One of Bi-FORK's two Beam3D baselines, STFlow, shares a senior author with Bi-FORK itself

single-source-capcorroborationphysics-mlbifurcation-theorygenerative-modelingsource-independencearxiv

Bi-FORK's Beam3D benchmark compares against two baselines, GeoTDM and STFlow. STFlow is Bennema ten Brinke, Minartz & Menkovski's "STFlow: Data-Coupled Flow Matching for Geometric Trajectory Simulation" (arXiv:2505.18647), whose author line reads "Kiet Bennema ten Brinke 1 Koen Minartz 1 Vlado Menkovski 1" under the "Machine Learning for Physical Sciences (ML4Sci/e) Group, Department of Mathematics & Computer Science, Eindhoven University of Technology." Bi-FORK's own author list includes "Vlado Menkovski† 4 Johannes Brandstetter† 1 5" (entity-vlado-menkovski, entity-johannes-brandstetter), affiliation 4 being the same department at the same university — the same person as a shared senior author on both the paper under evaluation and one of its two comparison baselines. GeoTDM, by contrast, shares no authors with Bi-FORK — its reference entry in Bi-FORK's own bibliography reads "Jiaqi Han, Minkai Xu, Aaron Lou, Haotian Ye, and Stefano Ermon. Geometric trajectory diffusion models. Advances in Neural Information Processing Systems, 37" — though the GeoTDM paper itself is not carried as a receipt here, so its affiliations are not independently recorded.

Bi-FORK's own bibliography describes STFlow's venue as "International conference on machine learning. PMLR, 2026," and the archived STFlow v3 preprint corroborates this itself: its first page carries the ICML camera-ready footnote "Proceedings of the 43rd International Conference on Machine Learning, Seoul, South Korea. PMLR 306, 2026. Copyright 2026 by the author(s)." (This sentence was corrected on 2026-10-16; the note originally claimed the archived preprint carried no proceedings or venue marker — see audit_status.) So STFlow is itself an ICML 2026-accepted paper, which bears on its weight as a baseline but not on the author overlap this note records.

This does not mean the Beam3D numbers are biased — STFlow is itself a method the Bi-FORK authors did not design, and labs reasonably benchmark against their own prior work — but "beats two independent baselines" overstates the independence of one of them. See question-corroborate-zimmel-2026-bi-fork-2610-12449-single-source-cap for the broader corroboration status of Bi-FORK's benchmark claims, which this finding does not resolve.