- Which baseline the community considers canonical — a named "why not X?" defines the comparison the next version cannot skip.
- Repeated confusion across reviewers — two reviewers misreading the same
mechanism marks a writing defect, not two reading defects
(
nsdi-writing-style). - The evidence rung the scores imply — enthusiasm plus low confidence often
means the idea landed but the evaluation did not reach the rung the claim needs
(
nsdi-experiments).
Confidentiality and conduct
Reviews and PC deliberations are confidential; submissions are read under double-blind rules with declared conflicts recused via HotCRP. Do not contact PC members about a pending paper; process questions go to the program co-chairs (named on the current conference page; the '27 pair 待核实). Decisions are not relitigated by email — the sanctioned channel for disagreement is the revision or the next eligible submission.
Output format
[Cohort] spring/fall + year; notification date; days remaining
[Outcome] accept / one-shot revision / reject
[If revision] issue list -> cost -> feasible target deadline
[If reject] ban window (dates); scope vs evidence diagnosis; re-route call
[Reviewer continuity] same-reviewer expectations affecting strategy
[Next action] single highest-leverage step, dated