sigmod-supplementary

SIGMOD Supplementary

SIGMOD's supplementary question is shaped by two venue facts: the body is

capped at 12 pages with unlimited pages only for references, and the

review cycle — including a possible revision — runs for months under

double-anonymity. Everything that supports the paper from outside the PDF

must therefore stay anonymous far longer than at one-shot conferences, and

whether appendices may ride inside the PDF varies by cycle: check the current

CFP and CMT form rather than assuming.

The three vehicles

| Vehicle | Carries | Multi-round hazard |

|---|---|---|

| In-PDF appendix (if the cycle allows) | Proofs, extra plots, tuning grids | Counting against the 12 pages by mistake |

| Anonymized extended report | Full proofs, complete evaluation | Hosting that expires or exposes authors mid-round |

| Anonymized code/data link | Implementation, workloads, scripts | Repo activity revealing identity during revision |

Decide per item, not wholesale. The default discipline: the 12 pages carry

every decision-critical argument; vehicles outside the PDF carry

verification depth, never the contribution itself.

Anonymous hosting that survives a round

• Use a purpose-made anonymous mirror (fresh account or an anonymization

service), not a scrubbed view of the real lab repository.

• Freeze it: pushes during the review window leak timezones, usernames in

commit metadata, and CI badges. Snapshot once, before the deadline.

• Test from a logged-out browser and a second network; institutional

single-sign-on redirects have unmasked authors before.

• Plan for the revision: if reviewers demand new experiments, you will need

a second frozen snapshot — name the first one v1 from the start.

• Remember late withdrawal counts as rejection; do not host anything you

would have to take down in a hurry mid-round.

What a database reviewer opens, in order

1. The PDF itself — if the 12 pages cannot stand alone, nothing else is

read charitably.

1. The extended report, when a proof or algorithm detail is contested;

mirror the paper's theorem and section numbering exactly so lookup takes

seconds.

1. The code link, most often to check whether a suspicious number has a

plausible harness behind it — a README naming which script produces

which figure answers this without a checkout.

1. Raw logs or datasets, rarely; include them for the ARI future, not for

review-time persuasion.

Extended-report etiquette

• Give the report the same title with "(Extended Version)" and keep it

anonymous during review — an arXiv posting under real names that the

paper links to is an anonymity breach, not a supplement.

• Do not fork content: the report is a superset of the paper, never a

different-numbers sibling. Divergent figures between the two versions is

a credibility wound reviewers do report.

• If the report is cited in the submission, cite it anonymously ("Full

proofs appear in the anonymous technical report [URL]").

Splitting a systems paper across vehicles

Worked split for a fictional adaptive-indexing engine submission: the body

keeps the architecture, the two headline comparisons, the correctness

argument sketch, and one ablation that carries the mechanism claim. The

extended report absorbs the remaining ablations, the full proof of the

maintenance-cost bound, and per-query latency tables. The anonymous

repository holds the engine, workload generators with seeds, and per-figure

run scripts. Each vehicle references the others by stable anonymous URL, and

nothing in the report contradicts a number in the body.

Body (12pp):   claim spine + headline evidence + mechanism ablation
Report:        proof detail + secondary ablations + full latency tables
Repo snapshot: engine v1 tag + generators + fig-to-script map

Size and format practicalities

• Keep any uploaded archive lean: source, scripts, configs, and small

sample data — not the 200 GB dataset, which belongs behind a fetch

script with checksums and a stated size.

• Prefer plain tar/zip over exotic formats; a reviewer who cannot open the

archive in ten seconds will not try twice.

• Put a one-screen INDEX file at the archive root mapping contents to paper

sections; reviewers grant minutes, not hours.

• Whatever CMT's current size ceiling is, verify it on the live form —

upload limits are among the least stable venue parameters.

Pre-upload sweep

• Grep the archive for names, employers, cluster hostnames, cloud account

IDs, and internal ticket references.

• Strip notebook execution metadata and PDF producer fields.
• Verify every URL in the PDF resolves to the anonymous mirror, not the

canonical repo you will publish at camera-ready.

Camera-ready conversion of the vehicles

Plan the vehicles' afterlife when creating them: at acceptance the extended

report becomes a named tech report or arXiv posting, the anonymous mirror

retires in favor of the canonical repository, and the paper's URLs all flip

(see sigmod-camera-ready). Vehicles built with this flip in mind — clean

history, no baked-in anonymous URLs inside scripts, version tags from day

one —

convert in an hour; vehicles built ad hoc convert over a painful week while

the PACMMOD metadata deadline looms.

Output format

[Vehicle map] item -> body / report / repo, with rationale
[Standalone test] does the 12-page body carry the claim spine alone
[Hosting audit] anonymity, freeze status, logged-out access verified
[Numbering mirror] report matches paper theorem/section numbers yes/no
[Sweep findings] identity artifacts found and removed
[Revision reserve] plan for a v2 snapshot if a revision verdict lands