KDD Camera Ready

Use this after acceptance. KDD publishes through ACM, which means the final version is not "the PDF you upload" — it is the output of ACM's production pipeline (e-rights, then TAPS) applied to your source files. Papers that compile locally but violate acmart conventions stall in TAPS, so start this process the week the decision arrives, not the week the proceedings deadline arrives.

The page arithmetic (2026 anchor — reverify per cycle)

Accepted KDD 2026 papers all received the same proceedings budget: 12 pages total, of which 9 are content pages. Since submissions were 8 content pages, acceptance buys exactly one extra content page, while references and appendix together must fit in the remaining 3 pages.

• Spend the ninth page deliberately: reviewer-requested ablations, the scoping sentences promised in rebuttal, and de-anonymized acknowledgements all compete for it.
• The 3-page references+appendix cap is the trap: submissions with long optional appendices must cut appendix material, not just add a page. Decide early what migrates into the body, what moves to the public repository, and what dies.
• Keep section, table, and algorithm numbering aligned with the reviewed version where possible, so the AC's conditional-acceptance notes remain checkable.

ACM production pipeline

The standard ACM sequence KDD rides on:

1. e-rights form — the contact author receives the rights email; the paper's final title/author list must be frozen before completing it, because the form stamps the copyright block and DOI metadata.
2. TAPS upload — a zip of LaTeX source (or Word), named per instruction, which TAPS compiles into both the PDF and the HTML5 Digital Library version.
3. Proof review — check the generated proofs; HTML rendering surfaces errors the PDF hides.

# Typical TAPS package hygiene before zipping (KDD-specific naming: see the
# acceptance email; the zip name encodes proceedings acronym + paper id)
grep -rn "anonymous" main.tex sections/*.tex   # leftover anonymity flags
grep -rn "\\\\vspace\|\\\\baselineskip" *.tex  # spacing hacks TAPS rejects
grep -rn "includegraphics" *.tex | wc -l       # every figure file must ship in the zip
biber --tool --validate-datamodel refs.bib     # broken .bib entries stall TAPS builds

Switch the document class options from [sigconf,anonymous,review] to [sigconf], add the rights commands TAPS emails you (\setcopyright, conference metadata), and never hand-edit the copyright block.

De-anonymization sweep

Test the public repository logged out; a camera-ready pointing at a still-private repo is the most common post-publication complaint at ACM venues.

Money and access

• KDD proceedings live in the ACM Digital Library under the ACM Open model: authors from institutions not participating in ACM Open face an article-processing charge unless they qualify for a waiver; ACM announced a temporary 65% subsidy for 2026. Check your institution's ACM Open status before budgeting — this surprises industrial and non-Western labs most.
• Exact APC amounts, waiver mechanics, registration pricing, and presentation obligations were not verifiable at pack-build time (待核实): confirm all four against the acceptance email and the current conference pages.

Worked example: spending the ninth page

An accepted Research Track paper has three claims on the extra content page: the AC's conditional note asked for a scoped limitations paragraph; R3 requested the sensitivity analysis currently in Appendix C; and the authors want a new "extensions" subsection. Allocation logic:

1. The AC's conditional item is non-negotiable — it lands first (≈0.25 page).
2. The reviewer-requested sensitivity table migrates from appendix to body — this also relieves the 3-page refs+appendix cap, solving two constraints at once (≈0.5 page).
3. The new extensions subsection loses: content never reviewed should not enter the version of record beyond clarification. It belongs in the repository README or the next paper.

The heuristic: the ninth page pays review debts before it buys new real estate.

Proceedings metadata checklist

• Title and abstract in TAPS metadata match the PDF exactly — the DL indexes the metadata, not the PDF.
• CCS concepts and keywords chosen deliberately; they drive Digital Library search and recommendation for the paper's whole life.
• Author name forms consistent with each author's prior DL records (initials vs full names split citation counts).
• ORCIDs attached; funding acknowledgements in the format your grants require.
• The repository/dataset DOI cited in the final text is the archival one, not a branch URL that will rot.
• If the conference collects presentation or poster assets separately, that pipeline is distinct from TAPS — track both.

Timeline discipline

• Camera-ready deadlines for the current cycle were not verifiable at pack-build time (待核实) — take them from the acceptance email, and assume the window is short: dual-cycle venues compress production schedules.
• Do the TAPS dry run in week one: compile the de-anonymized source with the rights commands stubbed in, and fix acmart violations before the real upload window.
• Author-order changes after e-rights are painful to impossible; settle order disputes before anyone touches the form.

Output format

[Camera-ready status] on-track / blocked-at-TAPS / blocked-at-rights / not started
[Page budget] content: <n>/9, refs+appendix: <n>/3, total: <n>/12
[Pipeline stage] e-rights / TAPS upload / proofs / done
[De-anonymization] <items restored, repo publicly reachable yes-no>
[ACM Open status] covered / APC due (amount 待核实) / waiver requested
[Conditional-acceptance items] <AC-required change -> where it landed>