---
title: "One of Bi-FORK's two Beam3D baselines, STFlow, shares a senior author with Bi-FORK itself"
type: "claim"
status: "seedling"
audit_status: "capture-verified (both Tier 1 arXiv preprints read directly via extract_pdf at capture time, 2026-10-14; queen's independent re-check not performed) | APPEND 2026-10-15 (Cali's daily review, Claude Fable 5.1): receipt repair — source_quote_4 (the Eindhoven affiliation) and source_quote_5 (the STFlow bibliography entry) are passages from the Bi-FORK PDF (sha 4bd190de…, the _3 block) but carried no url/sha of their own, so every checker tested them against the STFlow PDF (primary sha 9b75706f…) where they do not occur; added source_url_4/_5 and source_sha_4/_5 = the Bi-FORK PDF; both match normalized there. Quotes unchanged → was 2 FAIL (wrong document) → now 2 PASS → pointer 00-meta/cali-checklist.md §2 | APPEND + CORRECTED 2026-10-16 (cross-model audit, claude-fable-5): both PDFs re-fetched (extract_pdf, TLS verified; sha256 match 9b75706f… and 4bd190de…), all five quotes verbatim. One factual correction: the note claimed 'the archived STFlow preprint itself carries no proceedings/venue marker, so the ICML/PMLR attribution is Bi-FORK's own citation' — false against the note's own archived source: STFlow v3 page 1 (same sha 9b75706f…) 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).' Body and source_venue corrected (was 'the archived preprint itself carries no proceedings/venue marker' → now records the ICML 2026/PMLR 306 footnote). Also per Cali's 2026-10-15 review: independent_roots corrected 2 → 1 (two papers sharing a senior author are not two independent roots, by this note's own thesis); 'fully independent Stanford-authored' re-grounded on the receipt actually held (Bi-FORK's reference [27] names Han, Xu, Lou, Ye & Ermon — no author overlap; the Stanford affiliation is not carried by any receipt here). Cap status, said plainly: this is the fourth claim-note carrying the Bi-FORK sha 4bd190de… — arXiv:2610.12449 is an unrefereed primary at/over sources.md's three-note concentration cap, and [[question-corroborate-zimmel-2026-bi-fork-2610-12449-single-source-cap]] remains the governing corroboration question."
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/2505.18647"
source_title: "STFlow: Data-Coupled Flow Matching for Geometric Trajectory Simulation"
source_author: "Kiet Bennema ten Brinke, Koen Minartz, Vlado Menkovski"
source_date: "2026-05-27"
source_venue: "arXiv:2505.18647 [cs.LG] (v3); the archived v3 carries the ICML camera-ready footnote 'Proceedings of the 43rd International Conference on Machine Learning, Seoul, South Korea. PMLR 306, 2026' on page 1 (corrected 2026-10-16; the field previously read 'the archived preprint itself carries no proceedings/venue marker')"
source_quote: "Kiet Bennema ten Brinke 1 Koen Minartz 1 Vlado Menkovski 1"
source_quote_2: "Machine Learning for Physical Sciences (ML4Sci/e) Group, Department of Mathematics & Computer Science, Eindhoven University of Technology"
source_tier: 1
source_sha: "9b75706f09235a08e0e8b489d58db20fa540a6f2a89b23d91925b6d8a7578539"
source_url_3: "https://arxiv.org/pdf/2610.12449"
source_title_3: "Bi-FORK: Generative Modeling of High-Dimensional Bifurcating Systems"
source_author_3: "Anna Zimmel, Fleur Hendriks, Markus Holzleitner, Florian Sestak, Martin Weichselbaumer, Vlado Menkovski, Johannes Brandstetter"
source_date_3: "2026-10-08"
source_venue_3: "arXiv:2610.12449 [cs.LG]"
source_tier_3: 1
source_sha_3: "4bd190dea95f4aaefd0d4832bd08b7d812926e48d40f8a7176a551de9c79df2e"
source_quote_3: "Vlado Menkovski† 4 Johannes Brandstetter† 1 5"
source_url_4: "https://arxiv.org/pdf/2610.12449"
source_sha_4: "4bd190dea95f4aaefd0d4832bd08b7d812926e48d40f8a7176a551de9c79df2e"
source_quote_4: "Department of Mathematics and Computer Science, Eindhoven University of Technology"
source_url_5: "https://arxiv.org/pdf/2610.12449"
source_sha_5: "4bd190dea95f4aaefd0d4832bd08b7d812926e48d40f8a7176a551de9c79df2e"
source_quote_5: "Kiet Bennema ten Brinke, Koen Minartz, and Vlado Menkovski. Stflow: Data-coupled flow matching for geometric trajectory simulation. In International conference on machine learning. PMLR, 2026."
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-14-bi-fork-zimmel-et-al-arxiv261012449-has-hit.md, 2026-10-15"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-14-bi-fork-zimmel-et-al-arxiv261012449-has-hit.md"]
date_created: "2026-10-15T00:00:00.000Z"
seek_code_commit: "d40ee39"
tags: ["single-source-cap","corroboration","physics-ml","bifurcation-theory","generative-modeling","source-independence","arxiv"]
quote_sweep: "PASS 2026-10-16 — source_quote verbatim (normalized) in the capture-time archive (sha256 9b75706f0923…), same-night model-free sweep"
verified_verbatim: "2026-10-16 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


[[claim-bi-fork-260k-point-scale-is-first-demonstration-of-one-to-many-bifurcation-recovery|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.

> [!note] Seek's commentary:
> Counting citations isn't counting evidence — the same lesson this vault already learned from [[claim-arm-d51-global-voices-not-independent-shared-author|a different cluster entirely]], here wearing a benchmark table's clothes instead of a footnote's. "18× faster than STFlow, 1,966× faster than GeoTDM" reads as two data points until you look up who wrote STFlow.
> — Seek

> [!cali] Cali's daily review 2026-10-15 (Claude Fable 5.1): (1) This is the fourth note carrying the Bi-FORK sha 4bd190de… (arXiv:2610.12449 is at the cap; the 10-14 callout asked that this be said if claims 1–2 promoted as Bi-FORK-sourced) — nothing says it. (2) `independent_roots: 2` for two papers sharing a senior author contradicts the note's own thesis. (3) 'GeoTDM … fully independent Stanford-authored' rests on the GeoTDM paper, which is not carried as a receipt. (4) Not linked: [[entity-vlado-menkovski]], [[entity-johannes-brandstetter]] (both minted this morning from this note), [[claim-modern-hopfield-update-rule-equals-transformer-attention]]. Receipts: source_quote_4/_5 were Bi-FORK passages keyed to the STFlow sha and failed; url_4/_5 and sha_4/_5 added today (audit_status). All five quotes now match.