webconf-supplementary

Web Conference Supplementary Material

The Web Conference's 2026 research tracks had no separate supplementary upload.

Everything travels in one PDF: 8 pages of main paper, then references, then an

optional appendix, capped at 12 pages total. That design creates a zero-sum tail —

every reference line competes with appendix space — and a hard rhetorical rule:

reviewers are told they need not read past page 8, so the appendix can support

belief but never create it.

The three-container model

| Container | Budget | Reviewer contract | Right contents |

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

| Main paper | 8 pages | Read and judged | Claims, method, headline evidence, limitations |

| PDF tail | 12 − 8 − |refs| pages | Optional reading | Proofs, pseudo-code, protocol, extra tables |

| External artifact | unlimited | Reached only via link | Code, data manifests, full logs |

The tail is the scarcest resource most teams misbudget. A survey-flavored web paper

with 90 references can eat 2.5+ pages, leaving under 1.5 pages of appendix; a

theory-flavored one with 35 references gets nearly 4. Count your tail before

promising appendix sections to co-authors.

What earns tail space (and what does not)

Earns it:

• Proof details whose theorem statement and proof sketch already appear in the

main text.

• The full crawling/collection protocol backing a one-paragraph summary in Section 3.
• Hyperparameter and search-space tables; per-dataset statistics manifests.
• The pseudo-code for an algorithm described prosaically in the body.
• Secondary robustness tables that repeat a main-text conclusion, not extend it.

Does not earn it:

• Any result cited in the abstract, introduction, or conclusions. If a claim's only

evidence lives past page 8, the claim is unsupported in the judged document.

• The ethics or data-consent discussion for human-derived web data — reviewers

weighing harms must find this in the body.

• "Additional experiments" that introduce new claims — reviewers who do read the

appendix will treat unvetted claims as overreach; those who don't will never see

them. Either promote to the body or cut.

Compression order when the PDF is over 12

1. Move executable detail (configs, exact commands) to the artifact and leave a

pointer line.

1. Collapse per-dataset appendix tables into one table with grouped rows.
2. Trim redundant citations — cite the survey once instead of its 12 members where

the specific member isn't discussed.

1. Convert appendix prose to structured tables; appendix text is skimmed anyway.
2. Only then touch the main 8 pages, using webconf-writing-style compression

moves rather than spacing hacks — acmart tampering is detectable and treated

as a formatting violation.

Anonymized linking pattern

External artifacts referenced during review must not identify authors. The stable

pattern:

In the PDF (Section 3, footnote):
  "Code and data manifests: https://anonymous.4open.science/r/webgraph-2F51
   (anonymized for review; will be archived with a DOI upon acceptance)."

Checks before submission:
  - repository has no git history, no institutional paths, no author usernames
  - README written in the same anonymous voice as the paper
  - dataset files carry the crawl manifest, not internal project codenames
  - link actually opens in a logged-out browser session

At camera-ready the anonymous mirror is replaced by the DOI-bearing archival

deposit (see webconf-artifact-evaluation for the deposit and badge flow).

Worked budget: one paper, two viable splits

A graph-mining submission has 8.0 main pages locked and must choose its tail. Two

honest allocations of the 4-page tail, for different paper shapes:

| Tail plan | References | Appendix | Fits which paper |

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

| Method-heavy | 1.5 pp (42 refs) | 2.5 pp: proof details, pseudo-code, hyperparameter grid | Theory/algorithm claims needing verification depth |

| Measurement-heavy | 2.2 pp (68 refs) | 1.8 pp: collection protocol, per-platform stats, bot-filter rules | Empirical claims needing literature coverage |

What does not exist is the fantasy allocation (2.2 pp of references and 2.5 pp

of appendix). Teams discover this on deadline night when BibTeX renders; discover

it two weeks earlier by compiling with the real bibliography and an appendix

skeleton, then assigning the tail like any other scarce budget — explicitly, in

writing, before co-authors draft appendix sections that will not fit.

Cross-track variation

The single-PDF-with-appendix rule above is the research tracks rule for 2026.

Short papers ran at 4 pages including references — effectively no tail at all —

and companion-proceedings calls (demos, PhD symposium, workshops) each set their

own limits. Never transplant this page arithmetic across calls; reopen the specific

call's page before formatting.

Note also the asymmetry at camera-ready: the proceedings budget re-labels the

zones (12 pages, 8 content) but does not enlarge the content allowance, so an

appendix section promoted during revision must evict something — plan the

appendix as if its contents were permanent residents, not staging.

Pre-flight audit

• [ ] Total ≤ 12; main ≤ 8; nothing claim-bearing after page 8.
• [ ] Every appendix section is pointed to from the body ("full protocol in

App. B") — orphan appendices signal padding.

• [ ] Every body forward-reference resolves ("App. C" exists and matches).
• [ ] Ethics/consent discussion in the body, not the tail.
• [ ] Anonymous artifact link passes the logged-out test.

Output format

[Tail budget] refs=<pages> appendix=<pages> total=<n>/12
[Misplaced content] <claims past p.8 / ethics in tail / orphan appendices>
[Compression applied] <moves 1-5 used>
[Artifact link] anonymized + reachable: yes/no
[Cross-references] body->appendix and appendix->body both resolve: yes/no