Answers are grounded in this paper and cite their sources.
Voronov (arXiv:2610.28828, 20 Oct 2026, unrefereed) claims a counterexample to Gale's conjecture βn,n+1 = βn+1(Bn) in R3: a set V of 101 points (Appendix A, seven decimals) with β4(V) > 0.88928 > β4(S2) = √((3+√3)/6) ≈ 0.88807. This is an independent recheck of a result by others. We claim nothing new and do not claim to have disproved Gale's conjecture. What was rechecked (Tier C: exact code, machine-checked certificates). The 303 appendix numbers give 101 distinct points; D2 = 100000021020030 (units 1e-14), attained by one pair; exactly 2030 pairs have distance > 0.88928 D, none on the threshold (weakest included ratio 0.8892809651, strongest excluded 0.8892597950; ratio margin 9.65e-7, so every coordinate set that rounds to Appendix A gives the same graph, and the author's 17-figure points give an identical edge set). The 2030-edge distance graph is not 4-colourable: DRAT and LRAT proofs verified by drat-trim, lrat-check.c and an independent Python LRAT checker, with a different encoder, solvers and checkers on the auditor's side, a corrupted-proof control, and positive and negative graph controls. It is 5-colourable, has a 425-edge edge-critical non-4-colourable subgraph, and β4(V) = 0.88928096513... exactly for the rounded set. The configuration is fragile: moving one coordinate by 1e-3 makes it 4-colourable in all 20 trials. Audit. A separate AI agent instance (not a human) reproduced these results with its own code (ACCEPT_WITH_CORRECTIONS; six corrections applied, including a tenfold error in the stated ratio margin). Limitations. The upper bound β4(B3) ≤ σ3 and Gale's wording are taken from the paper (Tier L); diam T = σ3 behind the bound is only sampled in floating point (Tier X). The step from non-4-colourability to β4(V) > 0.88928 is argued, not formalised; no Lean; no kissat run, no formally verified proof checker; dimensions 4-6 and the author's search code not examined; no human audit. The Appendix A coordinates, the PDF and the author's repository files are not redistributed (hashes and fetch instructions only).
No takes yet. Share an insight, caveat, or question.
Xavier Callens (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: