sigir-camera-ready

SIGIR Camera Ready

SIGIR publishes through ACM: the version of record is produced by ACM's pipeline

(e-rights, then TAPS) from your source files, and it appears in the Digital

Library as both PDF and HTML (SIGIR 2025's volume is DOI 10.1145/3726302). The 2026

camera-ready window sat in late April for the tracks whose dates were verifiable

(April 29, 2026 for Resources / Low Resource Environments); take your own track's

date from the acceptance email, and note the July conference leaves little slack.

What changes between the reviewed PDF and the record

| Item | Reviewed submission | Version of record |

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

| Document class | sigconf + anonymous/review options | sigconf with rights commands from TAPS |

| Authors | Hidden | Restored; order frozen before e-rights |

| System name | Possibly a review alias | Real name, consistent with the public repo |

| Dataset/log provenance | Genericized for anonymity | As precise as data agreements allow |

| Repository link | Anonymous mirror | Public, licensed, stable (archival DOI best) |

| Acknowledgments | Absent | Restored — and they cost body pages at SIGIR |

| CCS concepts + keywords | Often placeholders | Final; they drive DL search forever |

| Leaderboard references | Scrubbed | Restorable — align run IDs with the paper |

Two SIGIR-specific pressure points hide in that table:

• Acknowledgments and restored details consume the page budget, because SIGIR

counts everything except references inside the limit. If the camera-ready allowance

for your track is the same as submission (whether acceptance buys an extra page was

not verifiable for 2026 — 待核实), something you wrote must shrink to make room.

• The anonymity aliases must be unwound consistently. A paper that called its

system "ANON-RANK" in review must not leave one stale alias in a figure caption or

a run-file name cited in a table. Grep, don't skim.

The ACM production sequence

1. e-rights: the contact author receives ACM's rights form; completing it stamps

the copyright block, so title and author list must be final first.

1. TAPS upload: zip your LaTeX (or Word) source per the instructions in the

acceptance email. TAPS compiles the DL's PDF and HTML5 — HTML is where homemade

macros, \vspace hacks, and missing figure files fail loudly.

1. Proofs: check both renderings. Tables that were hand-tuned for two-column PDF

are the usual HTML casualties, and IR papers are table-heavy.

# Sweep before zipping for TAPS
grep -rn "ANON\|anonym" *.tex sections/          # stale review aliases
grep -rn "4open\|anonymous.4open" *.tex          # anonymous mirror still linked?
grep -rn "vspace\|renewcommand{\\\\baselinestretch}" *.tex  # spacing hacks
ls figures/ | wc -l && grep -c includegraphics *.tex        # figure inventory match

Digital Library metadata that outlives the conference

• CCS concepts: pick from the ACM taxonomy deliberately (e.g., Information

systems → Retrieval models and ranking / Evaluation of retrieval results). They are

the DL's routing layer for your paper's entire life.

• Keywords: include the terms IR searchers actually use — task name, collection

names, method family ("dense retrieval," "learning to rank," "conversational

search"). Your future citers search the DL and Google Scholar with these.

• Author name forms: match each author's existing DL record; inconsistent forms

split citation identities.

• ORCID and funding: attach ORCIDs; format grants the way the funder mandates.
• The abstract in TAPS metadata must equal the PDF abstract — the DL indexes metadata.

Artifacts at publication time

The camera-ready is the moment review-time artifacts become public commitments:

• Flip the repository public before the camera-ready deadline and test it from a

logged-out browser; the paper will cite it for decades.

• Freeze the exact run files behind every table (runs/ with a manifest); IR

replication requests are usually "can you send the run file," and yes should be easy.

• If the paper introduced judgments, topics, or a corpus and it was not routed

through the Resources track, publish it with a license and a datasheet anyway —

the community treats missing licenses as unusable resources.

Worked example: absorbing review debts into the budget

An accepted full paper carries two commitments from the review phase: a promised

scope-limitation paragraph and a clarification of the tuning protocol. Camera-ready

also restores acknowledgments (~0.15 pp) and real provenance details (~0.1 pp).

With no verified extra page, the arithmetic forces trades:

1. Review commitments land first — they are quasi-contractual and reviewers of

record may check.

1. The restored acknowledgments displace the weakest analysis sentence-by-sentence,

not table-by-table: prose compresses, protocol does not.

1. Nothing new enters: results and claims not seen by reviewers do not belong in

the version of record; they belong in the repository changelog or the next paper.

The heuristic: camera-ready pays debts and restores identity; it does not buy new

real estate.

Registration and presentation

Author registration by the camera-ready date was confirmed for the Resources track in

2026 (April 29); whether every accepted full/short paper requires a registered,

in-person presenting author — and the fee schedule — was not verifiable at build time

(待核实). Read the acceptance email as the controlling document, and budget for

Melbourne travel early: July is conference high season in the IR calendar.

Timeline discipline

• Week 0 (decision): settle author order; start e-rights; unwind aliases in source.
• Week 1: TAPS dry run with rights commands stubbed; fix class-file violations.
• Week 2: metadata pass (CCS, keywords, ORCID); repository public; proofs checked.
• Slack: TAPS support queues lengthen as thousands of ACM camera-readies converge —

finishing a week early is the only reliable mitigation.

Output format

[Camera-ready status] on-track / blocked-at-rights / blocked-at-TAPS / not started
[Page budget] <pages>/<track allowance> (allowance verified from acceptance email? y/n)
[De-anonymization] aliases unwound y/n, repo public y/n, provenance restored y/n
[Metadata] CCS final / keywords final / ORCIDs attached / abstract == PDF
[Artifacts] run files frozen / license present / archival DOI minted
[Registration] requirement confirmed from acceptance email? y/n (fees 待核实)