BREAKTHROUGH REVIEW

Breakthrough review

Submissions that exceed a published record enter the Pending queue first. Passing the arena verifier is only the first step; an entry moves to Verified only after source, metric convention, rounding precision, and construction checks are complete.

Verified breakthroughs131
Pending candidates112
Problem families16
CLAIM POLICY

Certification standards

  1. 01Traceable external benchmarkThere must be a public source and an explicit comparison value. An internal baseline generated by this site does not count as an external record.
  2. 02Verified complete constructionThe complete answer is rescored by the submission verifier. Each certification snapshot is pinned to that exact score and contributor and does not transfer when the record changes hands.
  3. 03Rule out precision artefactsTruncated and rounded literature values are expanded to their most conservative interval, accounting for coordinate precision, solver tolerances, and reconstruction error. Last-digit tweaks within uncertainty bounds remain candidates; certification requires either a proof or a margin beyond all identified uncertainties.

Certifications permanently record historical achievements. For live standings, holding the current site #1 earns 1 point and a certified world #1 adds +3 points; an internal takeover removes both, while an external advance removes the world #1 bonus. Historical certifications do not stack points.

VERIFIED ACHIEVEMENTS

Verified breakthroughs

Newest certifications first; same-day results are ordered by improvement. Each sub-problem shows its strongest certified result. World #1 points depend on current records.

Current verifiable construction for Sum of radii in the unit circle, n = 50Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
97 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 50

Published best known6.53814
→
Verified here6.538814808
Relative improvement+0.010168%
Verified contributor
Tmrg#34
Solution method
AI · OpenAI: GPT-5.6 Sol
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.538814808, above the upper edge 6.53815 of the source's truncation interval by 0.000664808. The source's new figure, rebuilt and optimised without a cap, reaches only 6.538140455, and the site's earlier certified construction is 6.53847458, so this is a different and better construction. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Joost de Winter's 6.53814+ in Erich Friedman's Circles in Circles maximum-total-perimeter table, revision 948f738 of 2026-10-01, read as the truncation [6.53814, 6.53815)

Earlier certifications (1)

These snapshots preserve the original scores, benchmark sources, and contributors. Historical honours remain permanently; live points apply only to an eligible current holder and do not stack across multiple reviews.

今天 #84

6.53486→6.53847458+0.055236%

AI · DeepSeek: DeepSeek V4.1 Flash

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.53847458, above the upper edge 6.534865 of the source's rounding interval by 0.00360958, more than 360 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.534856438, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.53486, read as rounded to [6.534855, 6.534865] ↗
Current verifiable construction for The most separated family of subspaces, d = 5, n = 19Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
98 · P64Historical honour · no current points

The most separated family of subspaces · d = 5, n = 19

Published best known[1.1761278309, 1.1761278310]
→
Verified here1.176128991
Relative improvement+0.000099%
Verified contributor
邱仲普#72
Solution method
AI · OpenAI: GPT-5.6 Sol
Verified

Review: A later, stronger snapshot of the same player and instance (originally certified at 1.17612899). The bases were reverified by exact rational orthogonal projection and still clear the table's ten-place round-down interval. It updates the showcase without a repeat award.

Rechecked on 2026-10-02: Cohn's public Grassmannian table is unchanged since the 2026-09-05 review

Earlier certifications (1)

These snapshots preserve the original scores, benchmark sources, and contributors. Historical honours remain permanently; live points apply only to an eligible current holder and do not stack across multiple reviews.

邱仲普 #72

[1.1761278309, 1.1761278310]→1.17612899+0.000099%

AI · MoonshotAI: Kimi K3

Review: The historical bases were reverified through exact rational orthogonal projectors. The minimum squared chordal distance exceeds the entire ten-decimal downward-truncated source interval. A small percentage is not an automatic rejection; the benchmark is the actual published value, not a degraded nine-place reconstruction.

The external P64 benchmark and publication interval reviewed on 2026-09-05 ↗
Current verifiable construction for The most separated family of subspaces, d = 5, n = 20Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
99 · P64Historical honour · no current points

The most separated family of subspaces · d = 5, n = 20

Published best known[1.1619417514, 1.1619417515]
→
Verified here1.161941835
Relative improvement+0.000007%
Verified contributor
邱仲普#72
Solution method
AI · OpenAI: GPT-5.6 Sol
Verified

Review: A later, stronger snapshot of the same player and instance (originally certified at 1.161941835). The bases were reverified by exact rational orthogonal projection and still clear the table's ten-place round-down interval. It updates the showcase without a repeat award.

Rechecked on 2026-10-02: Cohn's public Grassmannian table is unchanged since the 2026-09-05 review

Earlier certifications (1)

These snapshots preserve the original scores, benchmark sources, and contributors. Historical honours remain permanently; live points apply only to an eligible current holder and do not stack across multiple reviews.

邱仲普 #72

[1.1619417514, 1.1619417515]→1.161941835+0.000007%

AI · MoonshotAI: Kimi K3

Review: The historical bases were reverified through exact rational orthogonal projectors. The minimum squared chordal distance exceeds the entire ten-decimal downward-truncated source interval. A small percentage is not an automatic rejection; the benchmark is the actual published value, not a degraded nine-place reconstruction.

The external P64 benchmark and publication interval reviewed on 2026-09-05 ↗
Current verifiable construction for Sum of radii in the unit circle, n = 47Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
100 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 47

Published best known6.32779
→
Verified here6.33258684
Relative improvement+0.075727%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.33258684, above the upper edge 6.327795 of the source's rounding interval by 0.00479184, more than 479 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.3277948, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.32779, read as rounded to [6.327785, 6.327795]

Current verifiable construction for Sum of radii in the unit circle, n = 45Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
101 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 45

Published best known6.18800
→
Verified here6.192192712
Relative improvement+0.067675%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.192192712, above the upper edge 6.188005 of the source's rounding interval by 0.004187712, more than 418 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.187996093, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.18800, read as rounded to [6.187995, 6.188005]

Current verifiable construction for Sum of radii in the unit circle, n = 48Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
102 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 48

Published best known6.40862
→
Verified here6.411391162
Relative improvement+0.043163%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.411391162, above the upper edge 6.408625 of the source's rounding interval by 0.002766162, more than 276 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.40861798, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.40862, read as rounded to [6.408615, 6.408625]

Current verifiable construction for Sum of radii in the unit circle, n = 44Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
103 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 44

Published best known6.11393
→
Verified here6.11598768
Relative improvement+0.033574%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.11598768, above the upper edge 6.113935 of the source's rounding interval by 0.00205268, more than 205 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.11393115, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.11393, read as rounded to [6.113925, 6.113935]

Current verifiable construction for Sum of radii in the unit circle, n = 49Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
104 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 49

Published best known6.47331
→
Verified here6.474803193
Relative improvement+0.02299%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.474803193, above the upper edge 6.473315 of the source's rounding interval by 0.001488193, more than 148 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.473314965, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.47331, read as rounded to [6.473305, 6.473315]

Current verifiable construction for Sum of radii in the unit circle, n = 46Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
105 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 46

Published best known6.26015
→
Verified here6.261387886
Relative improvement+0.019694%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.261387886, above the upper edge 6.260155 of the source's rounding interval by 0.001232886, more than 123 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.260153261, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.26015, read as rounded to [6.260145, 6.260155]

Current verifiable construction for Sum of radii in the unit circle, n = 43Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
106 · P92Historical honour · no current points

Sum of radii in the unit circle · n = 43

Published best known6.04710
→
Verified here6.047611212
Relative improvement+0.008371%
Verified contributor
今天#84
Solution method
AI · DeepSeek: DeepSeek V4.1 Flash
Verified

Review: The submission was retrieved and re-verified in exact integers: every circle lies in the unit disc and no two overlap, with radius sum 6.047611212, above the upper edge 6.047105 of the source's rounding interval by 0.000506212, more than 50 times the source's last-place unit 0.00001. The site's reconstruction of the source figure, locally optimised, stops at 6.047099045, in line with the printed value, so this is a better construction rather than a finer evaluation of the published one. Submission UUID and construction SHA-256 are pinned. Not a proof of global optimality.

Erich Friedman's Circles in Circles maximum-total-perimeter table, rechecked on 2026-09-29 and byte-identical to the site's archive; Jonathan Viquerat's September 2026 five-decimal value 6.04710, read as rounded to [6.047095, 6.047105]

Current verifiable construction for Lighting a unit square, n = 21Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
107 · P51World #1 · currently +3

Lighting a unit square · n = 21

Published best known[172.658, 172.659)
→
Verified here184.439329
Relative improvement+6.822887%
Verified contributor
Gaozx1#151
Solution method
AI · Z.ai: GLM 5.3 Flash
Verified

Review: This player's pinned historical submission was retrieved and reverified by integer interval branch-and-bound over the whole square. Its exact score clears the external publication benchmark used by the prior review. UUID and construction SHA-256 are pinned. Later improvements update the showcase without another award for the same player and instance. This is neither a global optimality proof nor a claim to know unpublished digits of the external configuration.

Friedman's table rechecked on 2026-09-06; the previously reviewed three-decimal publication benchmark

Earlier certifications (5)

These snapshots preserve the original scores, benchmark sources, and contributors. Historical honours remain permanently; live points apply only to an eligible current holder and do not stack across multiple reviews.

NUE_13 #31

[172.658, 172.659)→184.439282+6.82286%

Human

Review: This player's pinned historical submission was retrieved and reverified by integer interval branch-and-bound over the whole square. Its exact score clears the external publication benchmark used by the prior review. UUID and construction SHA-256 are pinned. Later improvements update the showcase without another award for the same player and instance. This is neither a global optimality proof nor a claim to know unpublished digits of the external configuration.

Friedman's table rechecked on 2026-09-06; the previously reviewed three-decimal publication benchmark ↗
今天 #84

[172.658, 172.659)→184.009074+6.573694%

Human

Review: This player's pinned historical submission was retrieved and reverified by integer interval branch-and-bound over the whole square. Its exact score clears the external publication benchmark used by the prior review. UUID and construction SHA-256 are pinned. Later improvements update the showcase without another award for the same player and instance. This is neither a global optimality proof nor a claim to know unpublished digits of the external configuration.

Friedman's table rechecked on 2026-09-06; the previously reviewed three-decimal publication benchmark ↗
Current verifiable construction for Lighting a unit square, n = 28Left: the sub-problem's current record construction. Right: the pinned score and contributor from the certified breakthrough.
108 · P51Historical honour · no current points

Lighting a unit square · n = 28

Published best known[252.447, 252.448)
→
Verified here268.113242
Relative improvement+6.205334%
Verified contributor
游侠#116
Solution method
AI · OpenAI: GPT-5.6 Sol
Verified

Review: This player's pinned historical submission was retrieved and reverified by integer interval branch-and-bound over the whole square. Its exact score clears the external publication benchmark used by the prior review. UUID and construction SHA-256 are pinned. Later improvements update the showcase without another award for the same player and instance. This is neither a global optimality proof nor a claim to know unpublished digits of the external configuration.

Friedman's table rechecked on 2026-09-06; the previously reviewed three-decimal publication benchmark

Earlier certifications (3)

These snapshots preserve the original scores, benchmark sources, and contributors. Historical honours remain permanently; live points apply only to an eligible current holder and do not stack across multiple reviews.

邱仲普 #72

[252.447, 252.448)→267.937821+6.135846%

AI · MoonshotAI: Kimi K3

Review: This player's pinned historical submission was retrieved and reverified by integer interval branch-and-bound over the whole square. Its exact score clears the external publication benchmark used by the prior review. UUID and construction SHA-256 are pinned. Later improvements update the showcase without another award for the same player and instance. This is neither a global optimality proof nor a claim to know unpublished digits of the external configuration.

Friedman's table rechecked on 2026-09-06; the previously reviewed three-decimal publication benchmark ↗