sosp-review-process
GitHub解析SOSP论文评审全流程，涵盖双盲HotCRP审稿、分阶段评审、作者回复及PC会议决策。提供评审重点分析（如正确性攻击）和会后分类模板，辅助作者把握关键节点与应对策略。
触发场景
安装
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill sosp-review-process -g -y
SKILL.md
Frontmatter
{
"name": "sosp-review-process",
"description": "Use when reasoning about how a SOSP submission is evaluated — double-blind HotCRP reviewing, staged review rounds, PC-chair-audited conflicts, the pre-PC-meeting author response, the single-track PC meeting where every decision is argued live, notification, and shepherded acceptance outcomes."
}
SOSP Review Process
Use this to model what happens to a SOSP paper between the April upload and the July
decision, and to time author-side actions to the phases that actually move outcomes.
Anchors below are from the 2026-cycle CFP and the sosp26.hotcrp.com schedule
(checked 2026-07-08): submission April 2026, author response ahead of the PC meeting,
notification July 3, 2026.
The pipeline
| Phase | Who acts | What authors can do |
|---|---|---|
| Conflict screening (from abstract registration) | PC chairs audit declared conflicts, add or remove entries | Declare defensibly; chairs treat strategic conflicts as misconduct-adjacent |
| Early reviewing | A subset of the PC reads every paper | Nothing — the paper speaks alone |
| Later-round reviewing | Papers still alive draw additional, often more senior, reviews | Nothing; round structure and counts vary by year (待核实 per cycle) |
| Author response | Authors correct factual errors, answer posed questions | The one sanctioned intervention (see sosp-author-response) |
| PC meeting | Full committee discusses candidates live, single track | Nothing — but the response was written for this room |
| Decision + shepherd assignment | Chairs notify July 3; accepted papers get a shepherding PC member | Camera-ready planning begins (see sosp-camera-ready) |
Two structural facts do most of the explanatory work. First, reviewing is staged:
a paper that reads as unsound in the first pass never reaches the reviewers most
likely to appreciate a subtle contribution, so the opening pages must survive a
skeptical fast read (see sosp-writing-style). Second, decisions are made in a
meeting, not by score averaging: one convinced advocate who can defend the paper
against live objections beats four mildly positive scores.
Reading SOSP reviews
Systems reviews concentrate their weight in predictable places:
- Soundness attacks ("the evaluation doesn't isolate the claimed mechanism", "this workload can't show the tail behavior claimed") are the score-drivers. Address these first in any response.
- Positioning complaints ("how is this different from X at OSDI two years ago")
signal a related-work gap the meeting will replay (see
sosp-related-work). - Taste judgments ("incremental", "not SOSP-worthy") rarely change through argument; they change when an advocate reframes the paper's principle at the meeting — which your response can quietly arm with one crisp sentence.
Post-review triage sheet (fill before drafting anything)
For each review R1..Rn:
overall lean: champion / positive / neutral / negative / kill
score-driving issue: <one sentence>
type: factual error | missed content | posed question | judgment
response-eligible?: yes (types 1-3) / no (type 4)
Meeting risk: <the objection an advocate could not answer alone>
Confidentiality and integrity edges
- Reviews and PC discussion are confidential; reviewers are anonymous forever, and guessing at reviewer identity in the response reads badly (reviews are anonymous in both directions).
- Double-blind holds until notification. Do not advertise the submission where PC members will see it; the CFP's anonymization duty does not lapse after upload.
- HotCRP is the channel of record. Deadline-adjacent questions go to the PC chairs through it, not to PC members you happen to know — that is a conflict violation in a culture that audits conflicts.
Calibrating expectations
SOSP is a low-acceptance venue where most submissions are competent; rejections
routinely carry reviews that would be accepts elsewhere. Per-year submission counts
and acceptance rates are published with each program (待核实 for 2026 until the
program appears; cross-check dl.acm.org/conference/sosp and dblp for history).
Practical consequences of the meeting-centric model:
- A rejected paper with one strong champion review is often a revise-and-retarget candidate for the next systems deadline, not a redesign candidate.
- Reviews written for the meeting compress; a terse negative review may encode a long live discussion. Mine every sentence — each one survived an argument.
- Because SOSP and its sibling OSDI both now run annually, the cost of a rejection
is measured in months, not years (see
sosp-workflowfor the retargeting map).
Output format
[Phase] where the paper sits in the pipeline now
[Review map] per-review lean + score-driving issue + response eligibility
[Meeting risk] the one live objection to defuse
[Action] respond / wait / prepare retarget
[Integrity check] anonymity holds? all contact via HotCRP?
版本历史
- 9f86f09 当前 2026-07-19 17:38
