You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hiroaki Hosono's closed form (wand125, 13 October 2026, T-180): s(N) >= a(N) for every non-square N >= 49, from a measure of line segments on 34 intervals of the fractional part. At n = 268, k = 16 and interval 12 give a(268) = 16.429141319661638627012326214822..., here rounded down at 12 and 30 places. The source replayed its certificate with its own checker; no replay has run here.
Unrestricted unit-square packing with independent rotations and disjoint interiors.
E-wand125-closed-form-report
value
exact_form
evidence
16.3823801742
1/2 + sqrt(268 - floor(sqrt(268)) + 1/4)
E-karakus-strip-lower
E-karakus-strip-measure-interval
property
assurance
method
scope
certificate
replay
evidence
not-rigid
numerically-checked
numerical-multiprecision
Square 0 of the retained witness (witness id 1) translates 0.00011 along (0, -1) with the packing still valid, so the configuration admits a non-trivial feasible motion; 233 of its 268 squares do. Every constraint is exactly affine in the slide parameter, so the arithmetic carries no linearization error, but the coordinates are the witness's own finite-precision transcription: this settles the retained configuration, not the true optimum. Rigidity and optimality are independent, and this bears only on the former.
atlas/known-best/translation-escape-screen.json
uv run --frozen --all-extras --group dev python -m devtools.screen_translation_escape --check
E-translation-escape-not-rigid
claim
claimed_by
published
year
s(268) <= 16.87895510935973675, Griffin Casson's packing, larger than the reported side
Earlier house 16.878814821006764724747121053018 from [evand exact optima 2026-10-05], the case’s ceiling until 17 October 2026; its certificate and replay remain retained.
Francisco Couzo
Evan Daniel
2026-10-05
2026
claim
claimed_by
published
year
Earlier house 16.8788148210067647246 from [PingYou PR 574 certificates 2026-10-13], the case’s ceiling from 17 October 2026 until 25 October 2026; its certificate and replay remain retained.
Francisco Couzo
Evan Daniel
PingYou Ltd
2026-10-13
2026
claim
claimed_by
published
year
Earlier house 16.8743617158738255 from [Couzo issue 763 2026-10-23], the case’s ceiling from 25 October 2026 until 26 October 2026; its certificate and replay remain retained.
Kirill Tymchishin
Francisco Couzo
2026-10-23
2026
claim
claimed_by
published
year
Earlier house 16.8743617158693459 from [Daniel Kaggle descents 2026-10-24], the case’s ceiling from 26 October 2026 until 5 November 2026; its certificate and replay remain retained.
Vladislav Lavrentev
Evan Daniel
2026-10-24
2026
claim
claimed_by
published
year
Earlier house 16.8724995181350307 from [Daniel issue 914 2026-11-05], the case’s ceiling from 5 November 2026 until 9 November 2026; its certificate and replay remain retained.
Evan Daniel
2026-11-05
2026
claim
claimed_by
published
year
Earlier house 16.8724698978362125 from [Özden s268 upper bound 2026-11-09], the case’s ceiling from 9 November 2026 until 11 November 2026; its certificate and replay remain retained.
Evan Daniel
Doğa Özden
2026-11-09
2026
claim
claimed_by
published
year
Earlier house 16.8720298490844502 from [Özden s268 upper bound 2026-11-10], the case’s ceiling from 11 November 2026 until 18 November 2026; its certificate and replay remain retained.
Regularized by the Squares Project with Evan Daniel's regularizer (evand/square-packing, MIT), with the project's modifications: port of his coincidence placement (regularize_grid_slack/2) rather than his code, same-side exact re-solve (devtools.regularize_same_side). A drawing of T-320's packing at the witness's own side: a pure regularization, which proves no new bound, so not a result.
by
credit
generator
modifications
side
of
view
project
Regularized by the Squares Project with Evan Daniel's regularizer (evand/square-packing, MIT)
regularize_grid_slack/2
port of his coincidence placement (regularize_grid_slack/2) rather than his code
Hiroaki Hosono’s closed form, 14 October 2026. Hiroaki Hosono’s closed form
(wand125, 13 October 2026, T-180) states $s(N) \ge a(N)$ for every non-square
$N \ge 49$, from a measure of line segments on 34 intervals of the fractional part.
At $n = 268$, $k = 16$ and interval 12 give
$a(268) = 16.429141319661638627012326214822\ldots$, so the reported lower bound here is
now $s(268) \ge 16.429141319661$, rounded down, above Nagamochi’s closed form,
$16.39480431834$. The source replayed its certificate with its own checker; no record of
it has been replayed here, so the verified lower bound is unchanged.
The source states that its search, certificates, checkers and text were produced with AI
assistance (Claude and Codex), under the direction of and reviewed by Hiroaki Hosono.
House adoption, 18 November 2026. Mykhailo Polkovskyi’s further arrangement of Doğa
Özden’s packing from a strip cut of PingYou Ltd’s n = 302 packing (T-320) is now this
case’s house: its exact certificate proves $s(268) \le 16.8718681351021783$, the
reported and the verified upper bound, rounded outward from its exact side.
The exact certificate side improves the previous verified exact side by about
1.62e-04. The earlier ceiling, displayed outward as $16.8720298490844502$ from Doğa
Özden’s packing (T-289), applied until this adoption and stays retained as history
(the certificate).
House adoption, 11 November 2026. Doğa Özden’s packing from a strip cut of PingYou
Ltd’s n = 302 packing as Francisco Couzo, Mishapolk and Evan Daniel improved it, below
his own further arrangement of Evan Daniel’s packing (T-289) was this case’s house until
18 November 2026: its exact certificate proves $s(268) \le 16.8720298490844502$, the
reported and the verified upper bound, rounded outward from its exact side.
The exact certificate side improves the previous verified exact side by about
4.40e-04. The earlier ceiling, displayed outward as $16.8724698978362125$ from Evan
Daniel’s packing, improved by Doğa Özden (T-276), applied until this adoption and stays
retained as history (the certificate).
House adoption, 9 November 2026. Doğa Özden’s further arrangement of Evan Daniel’s
packing built by his fault-line composer after Chris Deotte’s multi-gadget fault DP
(T-276) was this case’s house until 11 November 2026: its exact certificate proves
$s(268) \le 16.8724698978362125$, the reported and the verified upper bound, rounded
outward from its exact side.
The exact certificate side improves the previous verified exact side by about
2.96e-05. The earlier ceiling, displayed outward as $16.8724995181350307$ from Evan
Daniel’s packing (T-270), applied until this adoption and stays retained as history
(the certificate).
House adoption, 5 November 2026. Evan Daniel’s packing built by his fault-line
composer after Chris Deotte’s multi-gadget fault DP (T-270) was this case’s house until
9 November 2026: its exact certificate proves $s(268) \le 16.8724995181350307$, the
reported and the verified upper bound, rounded outward from its exact side.
The exact certificate side improves the previous verified exact side by about
1.86e-03. The earlier ceiling, displayed outward as $16.8743617158693459$ from
Vladislav Lavrentev’s packing, improved by Evan Daniel (T-242), applied until this
adoption and stays retained as history
(the certificate).
House adoption, 26 October 2026. Evan Daniel’s exact descent of Vladislav
Lavrentev’s public Kaggle packing (T-242) was this case’s house until 5 November 2026:
its exact certificate proves $s(268) \le 16.8743617158693459$, the reported and the
verified upper bound, rounded outward from its exact side.
The exact certificate side improves the previous verified exact side by about
4.48e-12. The earlier ceiling, displayed outward as $16.8743617158738255$ from Kirill
Tymchishin’s packing, improved by Francisco Couzo (T-240), applied until this adoption
and stays retained as history (the certificate).
House adoption, 25 October 2026. Francisco Couzo’s refinement of Kirill Tymchishin’s
Kaggle notebook packing (T-240) was this case’s house until 26 October 2026: its exact
certificate proves $s(268) \le 16.8743617158738255$, the reported and the verified upper
bound, rounded outward from its exact side.
The exact certificate side improves the previous verified exact side by about
4.45e-03. The earlier ceiling, displayed outward as $16.8788148210067647246$ from
Francisco Couzo’s packing, improved by Evan Daniel and PingYou Ltd (T-174), applied
until this adoption and stays retained as history
(the certificate).
House adoption, 17 October 2026. PingYou Ltd’s refinement of Evan Daniel’s exact
optimum of Francisco Couzo’s packing (T-174) was this case’s house until 25 October
2026: its exact certificate proves $s(268) \le 16.8788148210067647246$, the reported and
the verified upper bound, rounded outward from its exact side.
The exact certificate side improves the previous verified exact side by about
1.69e-19. The earlier ceiling, displayed outward as
$16.878814821006764724747121053018$ from Francisco Couzo’s packing, improved by Evan
Daniel (T-098), applied until this adoption and stays retained as history
(the certificate).
Held for the owner, 15 October 2026; lifted 16 October 2026. T-174 at
$16.8788148210067648$, 1.69e-19 below this case’s verified ceiling, is confirmed here
and accepted by the
review of 14 October 2026.
From 15 October 2026 until the owner’s house rule of 16 October 2026 lifted the hold, PR
#574’s houses were held for the owner, since their witnesses would add about 18.5 MB to
a checkout and their custody inputs about 5.8 MB to every worker’s snapshot.
Open. Previously, the best known published packing, Francisco Couzo’s at Evan Daniel’s
exact optimum (T-098), gives $s(268) \le 16.8788148210067647248\ldots$, and the
strongest lower bound independently verified here is $16.382380$ from Karakuş’s general
theorem, leaving a gap of $0.4964$. General bound: s(N) >= 1/2 + sqrt(N - floor(sqrt(N))
+ 1/4).
The exact optimum
Evan Daniel’s
square-packing published
on 5 October 2026 an exact rational certificate of this packing at its exact optimum
(T-098): the same 268 squares in a square of side $16.8788148210067647248\ldots$,
1.2e-11 below the side Francisco Couzo prints.
Square for square, its pose lies within 1.5e-4 of the binary64 pose the atlas pictures
for this count. 7 squares move by more than 1e-8, all of them squares the source lists
as carrying no force; every other square moves by at most 1.8e-11. The known-best
witness this record lists is that binary64 pose, posed at its finder’s larger side; the
side above is witnessed by the certificate itself.
Evan Daniel’s solver moves the binary64 pose to a nearby exact KKT point of the problem
of minimizing the side under non-overlap, computed at 80 digits, and rounds it outward
to rationals, each square a rational centre and a rational $t = \tan(\theta/2)$.
This repository decides the certificate exactly.
Converted without rounding, every pair and every wall is decided over $\mathbb{Q}$
twice, by sqpack’s exact separating-axis test and by an independent checker that
shares no code with it, and the source’s own two checkers, run here as retained, accept
it as well
(receipts).
That proves $s(268) \le 16.8788148210067647248\ldots$, the verified upper bound until 17
October 2026; it says nothing about optimality.
The source reports the exact point as a KKT local minimum: multipliers that keep it in
equilibrium, a reduced Hessian positive definite once its exact flat motions are set
aside (its per-count report), and no first-order descent across corner-to-corner
contacts, each computed numerically at that point; none of that is verified here, and it
bears on this packing alone, not on $s(268)$. On #375 its author wrote that
the solver, its checkers and the batch “were written with Claude (Anthropic) as a coding
and research agent, directed and reviewed by me.”
The packing
Francisco Couzo’s
square-packing
reports a packing of side $16.878814821018413$ for this count, dated 26 September 2026
and unchanged when this record retained the repository on 27 September 2026 (T-056). Its
first packing for this count, of side $16.879116936182307$, is dated 24 September 2026.
The repository names no method and no tolerance, and itself states no AI assistance; its
author said on issue #227 that he found
the 102 and 103 packings “with the help of Claude”.
This repository certifies it exactly.
The retained decimal pose rounds to an exact rational packing at centre dilation 1, of
side $16.8788148210184113883\ldots$, no larger than the printed side, and every pair and
every wall is decided over $\mathbb{Q}$ twice, by the promotion’s exact separating-axis
test and by an independent checker that shares no code with it
(receipt).
That proves $s(268) \le 16.878814821018413$, which was the verified upper bound until
Evan Daniel’s exact optimum above replaced it; it says nothing about optimality.
Interval arithmetic on the printed pose itself, with each angle’s true cosine and sine
and no rational rounding, decides every pair and wall again and gives the same bound
(interval route).
Griffin Casson’s
square-packing, dated
23 September 2026 and made with the help of Claude as its README says, reports a packing
of side $16.87895510935973675$ for this count, larger than Couzo’s by
$0.00014028834132375$. Casson’s is the earlier of the two by the timestamps this
record’s priority notes keep: Couzo’s first packing for this count, of side
$16.879116936182307$, is dated 24 September 2026. The record states both dates and
infers nothing about whether either packing derives from the other.
The previous best known packing
Before this intake the best known packing was the Kingbird catalogue’s, of side
$16.87933209237563$.
Found by David Ellsworth in 2024, via an unrecorded method.
Improved by David W. Cantrell in February 2025. Its side length is algebraic of degree
6 over $\mathbb{Q}$.
The lower bound
The strongest lower bound independently verified in this record is Karakuş’s general
bound, which applies to every nonsquare $N \ge 8$:
s(N) ≥ ½ + √(N − ⌊√N⌋ + ¼)
Correction, 2 October 2026. Until that date the verified lower bound here was
Nagamochi’s general closed form,
$s(N) \ge \min(\lceil\sqrt{N}\rceil, \sqrt{N - 2\lfloor\sqrt{N}\rfloor + 1} + 1)$, which
is stronger at this $n$ and is now recorded as a reported bound.
Its published proof rests on Nagamochi’s Lemma 1, which Karakuş showed false; nothing is
disproved, and no packing beating it is known
(review of 2 October 2026).
This register had recorded that proof as verified, its own error, logged as defect
D-516.
Source-reported bounds are recorded separately from this independently verified theorem.
Verification Code
The programs behind this case’s verified bounds, by their evidence.
The code column says how the code that ran stands to the code its producer used.
VERIFIERS.md says what each program is and whose it is.
PingYou Ltd’s refinement of Evan Daniel’s exact optimum of Francisco Couzo’s packing,
reported in pull request #574, at an exact
rational side, displayed upward as $16.8788148210067647246$ (T-174). The certificate
gives a rational side and, for each of the 268 unit squares, a rational centre and the
tangent of its half-angle; the
retained packet
holds it with the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and a third exact route, half-extent containment and exact
intersection area, decided it again.
The source ran both maintained routes itself on its corners witnesses; those runs are
its own, and the replay here is independently re-implemented.
The
review of 14 October 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8788148210067647246$, the verified upper bound until 25
October 2026: confirmed, independently re-implemented.
It says nothing about optimality, rigidity or a local minimum.
The certificate is PingYou Ltd’s tightening of the packing T-098 records, that packing
by the source’s statement, not decided here (T-174’s regularizations), so the packing
keeps its finder’s credit and PingYou Ltd is credited with the certificate.
The source makes no statement about AI assistance in this pull request; for its #503 and
#504 certificates the author said on 12 October 2026 that they were AI assisted by ‘PY’,
an unreleased agentic AI offering by PingYou Ltd.
The house it replaces, $16.878814821006764724747121053018$ from Francisco Couzo’s
packing, improved by Evan Daniel (T-098), stays retained with its own replay as history.
Francisco Couzo’s certificate (T-240)
Francisco Couzo’s refinement of Kirill Tymchishin’s Kaggle notebook packing, reported on
jlevy/squares#763, at exact side
$78645658458105249171306580101/4660659750118000000000000000$, displayed upward as
$16.8743617158738255$ (T-240). The certificate gives a rational side and, for each of
the 268 unit squares, a rational centre and the tangent of its half-angle; the
retained packet holds it with
the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and the 24 October review’s clean-room checker decided it
again on two routes of its own, in rationals by separating axes over every pair and by
the exact clipped area of every pair within reach; its controls at $10^{-30}$ refused
the closest pair moved past exact contact, the square nearest a wall enlarged past its
clearance and the box shrunk below the packing’s extent.
The source’s manifest names this repository’s sqpack/verify.py, through which one
maintained route decides, as its first checker, and both routes read the certificate
through an adapter transcribed from Evan Daniel’s checkers, from which the source’s
second checker derives (RF-4 of the 24 October review), so the replay shares the
producer’s components.
The review’s checker shares no code with either maintained route or the source, and the
independent re-implementation rests on it.
The
review of 24 October 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8743617158738255$, the verified upper bound until 26 October
2026: confirmed, independently re-implemented, by a Claude agent, a system family the
producer also used. It says nothing about optimality, rigidity or a local minimum.
The source’s commit that publishes the certificate and its two parents end with a
co-author trailer naming Claude (Anthropic), the earlier parent’s message describes
Codex’s reviewed intake, and the header of the source’s second checker says Francisco
Couzo modified Evan Daniel’s verifier with Codex.
The source’s manifest says Francisco Couzo’s live search refined Kirill Tymchishin’s
packing from his Kaggle notebook “CPU/GPU/MPS EDA+Seam LP” v4, which the packet retains
as the parent. Measured by the 24 October review, square by square, nearest centres pair
it one to one with that parent under a reflection of the box, no square moving beyond a
tenth (at most 0.059) or turning beyond a degree: a refinement of the notebook packing.
Against T-174’s packing, the house best packing here before this adoption, no symmetry
pairs nearest centres one to one and 118 squares lie beyond a tenth, so it is another
arrangement than T-174’s (T-240). Vladislav Lavrentev’s public Kaggle packing, which
Evan Daniel later certified $4.48 \times 10^{-12}$ below this side, is the same
arrangement in another placement, the mirror image of this one; whether the credit
should name Lavrentev is asked of the author.
The house it replaces, $16.8788148210067647246$ from Francisco Couzo’s packing, improved
by Evan Daniel and PingYou Ltd (T-174), stays retained with its own replay as history.
T-242’s exact side at this count 4.48e-12 below this house, is confirmed here and
accepted by the
review of 24 October 2026.
The owner’s house rule of 16 October 2026 adopted it: the case kept this house until its
adoption, staged under think-3anz, moved it to T-242’s side on 26 October 2026.
Evan Daniel’s certificate (T-242)
Evan Daniel’s exact descent of Vladislav Lavrentev’s public Kaggle packing, reported in
his repository on 24 October 2026, with no
issue, at exact side $8437180857934672912242411148233/500000000000000000000000000000$,
displayed upward as $16.8743617158693459$ (T-242). The certificate gives a rational side
and, for each of the 268 unit squares, a rational centre and the tangent of its
half-angle; the
retained packet holds it
with the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and the 24 October review’s clean-room checker decided it
again in rationals, by separating axes over every pair and by the exact clipped area of
every pair within reach; its controls at $10^{-30}$ refused the least-gap pair moved
past exact contact, every square moved past its near wall and the side lowered past the
far wall. Both routes read the certificate through an adapter whose line filter and
corner construction are transcribed from Evan Daniel’s own checkers, the ones the
certificate names (RF-4 of the 24 October review of issue 763, which holds here), so the
replay shares the producer’s components.
The review’s checker shares no code with either maintained route or the source, and the
independent re-implementation rests on it.
The
review of 24 October 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8743617158693459$, the verified upper bound until 5 November
2026: confirmed, independently re-implemented, by a Claude agent, a system family the
producer also used. It says nothing about optimality, rigidity or a local minimum.
The source’s commit names Claude as co-author, and its credits say the repository’s work
was produced by Claude (Anthropic) under human direction.
The source credits Vladislav Lavrentev’s public Kaggle packing and descends it exactly;
its report gives the descent’s input side as that of Lavrentev’s placement, which the
certificate is to $1.41 \times 10^{-4}$. Measured by the 24 October review, square by
square, it is one arrangement with T-240’s packing and another placement: nearest
centres pair the two one to one under a reflection of the box, none beyond a tenth and
135 beyond $10^{-3}$, so the house’s credit follows the placement adopted.
The house it replaces, $16.8743617158738255$ from Kirill Tymchishin’s packing, improved
by Francisco Couzo (T-240), stays retained with its own replay as history.
The other certificates confirmed here at this count, T-174 at $16.8788148210067648$,
stay verified and are not selected.
Evan Daniel’s certificate (T-270)
Evan Daniel’s packing built by his fault-line composer after Chris Deotte’s multi-gadget
fault DP, reported on jlevy/squares#914,
at exact side $2109062439766878830759866953431/125000000000000000000000000000$,
displayed upward as $16.8724995181350307$ (T-270). The certificate gives a rational side
and, for each of the 268 unit squares, a rational centre and the tangent of its
half-angle; the retained packet
holds it with the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and the 5 November review’s clean-room checker decided it
again on two methods of its own, by separating axes in rationals over every pair and by
the exact clipped area of every pair in GMP rationals; its fourteen controls, at exact
contact and $10^{-30}$ past it, were each decided as stated beforehand.
Both routes read the certificate through an adapter whose line filter and corner
construction are transcribed from Evan Daniel’s own checkers, the ones the certificate
names, so the replay shares the producer’s components.
The review’s checker shares no code with either maintained route or the source, and the
independent re-implementation rests on it.
The
review of 5 November 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8724995181350307$, the verified upper bound until 9 November
2026: confirmed, independently re-implemented, by a Claude agent, a system family the
producer also used. It says nothing about optimality, rigidity or a local minimum.
Issue 914 says the composer, the search, the checks and the request were done with
Claude; the source’s commit names Claude as co-author, and its credits say the
repository’s work was produced by Claude (Anthropic) under human direction.
The issue says Evan Daniel’s fault-line composer, his independent implementation of
Chris Deotte’s multi-gadget fault DP from Deotte’s Kaggle notebooks, built it, first on
1 November, with gadget donors not recorded and no public start known.
Measured by the 5 November review, no symmetry pairs it one to one with Evan Daniel’s
exact descent of Vladislav Lavrentev’s Kaggle packing, the house best packing here
before this adoption: under the best, a half turn, 80 squares lie beyond a tenth and 29
tilts beyond a degree are reversed, so it is another arrangement.
The certificate is the source’s 80-digit KKT point dilated by $1 + 10^{-20}$; the source
calls its local minimality inconclusive.
The house it replaces, $16.8743617158693459$ from Vladislav Lavrentev’s packing,
improved by Evan Daniel (T-242), stays retained with its own replay as history.
The other certificates confirmed here at this count, T-240 at $16.8743617158738255$ and
T-174 at $16.8788148210067648$, stay verified and are not selected.
Doğa Özden’s certificate (T-276)
Doğa Özden’s further arrangement of Evan Daniel’s packing built by his fault-line
composer after Chris Deotte’s multi-gadget fault DP, reported on
jlevy/squares#928, at an exact rational
side, displayed upward as $16.8724698978362125$ (T-276). The certificate gives a
rational side and, for each of the 268 unit squares, its four rational corners; the
retained packet holds it
with the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and the 9 November review’s clean-room checker, frozen
before any producer or witness code was read, decided it again by separating axes in
rationals and by exact clipped areas in GMP rationals over every square, wall and pair;
its controls, every near pair and every square at exact contact and $10^{-30}$ past it,
were each decided as stated beforehand.
The first route, sqpack’s rational witness verifier, is the packing-witness verify
the source ran, from the package whose promotion made the exact coordinates and ran its
exact verify_packing on them, so on that route the replay is reproduced with the
producer’s code; the independent rational corner checker shares no verification code
with it, and the review’s checker shares no code with either route or the source.
The
review of 9 November 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8724698978362125$, the verified upper bound until 11 November
2026: confirmed, independently re-implemented, by a Claude agent, a system family whose
code made the coordinates.
It says nothing about optimality, rigidity or a local minimum.
Issue 928 says the search was AI-assisted throughout, written and run by AI agents, and
that Doğa Özden’s AI agent, which uses the account kenaznia-doga, prepared and posted
the issue and the repository at his direction.
The issue says the packing starts from the registry’s 268, Evan Daniel’s packing of
issue 914 (T-270), built by his fault-line composer after Chris Deotte’s multi-gadget
fault DP, this case’s house best packing before this adoption, taken from this
repository’s regularized known-best atlas, and was found by basin hopping over local
minima of Evan Daniel’s ALM+SLP quench; this repository’s packing-witness promote made
its exact coordinates.
Measured by the 9 November review against T-270’s witness, it is a further arrangement
of T-270’s packing: 4 squares lie beyond a tenth, at most 0.274, 3 turn by more than 1
degree, the largest by 2.14 degrees, and 99 squares are tilted beyond 1 degree against
T-270’s 98. So the packing’s finder stays Evan Daniel, with Doğa Özden its improver; the
issue claims the upper bound only.
Exactly three squares touch a wall, the left, the bottom and the far wall that sets the
side, where the source says the wall squares touch.
The house it replaces, $16.8724995181350307$ from Evan Daniel’s packing (T-270), stays
retained with its own replay as history.
The other certificates confirmed here at this count, T-242 at $16.8743617158693459$,
T-240 at $16.8743617158738255$ and T-174 at $16.8788148210067648$, stay verified and are
not selected.
Doğa Özden’s certificate (T-289)
Doğa Özden’s packing from a strip cut of PingYou Ltd’s n = 302 packing as Francisco
Couzo, Mishapolk and Evan Daniel improved it, below his own further arrangement of Evan
Daniel’s packing, reported on
jlevy/squares#945, at an exact rational
side, displayed upward as $16.8720298490844502$ (T-289). The certificate gives a
rational side and, for each of the 268 unit squares, its four rational corners; the
retained packet holds it
with the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and the 11 November review’s clean-room checker, frozen
before any producer or witness code was read, decided it again by separating axes in
rationals and by exact clipped areas in GMP rationals over every square, wall and pair;
its 15 controls, each stated beforehand, were each decided as stated.
The first route, sqpack’s rational witness verifier, is the packing-witness verify
the source ran, from the package whose promotion made the exact coordinates, so on that
route the replay is reproduced with the producer’s code; the independent rational corner
checker shares no verification code with it, and the review’s checker shares no code
with either route or the source.
The
review of 11 November 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8720298490844502$, the verified upper bound until 18 November
2026: confirmed, independently re-implemented, by a Claude agent, a system family whose
code made the coordinates.
It says nothing about optimality, rigidity or a local minimum.
Issue 945 says the search and its code were written and run by AI agents, and that Doğa
Özden’s AI agent, which uses the account kenaznia-doga, prepared and posted the issue
and the repository at his direction.
The issue says the packing was found by transplant search from the registry’s n = 302
packing, PingYou Ltd’s as Francisco Couzo, Mishapolk and Evan Daniel improved it, a
horizontal and a vertical unit-wide strip cut out and closed up, then quenched with Evan
Daniel’s ALM+SLP code and basin-hopped over its local minima; the 11 November review’s
cut model fits it to that seed over most of the box, 235 squares and 24 of its 43 tilted
ones, where the house matches 137 and none.
Against T-276’s packing, the author’s own of issue 928 and this case’s house best
packing before this adoption, it is a new arrangement: no symmetry pairs nearest centres
one to one, and the optimal assignment moves 125 squares beyond a tenth, at most 1.041.
So Doğa Özden is the packing’s finder, from PingYou Ltd’s packing as the seed; the issue
claims the upper bound only.
Eight squares touch a wall, in groups the decimal pose ties to one coordinate, and its
two extents are equal, so the side is set at the right wall and the top at once.
The house it replaces, $16.8724698978362125$ from Evan Daniel’s packing, improved by
Doğa Özden (T-276), stays retained with its own replay as history.
The other certificates confirmed here at this count, T-270 at $16.8724995181350307$,
T-242 at $16.8743617158693459$, T-240 at $16.8743617158738255$ and T-174 at
$16.8788148210067648$, stay verified and are not selected.
Mykhailo Polkovskyi’s certificate (T-320)
Mykhailo Polkovskyi’s further arrangement of Doğa Özden’s packing from a strip cut of
PingYou Ltd’s n = 302 packing, reported on
jlevy/squares#1009, at exact side
$2108983516887772284036428313677/125000000000000000000000000000$, displayed upward as
$16.8718681351021783$ (T-320). The certificate gives a rational side and, for each of
the 268 unit squares, a rational centre and the tangent of its half-angle; the
retained packet holds it
with the complete native jobs that decided it.
Both maintained exact routes decided it here, sqpack’s exact separating-axis verifier
and the independent rational corner checker, with its duplicate-square and
outside-container controls, and the 17 November review’s clean-room checker, frozen
before any producer, witness or verifier code was read, decided it again by separating
axes in integers from its centres and half-angle tangents and by exact clipping in GMP
rationals from its corners, each run on both of its forms, over every square, wall and
pair; its 37 controls at this count, stated beforehand as rules, were each decided as
stated. Both routes read the certificate through an adapter transcribed from Evan
Daniel’s checkers, the family of the source’s own checker, and the second is the
independent rational corner checker the issue says the author ran on the same packing’s
Witness/v2 file, so the replay shares the producer’s components.
The review’s checker shares no code with either maintained route or the source, and the
independent re-implementation rests on it.
The
review of 17 November 2026
accepted the replay for finite feasibility.
That proves $s(268) \le 16.8718681351021783$, the verified upper bound: confirmed,
independently re-implemented.
It says nothing about optimality, rigidity or a local minimum.
Issue 1009 says the packings were discovered by Mykhailo Polkovskyi, assisted by Gemini.
The issue calls the packing an improved packing after Doğa Özden’s (T-289). Measured, no
symmetry pairs it one to one with T-289’s packing, and the optimal assignment moves 10
squares beyond a tenth, at most 0.809: a further arrangement of Özden’s packing.
So Doğa Özden is the packing’s finder and Mykhailo Polkovskyi its improver, as the issue
asks.
The house it replaces, $16.8720298490844502$ from Doğa Özden’s packing (T-289), stays
retained with its own replay as history.
The other certificates confirmed here at this count, T-276 at $16.8724698978362125$,
T-270 at $16.8724995181350307$, T-242 at $16.8743617158693459$, T-240 at
$16.8743617158738255$ and T-174 at $16.8788148210067648$, stay verified and are not
selected.