sosp-workflow

SOSP Workflow

Use this to place a project on the SOSP calendar and to decide what to do in the current phase. SOSP has been annual since 2024 (the SIGOPS announcement is at sigops.org/2023/sosp-is-going-annual/), which changed campaign planning from a two-year bet into a rolling circuit. All dates below were checked 2026-07-08.

The 2026 cycle as the template

As of 2026-07-08 the notification has just landed: teams are either five days into the acceptance track or five days into retargeting. SOSP 2027 dates are unpublished (待核实 at sigops.org and sosp.org); the annual pattern implies a spring 2027 deadline, but never back-schedule against an implied date — only against a posted one.

Accepted: three threads, two months

July 3 to August 28 runs three deadline-bound conversations in parallel, and the standing failure is one person owning all three:

1. Shepherd thread — revision plan within days, agreed scope, 13th/14th page negotiation (see sosp-camera-ready). Owner A.
2. Artifact thread — sysartifacts registration opens days after notification; the evaluation window is interactive (see sosp-artifact-evaluation). Owner B.
3. Logistics thread — ACM rights form, registration, Czech-entry visas where needed, talk preparation. Owner C, earliest start wins on visas.

Rejected: the retargeting circuit

Annual SOSP plus annual OSDI plus the EuroSys/ATC/NSDI ring means a systems paper is rarely more than a few months from its next natural deadline. Retarget on a decision tree, not on grief:

• Reviews attack soundness (evaluation holes, unisolated mechanisms): fix the evidence first; the next deadline is whichever one the new experiments make.
• Reviews attack fit or significance but concede soundness: the paper may be right for a sibling with a different center of gravity — networked-systems framing to NSDI, storage framing to FAST, deployment/practitioner framing to ATC.
• Reviews are strong with one blocker: an OSDI resubmission in the next window with a surgical fix and a cover-your-delta introduction paragraph is the highest expected-value move. Exact sibling deadlines shift every cycle (待核实 against each venue's current CFP before promising a date to co-authors).

Back-scheduling the next submission

T-9 months   problem locked; motivating measurement exists; skeleton design review
T-6 months   system runs end-to-end on one workload; claim list drafted
T-4 months   evaluation matrix frozen (sosp-experiments); baselines built and tuned
T-10 weeks   full experimental sweep begins; env capture automated (sosp-reproducibility)
T-6 weeks    complete draft; strict 15-minute read by an outside systems reader
T-3 weeks    evaluation freeze; figures one-command from logs
T-10 days    format compile, leak sweep, conflict list drafted (sosp-submission)
T-6 days     abstract registration
T-0          upload; verify the HotCRP-held PDF, not the local one

The two intervals teams misjudge: baselines (building and fairly tuning someone else's system routinely costs a month) and the draft-to-freeze gap (a SOSP draft that first exists three weeks before the deadline reads like it).

Cadence discipline for a group

• Track the circuit in one shared calendar: SOSP, OSDI, EuroSys, ATC, NSDI, FAST, plus HotOS in odd rhythm for early-stage ideas — each with abstract and paper gates, refreshed from official CFPs each fall and spring.
• One project, one primary venue bet, one named fallback — deciding the fallback before notification removes the week of post-rejection drift.
• Never submit the same work to overlapping review cycles; systems venues share PC members and treat dual submission as an integrity violation.

Output format

[Phase] where this project sits on the cycle today
[Owners] shepherd / artifact / logistics threads assigned?
[If rejected] soundness vs fit diagnosis -> named next venue + posted deadline
[Back-schedule] T-milestones with dates against the next posted deadline
[Calendar risk] the interval currently underestimated