SIGCOMM gives you one main-track deadline a year, so this audit is unforgiving by design.
Reopen the current CFP, submission page, and the HotCRP /deadlines clock before trusting
any number below; the 2026 figures are a dated snapshot.
SIGCOMM 2026 opened abstract registration on January 30, 2026 and full-paper upload on February 6, 2026, both Anywhere on Earth. The abstract step is a gate, not a courtesy: no registered abstract, no paper a week later. Register the title, author list, conflicts, and topics even at 60% confidence — a registered abstract can be abandoned, an unregistered one cannot be resurrected once the clock closes. Because there is no second gate this cycle, treat the AoE cutoff as the true horizon and convert each coauthor's local time from the HotCRP page rather than memory.
The two lanes are judged by different contracts, and choosing wrong wastes the whole year:
| Item | SIGCOMM 2026 rule | Audit action on the compiled PDF |
|---|---|---|
| Body | ≤ 12 single-spaced pages before references, including figures and tables | Count pages up to the references; over-length is not reviewed |
| References | as many pages as needed, outside the cap | Confirm nothing load-bearing hides there |
| Appendices | optional, after references, outside the cap | Body must stand without them (a shepherd later judges necessity) |
| Template | ACM acmart.cls (or ACM Word), unmodified |
Diff class options against the distributed file |
| Track | research / experience | Matches the evidence actually in the paper |
The page cap at SIGCOMM is figure pressure — figures and tables count against the 12 pages. The remedy is content triage, never margin, spacing, or font surgery; the ACM production path and reviewers both notice geometry games.
Author identity must be absent from the PDF text, plots, captions, metadata, and any linked artifact or repository. Self-citations stay in the third person, and references are never dropped to preserve anonymity. Measurement papers leak identity through the deployment more than the byline: a "our production backbone" trace plus a topology detail and an author count can triangulate to one operator, so neutralize to scale-descriptive language.
# Mechanical pass on the exact upload candidate (adapt tool names as available)
pdftotext paper.pdf - | grep -niE '(acknowledg|funded by|grant no|@[a-z]+\.(edu|com))'
pdfinfo paper.pdf | grep -iE 'author|creator' # metadata identity leak
grep -rniE '(github\.com/[a-z0-9_-]+|/home/|/Users/)' *.tex figures/
# then a human pass: a non-author reads the PDF asking only "who wrote this?"
Mechanical passes catch strings; only a human catches implication.
Title and abstract must match the PDF verbatim — drift confuses bidding and reads as carelessness. Topic selections steer the paper toward a sub-community; choose for the reviewer profile you want, not maximal coverage. Enter the full author list at abstract time for conflict computation even though the PDF stays blind, and declare conflicts by the site's definitions — completely and no wider than defined.
[Clock] HotCRP site + AoE cutoffs confirmed from /deadlines
[Track] research / experience — matches the evidence
[Format] 12-page pre-reference / template / references / appendix — pass or blocker
[Blindness] mechanical findings + human-pass verdict
[HotCRP fields] title/abstract verbatim match, topics, conflicts — cleared or flagged
[Blockers] ordered, with owner and time cost