---
title: "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?"
type: "question"
status: "answered"
date_raised: "2026-10-10T00:00:00.000Z"
answered_log: [{"date":"2026-10-13T00:00:00.000Z","by":["[[claim-beltran-v1-2024-contains-verbatim-neuralrgbd-fabrication-ansari-traced]]","[[claim-beltran-v2-v3-2025-carry-same-fabricated-citation-unchanged-correction-not-yet-landed]]","[[claim-beltran-v4-2026-corrects-neuralrgbd-fabrication-to-real-azinovic-2022-citation]]"],"settled_by":"A direct read of all four arXiv versions of 2412.13176 (not just v1/v2) found that v1 genuinely contains the exact fabricated citation Ansari describes, converting that leg from inference to direct confirmation. But the premise built into the question's own framing — that v2 is the corrected version — is false: v2 and v3 both carry the identical fabrication unchanged. The correction only lands in v4 (2026-01-23), roughly thirteen months after v1 and about two weeks before Ansari's own paper. Net verdict: the trace is confirmed in substance (fabrication present, later corrected), but not on the v1-vs-v2 diff the question named — 'later version' means v4, not v2."}]
tags: ["contamination-inheritance","citation-genealogy","samar-ansari","primary-source-verification","copy-fingerprint-test"]
---


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]