T-240: Couzo's 268 and 303 of #763, below T-174 and T-239, from Kaggle notebook packings

V3 C3 Upper bound confirmed

2026-10-23 published · Couzo after Lavrentev, Tymchishin, Muglich, Polkovskyi, Daniel, Fang, de Winter, Levy · n=268,303

Two complete rational source certificates of issue #763 report finite upper-bound improvements at the exact sides they state: s(268)≤78645658458105249171306580101/4660659750118000000000000000 = 16.87436171587383, 4.45e-3 below T-174, and s(303)≤55035101387261868285958438488216555614454417144023299/3073388523105741896887024831655729584500000000000000 = 17.90697823379890, 2.78e-3 below T-239, the case's house best packings when they were reported; both decimals are the exact sides rounded at 14 places.

Both certificates and their duplicate-square and outside-container controls were decided here on 24 October 2026 by both maintained exact routes. Those routes share components with the producer: the source's manifest names this repository's sqpack/verify.py, through which one route decides, as its first checker, and both routes read the certificates through the evand-cert adapter, transcribed from Daniel's checkers, from which the source's verify_bound_certificate.py derives (RF-4). The 24 October review decided both certificates again on a clean-room checker of its own, on two exact routes with ten controls each, by a Claude agent, a system family the producer also used, and found no blocking or major defect, so the bounds are confirmed, independently re-implemented.

The issue states that both maintained exact geometry routes (the corner replay and the projection check) accept each finite source witness, and the source's manifest records the author's checker receipts with sqpack/verify.py and the source's exact checker verify_bound_certificate.py (Daniel's MIT verifier, with modifications by Couzo with Codex, as its header says), pinned here by digest and not run. The source's commit the issue names and its two parents each end with a Co-Authored-By trailer naming Claude, and the earlier parent's message describes "Codex's reviewed intake" (RF-1).

The source's README selects its claims: each count credited to Couzo whose exact side, rounded up at 12 places, is strictly below the reviewed comparison's 12-place print where that comparison is Couzo's own, and more than 1e-10 below it where it is another author's. Those statements are the author's; no source program ran here.

Credit Couzo (https://github.com/franciscouzo/square-packing), after the packings his searches refined, as the source's manifest names them: at 268 Tymchishin's packing from his Kaggle notebook "CPU/GPU/MPS EDA+Seam LP" v4, the parent the manifest retains, and at 303 Couzo's own Kaggle push refinement of Lavrentev's packing from his Kaggle notebook "CUDA Packing Search" v11. Measured pose by pose by the review, each refines its parent, with no square beyond a tenth of it at 268 and three at 303, and each is another arrangement than the houses it replaces. The issue says the notebooks' own source notices carry their earlier construction credits; the source does not publish them, and the author is asked for them.

Significance, composition and next rung
Significance
New arrangements at 268 and 303 relative to the record, 4.45e-03 and 2.78e-03 below the houses T-174 and T-239, one to one with neither under any symmetry, with 118 and 133 squares beyond a tenth of them, as T-162's 2.02e-03 to 1.42e-02 (S3) are, and far above T-230's 1.86e-04 (S3). Each refines the Kaggle notebook packing its search started from, with none beyond a tenth of Tymchishin's v4 at 268 and three of Couzo's push packing at 303, so the arrangements come from the notebooks and the sides from the refinements, as the credit records. No lower bound or optimum.
Next rung
None for the bound. V4 and C4 need a second accepted adversarial review by a distinct reviewer and a retained human oversight record; one AI review, of 24 October, is retained. The review's six questions for the author (the credit at 268, whether #763 claims both counts, the 303 lineage and its unpublished steps, and the notebooks' source notices) hold no rung. At 268 Daniel's later certificate, 4.48e-12 below this side (T-242), replaced it as the house on 26 October 2026; it remains the house at 303.
Regularized drawings
n = 303: a regularization of T-240’s packing by Levy (regularize_grid_slack/1: Daniel's coincidence placement (evand/square-packing search/regularize reg.coincidences and reg.place, MIT), ported to start from the witness's exact frame and to keep its contacts, then the same-side exact re-solve, through devtools.regularize_grid_slack; the project’s modifications: port of his coincidence placement (regularize_grid_slack/1) rather than his code, same-side exact re-solve (devtools.regularize_same_side); at its own side; identity: measured here; nearest centres one to one; 28 squares moved, at most 0.0978646; turns up to 0.033049°).
Novelty
previously-published Present in an identified source

The cases

Case record
Regularized by the Squares Project with Daniel's regularizer (evand/square-packing, MIT)
As published
The packing as published, in its source’s pose, Vladislav Lavrentev 2026 [Daniel Kaggle descents 2026-10-24]

n=268

16.3816.874
161718
16.37117.371
nn+1

Proven

16.38238≤s(268)≤16.874362

  • exact

Citation record n-268

lowerKarakuş 2026, arXiv corrects Nagamochi 2005 (confirmed T-083)

upperLavrentev & Daniel, GitHub (confirmed T-242)

Open

  • optimality

The case record

LowerUpper
Gap0.49198154…
Case record
Regularized by the Squares Project with Daniel's regularizer (evand/square-packing, MIT)
As published
The packing as published, in its source’s pose, Vladislav Lavrentev 2026 [Couzo issue 763 2026-10-23]

n=303

17.4117.907
171819
17.40718.407
nn+1

Proven

17.41892≤s(303)≤17.906979

  • exact

Citation record n-303

lowerKarakuş 2026, arXiv corrects Nagamochi 2005 (confirmed T-083)

upperLavrentev & Couzo, GitHub (confirmed T-240)

Open

  • optimality

The case record

LowerUpper
Gap0.48805391…

Results on these cases

19 results in the register on n=268,303, oldest first, each with what it established and how it stands now.

  1. 2005 published T-007 · n=268,303

    s(n)≥min(⌈n⌉,n−2⌊n⌋+1+1) for 4≤n≤324

    V0 C1 Lower bound incomplete on these cases, superseded by T-083 and T-180 (reported)

    Nagamochi · Nagamochi 2005 · source · register

  2. 2026-09-04 published T-085 · n=268,303

    Nagamochi 2005, Lemma 1 is false for every container with a>3 and b>2

    V3 C3 Correction confirmed

    Karakuş; Vlasenko · Karakuş 2026 · chelokot Nagamochi counterexample 2026 · packet · register

  3. 2026-09-27 published T-056 · n=268,303

    Smaller packings for 49 counts from n=68 to 307, each certified two independent ways

    V3 C3 Upper bound confirmed superseded by T-240 and T-242

    Couzo · franciscouzo square-packing 2026-09-27 · packet · register

  4. 2026-09-29 published T-083 · n=268,303

    s(n)≥1/2+n−⌊n⌋+1/4 for every nonsquare 8≤n≤324

    V3 C3 Lower bound confirmed

    Karakuş · Karakuş 2026 · source · register

  5. 2026-10-03 published T-092 · n=303

    Smaller packings again at seven counts from n=208 to 306, each certified two independent ways

    V3 C3 Upper bound confirmed superseded by T-240

    Couzo · franciscouzo square-packing 2026-10-03 · packet · register

  6. 2026-10-05 published T-098 · n=268,303

    Exact optima of 48 known-best packings: s(n) ≤ S', 3.5e-13 to 5.0e-11 below each printed side

    V3 C3 Upper bound confirmed on these cases, superseded by T-240 and T-242

    Daniel after Couzo, de Winter, Ellsworth, Levy · evand exact optima 2026-10-05 · packet · register

  7. 2026-10-07 published T-113 · n=303

    s(n)≤Sn at ten counts, n=108 to 303, from new SQUISH rational packings

    V3 C3 Upper bound confirmed superseded by T-240

    Chaoweeraprasit after Ellsworth · SQUISH ten packings 2026-10-07 · packet · register

  8. 2026-10-09 published T-134 · n=303

    Six exact rational certificates from Couzo

    V3 C3 Upper bound confirmed on these cases, superseded by T-240

    Couzo after Polkovskyi, Chaoweeraprasit, Gupta, Xu, Daniel, Ellsworth, Levy · Couzo exact certificates 2026-10-09 · packet · register

  9. 2026-10-09 published T-136 · n=303

    Seventeen exact rational packings from SQUISH's third request

    V3 C3 Upper bound confirmed superseded by T-240

    Chaoweeraprasit after Couzo, Xu, Göbel, Schadt, Ellsworth, Hajba, Levy · SQUISH third request 2026-10-09 · packet · register

  10. 2026-10-09 published T-138 · n=303

    Exact witnesses at Mishapolk's printed ceilings, the smallest known at 103 and 258 when registered

    V3 C3 Upper bound confirmed superseded by T-240

    Polkovskyi after Stenlund, Friedman, Ellsworth, Chaoweeraprasit, Couzo, Xu, Gupta, Daniel, Levy · Mishapolk decimal poses 2026-10-09 · packet · register

  11. 2026-10-10 published T-146 · n=303

    Exact rational certificates at fifteen counts from Daniel's regularized record lists

    V3 C3 Upper bound confirmed on these cases, superseded by T-240

    Daniel after Xu, Chaoweeraprasit, Polkovskyi, Levy · Daniel regularized lists 2026-10-10 · packet · register

  12. 2026-10-12 published T-162 · n=303

    New packings of 129, 204 and 303 squares, a cut-out and two compositions, each certified exactly

    V3 C3 Upper bound confirmed superseded by T-240

    Ostashev after Nakajima, Chaoweeraprasit, Daniel, Xu, Levy · Ostashev compositions 2026-10-12 · packet · register

  13. 2026-10-13 published T-174 · n=268,303

    Tighter certificates at 68 counts from PingYou Ltd, of others' packings and its own

    V3 C3 Upper bound confirmed on these cases, superseded by T-240 and T-242

    PingYou Ltd after Friedman, Ellsworth, Daniel, Xu, Barnett, Couzo, Ostashev, Chaoweeraprasit, Gupta, Fang, Polkovskyi, Levy · PingYou PR 574 certificates 2026-10-13 · packet · register

  14. 2026-10-13 published T-180 · n=268,303

    Closed-form lower bound s(N)≥a(N) for every non-square N≥49, reported

    V0 C1 Lower bound reviewed

    Hosono · wand125 closed form 2026-10-14 · packet · packet · register

  15. 2026-10-14 published T-186 · n=303

    Exact certificates at twelve counts from Polkovskyi, tightening the smallest reported there

    V3 C3 Upper bound confirmed on these cases, superseded by T-240

    Polkovskyi after Ostashev, Couzo, PingYou Ltd, Levy · Mishapolk KKT refinements 2026-10-14 · packet · register

  16. 2026-10-22 published T-230 · n=303

    Daniel's smaller n303 packing, started from Ostashev's packing, certified exactly

    V3 C3 Upper bound confirmed superseded by T-240

    Daniel after Ostashev, Levy · Daniel n303 2026-10-22 · packet · register

  17. 2026-10-23 published T-239 · n=303

    Kristiansen's smaller n303 packing, found from Daniel's packing, certified exactly

    V3 C3 Upper bound confirmed superseded by T-240

    Kristiansen after Daniel, Ostashev, Levy · Kristiansen n303 2026-10-23 · packet · register

  18. 2026-10-23 published T-240 this result · n=268,303

    Couzo's 268 and 303 of #763, below T-174 and T-239, from Kaggle notebook packings

    V3 C3 Upper bound confirmed

    Couzo after Lavrentev, Tymchishin, Muglich, Polkovskyi, Daniel, Fang, de Winter, Levy · Couzo issue 763 2026-10-23 · packet · register

  19. 2026-10-24 published T-242 · n=268

    Daniel's exact descents of public Kaggle packings at 268 and 304, below the reports before them

    V3 C3 Upper bound confirmed

    Daniel after Lavrentev, Tymchishin · Daniel Kaggle descents 2026-10-24 · packet · register