Retractions

Published retractions

No published result has been retracted. If one is, it will be listed here with what was claimed, what was found, the date, and what still stands.

Caught before publication

Six claims that reached internal review or rating were dropped, refuted, or had their priority withdrawn before publication. None is counted among the results on the home page.

  • The 3k+13k + 1 greedy that never stops (Erdős Problem #282) — Priority withdrawn, 2026-11-20. A proved theorem showing the greedy Egyptian-fraction expansion with denominators 1(mod3)1 \pmod 3 never stops on 3/173/17 or 1/771/77 was dropped after prior-work review found an open Lean pull request (formal-conjectures #7716, opened 2026-11-01) proving the same negative answer on 1/1191/119 by the same mechanism. The theorem itself still holds.
  • The general form of the base-6 Lychrel argument — Priority withdrawn, 2026-11-14. A general-base pattern theorem and a corollary for bases B=2kB = 2^k (k≥2k \ge 2) were withdrawn the day they were written after a public repository (taco-jpg/p196-algorithm, committed 2026-10-18 and 2026-10-22; family dating to Seal 1996) was found to state both. The two base-6 families on Two base-6 Lychrel families are unaffected.
  • The "below k2k^2" law for taking kk pins — Refuted, 2026-11-20. The claim that the only losing lengths below k2k^2 in the game taking kk adjacent pins are 0,…,k−10,\allowbreak \dots,\allowbreak k - 1 and the twelve linear families fails for every k≥8255k \ge 8255, where 8255k−24418255k - 2441 lies below k2k^2. The theorem on Flammenkamp's twelve losing lengths is unaffected.
  • Halve or absorb: doubling law for coin piles — Refuted, 2026-11-18. The claim L(2n)=L(n)L(2n) = L(n) for the number of cycles on nn coins held through n≤64n \le 64 (and under an unproved assumption to n=980n = 980), but an exhaustive census of all 342,325,709 arrangements at n=105n = 105 gave L(105)=8L(105) = 8 while n=210n = 210 has a ninth cycle of 58 moves (L(210)≥9L(210) \ge 9; likewise L(115)=5L(115) = 5 and L(230)≥6L(230) \ge 6).
  • The ends game — No longer holds up, 2026-11-18. A candidate pairing an infinitely-many-loops theorem (cycles of length 3940+3560k3940 + 3560k) with an open no-growth conjecture had its confirmation withdrawn because the growth claim stays unproved. Nothing in it was refuted.
  • Few lines through every point: the 12j+112j + 1 row — Sent back before review, 2026-11-20. The claim that for every j≥2j \ge 2 there are 12j+112j + 1 non-collinear points with every point on at most 6j−26j - 2 connecting lines (checked for j=2j = 2 to 10) had been counted without a direct falsifier pass on the row itself and was returned to pending. The proved result on Few lines through every point is unaffected.