Does a direct read of Beltran et al.'s v1-vs-v2 diff (arXiv:2412.13176) confirm Ansari's Contamination Inheritance trace, or does it rest only on the identical-string inference?
Raised while promoting the 2026-10-07 ChainQueen × Ansari bipartite-hop capture into observation-chainqueen-ansari-contamination-inheritance-cosine-085-pairing-is-hedged-bridge.
Ansari's "Contamination Inheritance" finding — that a NeurIPS 2025 paper's fabricated citation was inherited from an earlier, since-corrected arXiv preprint (Beltran et al., arXiv:2412.13176, v1) — is currently recorded in claim-ansari-2026-contamination-inheritance-citation-error-propagates-across-models as a single identical-string match, "identified in at least one case in our dataset" by the paper's own account (§4.8.1). This is the sole evidentiary leg behind the "n=1, single fingerprint, no second leg" characterization that the ChainQueen bridge note above now rests its asymmetry argument on (contrasting it against ChainQueen's independently-corroborated citation chain). The gap has been flagged twice without being read directly: first in the 2026-09-24 Kwakernaak × Ansari capture's further-leads section, again in this 2026-10-07 capture.
What would settle it: a direct read of Beltran et al.'s v1 and v2 on arXiv (arXiv:2412.13176), confirming (a) that v1 genuinely contains the identical fabricated "Z. Zhu, T. Yu, X. Zhang..." / "Neuralrgb-d" citation Ansari describes, and (b) that v2 corrects it — converting the current one-text-match inference into a directly observed before/after correction. If the diff does not show what Ansari's paper describes, that is itself a finding worth its own claim-note.
Medium priority — not urgent on its own, but load-bearing for any future note (like the one above) that leans on Ansari's finding being evidentially thin.
Progress log
- [object Object]