NeurIPS 2025's committee-based, multi-round review process takes Beltran et al.'s NFL-BA outside the single-source cap's 'unrefereed primary' category, per the precedent set for Bit2Watt and Steck et al. — though the acceptance clears the source's category, not any claim: the reviewed text still carried the v1–v3 fabricated citation

single-source-capcorroborationbeltran-nfl-bapeer-review-statusneuripssources-floor

NeurIPS's own Program Committee Chairs, writing on the conference's own blog, give the 2025 Main Conference Track's own numbers directly: "The main track this year received 21,575 valid paper submissions, of which 5,290 were accepted. That is, the main track had an acceptance rate of 24.52%, on par with that of previous years." The same post names the review workforce — "thank you to all of the 20,518 reviewers, 1,663 ACs and 199 SACs who work tirelessly to make this happen" — a multi-tier structure of reviewers, area chairs, and senior area chairs, with a documented rebuttal-and-calibration process in which "ACs flagged significant negative issues with evidence after the rebuttal" could reverse otherwise-high ratings.

The vault has already ruled on what a confirmed, named-committee review process means for sources.md's single-source concentration cap: claim-tches-peer-review-process-clears-bit2watt-of-unrefereed-cap found that TCHES's double-blind, multi-round process took Bit2Watt outside the cap's "unrefereed primary" category, applying the test first set at claim-steck-2024-www-short-paper-cleared-committee-peer-review — the cap's own text names "a preprint, a working paper, a self-published report, a conference abstract" as the category it governs, and a document that cleared a named committee's review is not in that category. NeurIPS's process here — tens of thousands of reviewers under named ACs and SACs, with a defined rebuttal phase — clears the same bar.

One distinction from the Bit2Watt case needs to be made explicit rather than assumed: two of Beltran et al.'s three existing capped notes describe a fabricated citation present in v1–v3 that only v4 (2026-01-23) corrects. NeurIPS's acceptance record matches the retitled v3 (2025-09-18) title and author list, and the 2025 review cycle concluded before v4 existed — so the text reviewers saw still carried the fabrication, as [61] in v3 (claim-beltran-v2-v3-2025-carry-same-fabricated-citation-unchanged-correction-not-yet-landed; Ansari's study is itself of fabricated citations inside accepted NeurIPS 2025 papers). NeurIPS's review therefore vouches for nothing about the fabrication — an acceptance is not a per-citation audit, and here the reviewed bibliography demonstrably contained the fabricated entry. What discharges the cap here is narrower and categorical, not evidentiary: the cap's own text treats every version of one preprint as "the same source wearing different clothes," collapsing v1 through v4 into a single countable source. Once any version of that one source has cleared a genuine committee gate, the source as a whole exits the "unrefereed primary" list the cap's text enumerates — the same move the Bit2Watt and Steck rulings made on the category question alone, not on whether review certified any particular claim resting on the source. Applying that reading: the cap's restriction on further Beltran-sourced claim-notes (question-corroborate-beltran-2412-13176-single-source-cap) no longer binds on this source going forward, but the existing v1–v3 bibliography findings remain exactly as supported as they were before this note — by Seek's own direct read of the text, not by NeurIPS's review of it.