brycewang-stanford SODA Review Process
SODA is run jointly by the SIAM Activity Group on Discrete Mathematics and ACM SIGACT, with a fresh program committee appointed for each edition — there is no standing editorial board, and every policy below is re-decided annually. Anchor facts for the 2027 cycle (checked 2026-07-08 via SIAM SODA27 pages read through search renderings, plus organizer announcements): submission July 9, 2026 AoE on HotCRP; reviews to authors by September 1; responses due September 4; decisions and instructions in October 2026. The 2027 PC chairs were not yet indexed at check time: 待核实. For scale context, SODA 2026's chair reported the largest submission pool in SODA history, roughly a 25% jump year-over-year (reported via the PC chair's public announcement; treat as reported, not official).
Pipeline anatomy
| Phase (2027 anchors) | What happens | Author-side reality |
|---|---|---|
| July 9 | HotCRP closes | The record is frozen; nothing is added later |
| July | Bidding and assignment | Your title-page abstract and topic tags determine who reads you |
| July-August | Refereeing, often with external subreviewers | The full version is the reviewed object; depth checks are real |
| By September 1 | Initial reviews released | Read for factual errors first, opinions second |
| September 4 | Responses due | Three days; preparation happened in August (soda-author-response) |
| September-October | PC discussion and ranking | Reviews are debated; the response is evidence in that debate |
| October | Decisions + final instructions | Accept/reject; no shepherding tier by default (待核实 per cycle) |
Who reads you: the subreviewer economy
A large-pool theory conference distributes much of its reading to external subreviewers — domain experts recruited per paper by PC members. Consequences:
- Expect one deep technical review (the subreviewer who works in your niche) and
broader reviews weighing significance. Write for both: airtight proofs and a
first-ten-pages case for why the bound matters (
soda-writing-style). - Subreviewers check lineage. A comparison table that omits the actual state-of-the-art bound will be caught by the one person who proved it.
- Expertise variance is the norm, not a scandal. The rebuttal exists partly to arbitrate when the deep review and a skim review disagree.
Lightweight double-blind, referee's view
Referees see no author names, but SODA's rules keep the literature intact: authors post to arXiv freely and cite themselves in third person. In practice many referees can guess authorship; the process is designed to remove the default weight of names from triage, not to guarantee ignorance. Author-side consequences: never rely on reputation carrying a thin section, and never burn words hiding a self-citation the rules told you to keep visible.
What moves a borderline paper
PC discussions compress each paper into a few sentences. The compressible virtues:
- A named open problem or barrier the result resolves or bypasses — with the citation where it was posed.
- A clean headline bound: "first deterministic near-linear time" survives compression; "improves several parameters in some regimes" does not.
- A transferable technique a PC member can imagine using elsewhere.
- A decisive rebuttal correction of the most negative review's factual core.
And the compressible sins: a bound worse than known work (fatal), a proof the
deep review could not verify (near-fatal), scope mismatch better served by a
satellite or sibling venue (soda-topic-selection).
Interpreting the calendar's silences
Author anxiety maps poorly onto PC reality; this table converts one to the other for the 2027 anchors:
| What you observe | What it usually means | What it does not mean |
|---|---|---|
| No confirmation beyond HotCRP receipt in July | Normal; assignment is in progress | Nothing about your paper's fate |
| Reviews arrive well before September 1 | A punctual referee pool | Neither enthusiasm nor doom |
| One review far shorter than the others | A triage-level read alongside deep ones | Not automatically negative — check its F/L content |
| A referee question phrased as a request for computation | Your rebuttal's highest-value target | Not an acceptance signal |
| Silence from September 5 to October | Deliberation; scores are being argued | No news is literally no news |
| Decision earlier or later within October | PC logistics | Nothing paper-specific |
The single decision authors control in this period is rebuttal quality; the single mistake available is contacting PC members outside HotCRP, which converts a borderline paper into a conflict-of-interest incident.
Reading a SODA review packet
For each review, classify every substantive sentence:
F factual claim about your paper -> verifiable; rebut if wrong
L literature claim (prior bound, who did what) -> check the citation yourself
S significance judgment -> rebuttal leverage is low
C correctness doubt -> highest priority; answer with page numbers
Then decide: does the packet's most negative review rest on F/L errors
(rebuttable) or on S consensus (usually terminal for this cycle)?
After the decision
- Accept: October instructions start the camera-ready track
(
soda-camera-ready). - Reject with correctness doubts: repair the proof before any resubmission; theory reviewer pools overlap heavily across SODA, STOC, FOCS, ESA, and ICALP, and a resubmitted known bug follows the paper.
- Reject on significance: re-route rather than re-argue — ITCS for
conceptual framing, ESA/ICALP for strong-but-specialized algorithmics, SOSA if
the real contribution is a simplification (
soda-workflowhas the calendar).
Output format
[Process stage] <where the paper sits in the 2027 pipeline>
[Review triage] <F/L/S/C classification summary per review>
[Rebuttal leverage] <high/medium/low, with the deciding factor>
[Post-decision route] <camera-ready / repair / re-route target>