sensys-submission
GitHub用于SenSys论文提交前的最终预上传审计。检查正确HotCRP站点、截止日期及AoE时区转换；验证12/6页双栏格式限制；执行严格的双盲审查，排查硬件和部署中的身份泄露；确认二稿重投需包含实质性修改及回复信。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill sensys-submission -g -y
SKILL.md
Frontmatter
{
"name": "sensys-submission",
"description": "Use when running the final pre-upload audit of a SenSys submission — confirming the right per-edition HotCRP site and which of the two deadlines is open, the AoE cutoff in local time, the 12\/6-page double-column cap, the double-blind sweep including hardware and deployment leaks, and the resubmission-with-revision contract at the second deadline."
}
SenSys Submission
Order the checks by how permanent the mistake is. A paper aimed at the wrong deadline, or a
first-deadline reject re-entered without its required revision, is bounced before a reviewer ever
opens it — nothing recovers that. Formatting and anonymity slips, by contrast, are fixable right
up to upload. So the audit below front-loads the irreversible gates and leaves the mechanical
sweeps for last. Treat every number as the 2026/2027 snapshot; the live HotCRP /deadlines page
overrides it.
Gate 1 — the deadline is real and yours
SenSys submits through a per-edition HotCRP instance (e.g. sensys27.hotcrp.com) and offers
two deadlines per cycle, each fully reviewed on its own. Confirm on the deadlines page:
- Whether the open deadline is the first or the second, and which edition it feeds — a mid-cycle submission competes for that edition's program, not one already decided.
- Whether this edition requires abstract registration ahead of the paper, and its date.
- That the cutoff is AoE, then convert it to each coauthor's wall clock — a Paris or New York time printed on the page is nobody's local midnight.
The resubmission gate is the one authors miss: if you are re-entering a first-deadline reject at
the second deadline, the submission is invalid without a substantive revision and a "Response
to Reviewers" tying each prior concern to a concrete change (sensys-author-response). A
repackaged reject is a desk reject.
Gate 2 — the geometry is legal
| Item | Rule (2027 cycle) | How to check |
|---|---|---|
| Full paper | ≤ 12 pages including figures and tables | Count the compiled PDF |
| Short paper | ≤ 6 pages (vision/experience track) | Confirm which track you entered |
| References | unlimited, beyond the cap | Nothing load-bearing tucked among them |
| Appendices | optional, after references, beyond the cap | Nothing a reviewer must read to believe the claim |
| Template | ACM double-column, geometry from the class file | Diff against the template; never resize by hand |
At SenSys the cap pressure is figures — power traces, latency CDFs, and testbed schematics eat
the body. Relieve it by cutting content, not by \vspace tricks or shrinking a measurement plot
past legibility; production and reviewers both notice. The exact template version and font floor
are 待核实 for the cycle — read the CFP.
Gate 3 — nobody can tell whose paper it is
Beyond the standard anonymization (no names, affiliations, funding lines, or acknowledgments; third-person self-citation; references intact), SenSys papers leak identity through their hardware and deployments. Sweep the exact upload candidate:
# 1. text + metadata identity
pdftotext final.pdf - | grep -niE 'acknowledg|funded by|grant no|@[a-z]+\.(edu|com)'
pdfinfo final.pdf | grep -iE 'author|creator|producer'
# 2. paths and un-blinded artifact hosts
grep -rniE 'github\.com/[a-z0-9_-]+|zenodo\.org|/home/[a-z]+|/Users/[a-z]+' *.tex figures/
# 3. the SenSys-specific tell: a nameable site, testbed, or deployment
pdftotext final.pdf - | grep -niE 'our (lab|building|campus|deployment|testbed)|[A-Z][a-z]+ Hall|room [0-9]+|IRB'
Strings are the easy part. A "30-node testbed across our engineering building" beside a photo of a distinctive atrium narrows to one group; soften identifying specifics to scale-descriptive language. And every artifact URL must resolve to a blinded repo or an author-stripped Zenodo deposit. A machine finds the strings; only a human reading for implication catches the rest — hand the PDF to a non-author and ask what lab they'd guess.
Gate 4 — the form and the policies
The HotCRP record is read before the PDF: title and abstract must match the paper word for word; topic tags steer the paper toward the sensing / embedded / IoT / on-device-AI reviewers you want bidding, so pick for reviewer fit rather than breadth; and the full author list is registered for conflict computation even while the PDF stays blind — a late author addition after assignment is a chairs-level problem. Declare conflicts exactly as the site defines them. Then clear the policy gates: no concurrent submission elsewhere; a stated delta for any overlapping workshop or poster; human-subjects and third-party-infrastructure handling addressed (wording 待核实); and the cycle's AI-use rule read live rather than assumed.
Run-of-show
T-8d account on the right instance + deadline; form fields drafted; conflicts entered
T-7d abstract registered if required; topic tags chosen for reviewer fit
T-2d content freeze; Gates 2-3 run on the near-final PDF; resubmission packet attached
T-1d coauthors sign off; HotCRP title/abstract match the PDF verbatim
T-0 upload → status ready; then re-download the stored file and open it — only server bytes count
Report back
[Gate1] instance + first/second deadline + edition fed + AoE cutoff in local time
[Gate2] full/short track + page/column/template verdict
[Gate3] mechanical hits + human-read verdict (deployment + artifact-link giveaways)
[Resub] revision + Response-to-Reviewers attached? (second-deadline re-entry only)
[Gate4] form verbatim-match + conflicts + dual-sub/ethics/AI each cleared or flagged
[Open] live blockers, most-permanent first, each with an owner and hours-to-fix
Version History
- 9f86f09 Current 2026-07-19 17:41
