Erdős minimum overlap constant

Upper bound, last-listed row

| $0.380868$ | [YLTLYSTYLLGDHZSWZSHMELCZX2026] | SimpleTES |tracked since 2026-09-03 — no movement seen yet

Lower bound, last-listed row

| $0.379005$ | [W2022] |tracked since 2026-09-03 — no movement seen yet

Checked or tried against material for its cited source

• Another row, not one shown above: bound row $0.380924$ line 23 citing [GGSWT2025] — the claim we checked: that [GGSWT2025] gives the upper bound 0.380924 listed in this row, with the credit to AlphaEvolve stated in the row. What we read supports it: the paper states that AlphaEvolve improved the upper bound on the minimum overlap constant to 0.380924, and prints no longer form of that figure. source link (read in arXiv version 3 (22 Dec 2025), the current version of the unversioned link the entry gives, at Problem 6.5 on p. 16, from the page image; version 1's HTML carries the same sentence and figure). Selected: chosen because something already looked wrong.
• Another row, not one shown above: bound row $0.380876$ line 24 citing [YKLBMWKCZGS2026] — the claim we checked: that [YKLBMWKCZGS2026] gives the upper bound 0.380876 listed in this row, with the credit to TTT-Discover stated in the row. What we read supports it: the paper states that TTT-Discover, the method it introduces, improved the upper bound to 0.380876 with a 600-piece step function. source link (read in arXiv version 2 (5 Feb 2026), the version the entry's unversioned link resolves to, at Section 4.1.1, Table 2, Figure 2 and the Results paragraph on pp. 7–8, from page images; version 1 was not opened). Selected: chosen because something already looked wrong.
• reference entry citing [YKLBMWKCZGS2026] — the cited source could not be read at all (tried 2026-09-05). the source we could not read (on 2026-09-05 the URL this reference entry gave returned GitHub Pages' "Site not found" page (HTTP 200, about 9 KB, title "Site not found"), so a reader following the citation reached nothing; the paper itself is live at arXiv:2601.16175v2 and was read there (A-47-0015). Upstream replaced the link with arXiv:2601.16175 on 2026-09-26 (pull request #194, which we opened); that day the old URL was serving a PDF again (HTTP 200, application/pdf, 908,032 bytes), so the link was down and then up, not permanently dead). Selected: chosen because something already looked wrong.
• Another row, not one shown above: bound row $0.380871$ line 25 citing [T2026] — the claim we checked: that the cited reference gives the upper bound this row lists, 0.380871, credited to TogetherAI. What we read supports it: the repository, published by Together AI, states 0.380871 as its upper bound, its notebook's 0.380870310586220 rounded up, and credits the construction to Together AI's agents. source link (read in the cited repository at commit c388c6f of that repository (2026-04-12, the head of its main branch): the root README, the problem folder's README, the solution file's docstring and the committed outputs of its analysis notebook; no verifier or code was run). Selected: chosen because something already looked wrong.
• The row shown above: bound row $0.380868$ line 26 citing [YLTLYSTYLLGDHZSWZSHMELCZX2026] — the claim we checked: that [YLTLYSTYLLGDHZSWZSHMELCZX2026] gives the upper bound 0.380868 listed in this row, with the credit to SimpleTES stated in the row. What we read supports it: both editions give 0.380868 as SimpleTES's reported result in their results tables; version 1 also mentions 0.380856 from an ablation run, which it keeps out of its comparison table. source link (read in arXiv version 1 (21 Apr 2026), whose title is the one the entry quotes, at Table 10 and the text under it on p. 36 and the highlights on p. 1, from page images; version 2 (27 Jul 2026, retitled), which the entry's unversioned link now resolves to, was read at p. 14 and states the same figure). Selected: chosen because something already looked wrong.
• Another row, not one shown above: bound row $0.380926$ line 22 citing [H2016] — the claim we checked: that the cited 2016 reference gives the upper bound this row lists, 0.380926 (the row's comment is empty, so the value is the whole claim). What we read could not settle it: the abstract gives 0.380926... and the body 0.3809268534330870, so the row's 0.380926 is the paper's figure cut to six decimals, slightly below the bound the paper exhibits, and the paper does not settle whether that shortening counts as the listed upper bound. source link (read in arXiv version 1 (23 Sep 2016), the only version and the edition the entry cites, at the abstract and the final section on p. 2, from page images; the row carries the abstract's six decimals without its ellipsis, just below the value the paper exhibits). Selected: chosen because something already looked wrong.
• Another row, not one shown above: Another reading of this same row is listed above. bound row $0.380876$ line 24 citing [YKLBMWKCZGS2026] — the claim we checked: that the cited TTT-Discover paper states the upper bound 0.380876. What we read supports it. source link (read in the body: arXiv v2 (the version the entry's unversioned link resolves to) and v1, p. 8 (the Results paragraph, Table 2 and Figure 2) and the summary table on p. 1, p. 8 of v2 read from the page image). Selected: drawn by its position in the list of cited rows, not because it looked wrong.

5 bound row(s) here have been checked against material for their cited source (1 drawn by position, 4 chosen because something already looked wrong). Across this whole ledger 70 row(s) have been drawn by position, against 691 rows that name a source, counted on 2026-09-27: 64 were checked against material for their cited sources and for 6 the cited source could not be read at all. 5 more were chosen for suspicion and checked; they are counted apart, because a set selected for suspicion carries no rate. 1 reference entry here was also checked; those are not bound rows and are not counted against that figure. A row not listed here has NOT been checked.