brycewang-stanford SIGMOD Workflow
SIGMOD project management differs from single-deadline conferences in kind: with four research rounds a year feeding PACMMOD issues, the strategic variable is which round, and the punishment structure (a 12-month embargo on rejection, one month to execute a revision) rewards teams that plan the whole round, not just the deadline. Anchor every date below to the live important-dates page before acting; the quarterly pattern is durable, the dates are not.
The round as a five-act unit
Act 1 Abstract + COIs (deadline minus ~1 week)
Act 2 Paper deadline (the 17th of the round month, AoE, for 2027)
Act 3 Feedback phase (~2 months in; one-week reply window)
Act 4 First verdict (~3 months in: accept / reject / revision)
Act 5 Revision + final (+1 month to resubmit, ~+3 weeks to decide)
Block engineer time for Acts 3 and 5 when you submit, not when they arrive:
the feedback week may need a fast counter-experiment, and the revision month
is a sprint that assumes the original evaluation was scripted end to end
(see sigmod-reproducibility).
Round selection under the embargo
| Situation | Recommended move |
|---|---|
| Evaluation complete, writing mature | Nearest round |
| Core result solid, one comparison missing | Next round; a missing baseline reads as reject, not revision |
| System working, evaluation thin | Skip two; thin evaluations burn the embargo |
| Racing a known competing group | Weigh nearest round against reject risk; PVLDB's monthly gate may beat waiting |
| Paper rejected last round | Different venue — the track is closed for ~3 rounds |
The asymmetry to internalize: submitting early costs a possible year; waiting costs three months. Under uncertainty, waiting is usually the cheaper mistake at this venue.
Backward plan from a paper cutoff
- T-16 weeks: freeze the system's architecture; feature work after this point invalidates measurements.
- T-10: evaluation matrix locked (workloads × baselines × sweeps); harness and seeds committed.
- T-6: full experiment pass complete once; draft carries real numbers, not placeholders.
- T-4: internal adversarial read by a colleague who builds systems; schedule their week now.
- T-2: rerun-from-scratch of every figure; repro ledger shows zero DEBT.
- T-1 week + a few days: abstract, authors, topics, COIs into CMT.
- T-0: PDF up; verify the uploaded file, not the local one.
Team coordination specifics
- Name one owner for the measurement harness and a different owner for the anonymity sweep; both jobs fail under shared custody.
- Keep a submissions ledger across the group: PACMMOD's exclusivity rule and the 10-papers-per-author annual cap are portfolio constraints, not per-paper ones.
- Archive the exact submitted PDF, the CMT confirmation, and the frozen artifact snapshot; the revision letter must quote and diff against them.
- Pre-agree the reject plan (PVLDB? ICDE? rework?) while everyone is calm; deciding it the day the verdict lands produces venue-shopping errors.
Interleaving multiple projects
Quarterly rounds let a group pipeline: project A in Round 1, its revision landing while project B targets Round 3. The trap is revision collision — two revisions in the same month exceed most teams' experiment bandwidth. When two papers are near-ready, split them across non-adjacent rounds deliberately rather than shipping both into one.
Small-team and solo variants
The five-act structure assumes bench depth; without it, adjust:
- A solo author cannot staff a feedback week and a revision month on top of other obligations — prefer the round after the one that looks feasible, and treat the slack as the plan, not padding.
- Compress the evaluation matrix before compressing run hygiene; fewer workloads with clean provenance beat a wide matrix that cannot be regenerated during a revision.
- Recruit the adversarial reader early even when solo — the T-4 external read is the step small teams skip and most regret at verdict time.
- Consider whether PVLDB's monthly gate suits a solo pipeline better; a missed monthly deadline costs weeks, a failed SIGMOD round costs a year.
Conference layer
Publication and presentation decouple: your issue ships on the round's
schedule, and the talk happens at the next conference edition (SIGMOD 2027:
June 13–19, 2027, Huntington Beach; SIGMOD 2026 was May 31–June 5,
Bengaluru). Budget registration and travel when the acceptance lands, and
schedule the ARI artifact push (sigmod-artifact-evaluation) into the gap
between issue and conference.
Calendar-keeping discipline
Because every date in this skill perishes, institutionalize the refresh: at the start of each quarter, one team member re-opens the current important-dates page, diffs it against the team's planning calendar, and posts the delta. Rounds have shifted between cycles before; a plan anchored to a stale calendar fails silently until the abstract deadline passes. The same quarterly sweep should touch the ARI edition page and, for groups with industrial collaborations, the industrial-track CFP status.
Output format
[Target round] round + year, with selection rationale
[Stage now] pre-abstract / submitted / feedback / verdict / revision
[Backward plan] next three milestones with owners
[Bandwidth check] feedback + revision windows staffed yes/no
[Portfolio constraints] cap headroom, exclusivity conflicts, collisions
[Reject plan] pre-agreed next venue