SUDOKU × PERMUTATION REPRESENTATIONSEED 42 · LIVE CHECKPOINT

ROSE FAMILY AS THE PROTAGONIST

Can relational order learn its way out of Sudoku?

Nine block permutations preserve every 3×3 box. The algorithms must learn how to remove row and column conflicts—without obvious-fill assistance.

100.00%384 / 384 one-seed runs
144/144ROSE family
216/216COIN family
24/24CARE/WT
800generation cap
Updated 9/9/2569 20:05:118 workers · early stop at exact solution
WHY ROSE?REPRESENTATION HYPOTHESIS

Sudoku is local in its rules—but relational in its failures.

Edge

COIN remembers which digit follows another inside a block permutation.

Position

NB-COIN remembers which digit belongs at each missing position.

Relative order

ROSE learns where digits sit relative to several references—direction, distance, and placement together.

Context

CARE will decide how much generation authority each representation deserves.

Reported evaluatorpaper penalty = 162 − Σ unique(row) − Σ unique(column)0 = exact solution; 3×3 blocks are valid by construction
Primary question: solution probability and evaluations-to-solutionobvious-fill disabled
SUDOKU 01 / 24missing-12

ONE PUZZLE · ONE EVIDENCE PAGE

Controlled · 12 missing

12 missing3.9 log |Ω|100 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCOIN/WTCOINSNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT3/33/33/3
CNB-COIN3/33/33/3
CNE-COIN3/33/33/3
COIN3/33/33/3
COIN/WT3/33/33/3
HNE-COIN3/33/33/3
NB-COIN3/33/33/3
NB-COIN/WT3/33/33/3
ROSE3/30/30/3
ROSE/WT3/33/33/3
SH-ROSE3/30/30/3
SH-ROSE/WT3/33/33/3
SNE-COIN3/33/33/3
SNE-COIN/WT3/33/33/3
SR-ROSE3/30/30/3
SR-ROSE/WT3/33/33/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 02 / 24missing-18

ONE PUZZLE · ONE EVIDENCE PAGE

Permutation warm-up · 18 missing

18 missing8.1 log |Ω|100 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCOIN/WTCOINCNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT3/33/33/3
CNB-COIN3/33/33/3
CNE-COIN3/33/33/3
COIN3/33/33/3
COIN/WT3/33/33/3
HNE-COIN3/33/33/3
NB-COIN3/33/33/3
NB-COIN/WT3/33/33/3
ROSE3/30/30/3
ROSE/WT3/33/33/3
SH-ROSE3/30/30/3
SH-ROSE/WT3/33/33/3
SNE-COIN3/33/33/3
SNE-COIN/WT3/33/33/3
SR-ROSE3/30/30/3
SR-ROSE/WT3/33/33/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 03 / 24missing-36

ONE PUZZLE · ONE EVIDENCE PAGE

Controlled · 36 missing

36 missing29.3 log |Ω|100 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINNB-COINHNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT3/33/30/3
CNB-COIN3/31/33/3
CNE-COIN2/31/33/3
COIN0/30/30/3
COIN/WT1/32/30/3
HNE-COIN2/32/33/3
NB-COIN3/33/33/3
NB-COIN/WT3/33/32/3
ROSE0/30/30/3
ROSE/WT3/33/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT3/33/30/3
SNE-COIN1/30/31/3
SNE-COIN/WT3/33/31/3
SR-ROSE0/30/30/3
SR-ROSE/WT3/33/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 04 / 24missing-42

ONE PUZZLE · ONE EVIDENCE PAGE

Controlled · 42 missing

42 missing53.4 log |Ω|120 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINCNE-COINCARE-EDA/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT3/33/30/3
CNB-COIN3/31/32/3
CNE-COIN2/31/31/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN1/31/31/3
NB-COIN1/32/33/3
NB-COIN/WT3/32/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/32/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 05 / 24paper-01

ONE PUZZLE · ONE EVIDENCE PAGE

Paper No. 1 · Easy · 43 missing

43 missing40.4 log |Ω|120 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINHNE-COINNB-COIN/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT2/32/30/3
CNB-COIN1/33/33/3
CNE-COIN2/32/33/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN3/33/33/3
NB-COIN3/33/33/3
NB-COIN/WT3/33/30/3
ROSE0/30/30/3
ROSE/WT3/33/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT3/33/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT3/33/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT1/31/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 06 / 24paper-11

ONE PUZZLE · ONE EVIDENCE PAGE

Paper No. 11 · Easy · 47 missing

47 missing47.9 log |Ω|120 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINCNB-COINNB-COIN/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT1/30/30/3
CNB-COIN2/33/33/3
CNE-COIN0/31/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN1/32/31/3
NB-COIN2/32/32/3
NB-COIN/WT3/33/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT1/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 07 / 24paper-27

ONE PUZZLE · ONE EVIDENCE PAGE

Paper No. 27 · Medium · 51 missing

51 missing54.0 log |Ω|120 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNE-COINCNB-COINNB-COIN/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/31/30/3
CNB-COIN0/30/30/3
CNE-COIN0/31/31/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/31/3
NB-COIN/WT1/31/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 08 / 24paper-29

ONE PUZZLE · ONE EVIDENCE PAGE

Paper No. 29 · Medium · 52 missing

52 missing57.3 log |Ω|140 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINHNE-COINCARE-EDA/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT1/31/30/3
CNB-COIN0/30/33/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/31/32/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 09 / 24paper-77

ONE PUZZLE · ONE EVIDENCE PAGE

Paper No. 77 · Difficult · 53 missing

53 missing57.6 log |Ω|140 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINHNE-COINNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT1/31/30/3
CNB-COIN2/31/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN1/30/30/3
NB-COIN2/31/32/3
NB-COIN/WT1/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 10 / 24paper-106

ONE PUZZLE · ONE EVIDENCE PAGE

Paper No. 106 · Difficult · 57 missing

57 missing65.6 log |Ω|200 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCARE-EDA/WTCNB-COINNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/31/31/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/31/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 11 / 24missing-24

ONE PUZZLE · ONE EVIDENCE PAGE

Intermediate · 24 missing

24 missing14.4 log |Ω|100 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINHNE-COINCNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT3/33/33/3
CNB-COIN3/33/33/3
CNE-COIN3/33/33/3
COIN0/30/30/3
COIN/WT3/33/33/3
HNE-COIN3/33/33/3
NB-COIN3/33/33/3
NB-COIN/WT3/33/33/3
ROSE3/30/30/3
ROSE/WT3/33/33/3
SH-ROSE3/30/30/3
SH-ROSE/WT3/33/33/3
SNE-COIN3/33/33/3
SNE-COIN/WT3/33/33/3
SR-ROSE3/30/30/3
SR-ROSE/WT3/33/33/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 12 / 24missing-30

ONE PUZZLE · ONE EVIDENCE PAGE

Challenge · 30 missing

30 missing24.6 log |Ω|100 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINCNE-COINHNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT3/33/30/3
CNB-COIN3/33/33/3
CNE-COIN2/32/31/3
COIN0/30/30/3
COIN/WT3/33/30/3
HNE-COIN3/33/33/3
NB-COIN3/33/33/3
NB-COIN/WT3/33/31/3
ROSE0/30/30/3
ROSE/WT3/33/30/3
SH-ROSE2/30/30/3
SH-ROSE/WT3/33/30/3
SNE-COIN2/31/32/3
SNE-COIN/WT3/33/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT3/33/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 13 / 24exchange-easy-0000183b305c

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Easy · 1.2 · #0000183b305c

51 missing54.7 log |Ω|140 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINSNE-COIN/WTCARE-EDA/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN1/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN1/30/31/3
NB-COIN1/30/31/3
NB-COIN/WT1/31/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN1/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 14 / 24exchange-easy-0001d5d6314e

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Easy · 1.2 · #0001d5d6314e

53 missing57.8 log |Ω|160 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COIN/WTCNB-COINNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT1/30/30/3
CNB-COIN2/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT1/32/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 15 / 24exchange-easy-000212406270

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Easy · 1.2 · #000212406270

53 missing58.6 log |Ω|160 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINCARE-EDA/WTCNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN1/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 16 / 24exchange-medium-0000847b216e

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Medium · 2.3 · #0000847b216e

52 missing55.8 log |Ω|140 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINSNE-COIN/WTHNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 17 / 24exchange-medium-0000c45fb232

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Medium · 2.0 · #0000c45fb232

55 missing61.5 log |Ω|200 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COIN/WTCNB-COINNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 18 / 24exchange-medium-0000fc7c6b96

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Medium · 2.3 · #0000fc7c6b96

52 missing56.1 log |Ω|140 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNE-COINNB-COINNB-COIN/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/32/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 19 / 24exchange-hard-00005f662e09

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Hard · 3.4 · #00005f662e09

54 missing60.0 log |Ω|160 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COIN/WTCNE-COINCNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 20 / 24exchange-hard-00009e4900fa

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Hard · 2.6 · #00009e4900fa

57 missing65.1 log |Ω|200 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WONB-COINNB-COIN/WTCARE-EDA/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/31/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 21 / 24exchange-hard-0000b2c3fc62

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Hard · 4.3 · #0000b2c3fc62

53 missing58.3 log |Ω|160 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOSNE-COIN/WTNB-COINCNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 22 / 24exchange-diabolical-00015097c6c3

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Diabolical · 7.2 · #00015097c6c3

53 missing57.7 log |Ω|160 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINNB-COIN/WTSNE-COIN/WT
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 23 / 24exchange-diabolical-0001d2888928

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Diabolical · 7.1 · #0001d2888928

55 missing61.3 log |Ω|200 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOCNB-COINNB-COINCNE-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/30/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN0/30/30/3
NB-COIN0/30/30/3
NB-COIN/WT0/30/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT0/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU 24 / 24exchange-diabolical-000274921f39

ONE PUZZLE · ONE EVIDENCE PAGE

Exchange Diabolical · 7.1 · #000274921f39

49 missing50.3 log |Ω|120 population16 runs ready
Generation 1 → 800Best penalty ↓
ROSE/WTSR-ROSE/WTSH-ROSE/WTROSE/WOSR-ROSE/WOSH-ROSE/WOHNE-COINCNE-COINCNB-COIN
AlgorithmUnique-count (paper)solved seedsPrime-product hierarchicalsolved seedsMO row × columnsolved seeds
CARE-EDA/WT0/30/30/3
CNB-COIN0/30/31/3
CNE-COIN0/30/30/3
COIN0/30/30/3
COIN/WT0/30/30/3
HNE-COIN1/32/30/3
NB-COIN0/30/31/3
NB-COIN/WT0/32/30/3
ROSE0/30/30/3
ROSE/WT0/30/30/3
SH-ROSE0/30/30/3
SH-ROSE/WT0/30/30/3
SNE-COIN0/30/30/3
SNE-COIN/WT1/30/30/3
SR-ROSE0/30/30/3
SR-ROSE/WT0/30/30/3
Three objectives · 16 algorithms × 3 matched seeds eachearly stop · exact Sudoku certificate
SUDOKU × OBJECTIVE TYPE3 SEEDS · 1,152 MATCHED RUNS

Unique-count (paper)

Node-based learning leads the scalar paper objective.

162 - Σ unique(row) - Σ unique(column); prime product not used in fitness

Generation 1 → 800Median residual ↓
NB-COIN/WTCNB-COINNB-COINCARE-EDA/WTHNE-COINSNE-COIN/WT
All 16 algorithms · 24 puzzles · identical seed and budgetResidual is comparable only within this evaluator
SUDOKU × OBJECTIVE TYPE3 SEEDS · 1,152 MATCHED RUNS

Prime-product hierarchical

Template-assisted node learning wins the hierarchy.

duplicate excess → prime-product invalid groups → squared excess → worst group

Generation 1 → 800Median residual ↓
NB-COIN/WTNB-COINHNE-COINCARE-EDA/WTCNB-COINSNE-COIN/WT
All 16 algorithms · 24 puzzles · identical seed and budgetResidual is comparable only within this evaluator
SUDOKU × OBJECTIVE TYPE3 SEEDS · 1,152 MATCHED RUNS

MO row × column

Pure NB-COIN leads the row × column search.

row duplicate excess × column duplicate excess

Generation 1 → 800Median residual ↓
NB-COINCNB-COINHNE-COINCNE-COINSNE-COINNB-COIN/WT
All 16 algorithms · 24 puzzles · identical seed and budgetResidual is comparable only within this evaluator
BENCHMARK × OBJECTIVERAW SOLVE STATUS · 01–12 / 24

Which objective solves each puzzle across seeds?

A seed counts as solved when at least one of the 16 algorithms reaches the exact Sudoku certificate.

BenchmarkUnique-count (paper)162 - Σ unique(row) - Σ unique(column); prime product not used in fitnessPrime-product hierarchicalduplicate excess → prime-product invalid groups → squared excess → worst groupMO row × columnrow duplicate excess × column duplicate excess
Exchange Diabolicalexchange-diabolical-00015097c6c30/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +40/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +20/3 SEEDS0/48 algorithm-runs solvedbest residual 2HNE-COIN
Exchange Diabolicalexchange-diabolical-0001d28889280/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / HNE-COIN +20/3 SEEDS0/48 algorithm-runs solvedbest residual 2HNE-COIN / NB-COIN +20/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / NB-COIN
Exchange Diabolicalexchange-diabolical-000274921f392/3 SEEDS2/48 algorithm-runs solvedbest residual 0HNE-COIN / SNE-COIN/WT3/3 SEEDS4/48 algorithm-runs solvedbest residual 0HNE-COIN / NB-COIN/WT2/3 SEEDS2/48 algorithm-runs solvedbest residual 0CNB-COIN / NB-COIN
Exchange Easyexchange-easy-0000183b305c2/3 SEEDS5/48 algorithm-runs solvedbest residual 0CNB-COIN / HNE-COIN +31/3 SEEDS1/48 algorithm-runs solvedbest residual 0NB-COIN/WT2/3 SEEDS2/48 algorithm-runs solvedbest residual 0HNE-COIN / NB-COIN
Exchange Easyexchange-easy-0001d5d6314e3/3 SEEDS4/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +12/3 SEEDS2/48 algorithm-runs solvedbest residual 0NB-COIN/WT0/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / CNE-COIN +2
Exchange Easyexchange-easy-0002124062701/3 SEEDS1/48 algorithm-runs solvedbest residual 0NB-COIN0/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / HNE-COIN +20/3 SEEDS0/48 algorithm-runs solvedbest residual 2HNE-COIN / NB-COIN
Exchange Hardexchange-hard-00005f662e090/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +30/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +50/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / CNE-COIN +1
Exchange Hardexchange-hard-00009e4900fa0/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +21/3 SEEDS1/48 algorithm-runs solvedbest residual 0NB-COIN0/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / HNE-COIN +1
Exchange Hardexchange-hard-0000b2c3fc620/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / NB-COIN/WT +10/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / HNE-COIN +20/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / NB-COIN
Exchange Mediumexchange-medium-0000847b216e0/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +40/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +50/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / HNE-COIN +1
Exchange Mediumexchange-medium-0000c45fb2320/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +30/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +30/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / CNE-COIN +2
Exchange Mediumexchange-medium-0000fc7c6b960/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +42/3 SEEDS2/48 algorithm-runs solvedbest residual 0NB-COIN0/3 SEEDS0/48 algorithm-runs solvedbest residual 2CNB-COIN / CNE-COIN +2
3 matched seeds · 800-generation cap · early stopMissing = variables to infer
BENCHMARK × OBJECTIVERAW SOLVE STATUS · 13–24 / 24

Which objective solves each puzzle across seeds?

Each cell reports seed repeatability first, then successful algorithm-runs and the best residual for that benchmark/objective pair.

BenchmarkUnique-count (paper)162 - Σ unique(row) - Σ unique(column); prime product not used in fitnessPrime-product hierarchicalduplicate excess → prime-product invalid groups → squared excess → worst groupMO row × columnrow duplicate excess × column duplicate excess
Controlled missingmissing-123/3 SEEDS48/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +143/3 SEEDS48/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +143/3 SEEDS48/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +14
Controlled missingmissing-183/3 SEEDS48/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +143/3 SEEDS48/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +143/3 SEEDS48/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +14
Controlled missingmissing-243/3 SEEDS45/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +133/3 SEEDS45/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +133/3 SEEDS45/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +13
Controlled missingmissing-303/3 SEEDS36/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +113/3 SEEDS33/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +103/3 SEEDS13/48 algorithm-runs solvedbest residual 0CNB-COIN / CNE-COIN +4
Controlled missingmissing-363/3 SEEDS30/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +103/3 SEEDS27/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +93/3 SEEDS16/48 algorithm-runs solvedbest residual 0CNB-COIN / CNE-COIN +5
Controlled missingmissing-423/3 SEEDS13/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +43/3 SEEDS12/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +53/3 SEEDS7/48 algorithm-runs solvedbest residual 0CNB-COIN / CNE-COIN +2
Published paperpaper-013/3 SEEDS24/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +83/3 SEEDS26/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +83/3 SEEDS12/48 algorithm-runs solvedbest residual 0CNB-COIN / CNE-COIN +2
Published paperpaper-1060/3 SEEDS0/48 algorithm-runs solvedbest residual 2CARE-EDA/WT / CNB-COIN +11/3 SEEDS2/48 algorithm-runs solvedbest residual 0NB-COIN / SNE-COIN/WT1/3 SEEDS1/48 algorithm-runs solvedbest residual 0NB-COIN
Published paperpaper-113/3 SEEDS10/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +43/3 SEEDS11/48 algorithm-runs solvedbest residual 0CNB-COIN / CNE-COIN +33/3 SEEDS6/48 algorithm-runs solvedbest residual 0CNB-COIN / HNE-COIN +1
Published paperpaper-271/3 SEEDS1/48 algorithm-runs solvedbest residual 0NB-COIN/WT2/3 SEEDS3/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNE-COIN +12/3 SEEDS2/48 algorithm-runs solvedbest residual 0CNE-COIN / NB-COIN
Published paperpaper-291/3 SEEDS1/48 algorithm-runs solvedbest residual 0CARE-EDA/WT2/3 SEEDS2/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / NB-COIN3/3 SEEDS5/48 algorithm-runs solvedbest residual 0CNB-COIN / NB-COIN
Published paperpaper-773/3 SEEDS7/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +32/3 SEEDS3/48 algorithm-runs solvedbest residual 0CARE-EDA/WT / CNB-COIN +12/3 SEEDS2/48 algorithm-runs solvedbest residual 0NB-COIN
3 seeds · 16 algorithms per seed · 48 runs per cellRaw fitness values are not compared across objectives
OBJECTIVE × REPRESENTATIONTHREE MATCHED SEEDS · NOT A LEADERBOARD

Different objectives expose different learnable structure.

UNIQUE-COUNT (PAPER)

Learns absolute node placement.

Supporting algorithmsNB-COIN · NB-COIN/WT · CNB-COINRepresentationdigit/node × missing-position relation

The objective rewards completing valid row and column symbol sets. Position-aware models can accumulate where each missing digit belongs; edge-only order is less direct.

PRIME-PRODUCT HIERARCHICAL

Still favors node/position evidence.

Supporting algorithmsNB-COIN/WT · NB-COIN · HNE-COINRepresentationabsolute placement + inherited partial structure

Prime-product validity changes the evaluation signal, but not the useful representation. /WT preserves promising assignments while the model refills uncertain positions.

MO ROW × COLUMN

Learns two compatible views of placement.

Supporting algorithmsNB-COIN · CNB-COIN · HNE-COIN · CNE-COIN · SNE-COINRepresentationnode position, conditioned position, and start-node/edge context

Separating row and column conflict preserves directional evidence. Hybrid COIN variants combine placement with contextual order instead of collapsing both signals into one scalar.

Main finding: Sudoku is primarily learnable as a node-to-position assignment problem. Scalar objectives strongly reward NB-style models; the MO objective additionally reveals value from conditioned and edge-context representations. Three seeds preserve this pattern even when individual ranks move.
Evidence: 3 seeds × 24 puzzles × 16 algorithms per objectiveInterpret families and representations—not isolated rank numbers
HUMAN DIFFICULTY ≠ EDA DIFFICULTYEXCHANGE PUZZLES · 3 MATCHED SEEDS

“Diabolical” can be easier to learn than “Medium.”

DIABOLICAL · 000274921f39

Learnable by every objective

Missing
49
log |Ω|
50.3
Population
120

Unique 2/3 · Prime 3/3 · MO 2/3

MEDIUM · 0000847b216e

Not solved by any objective

Missing
52
log |Ω|
55.8
Population
140

Unique 0/3 · Prime 0/3 · MO 0/3

MEDIUM · 0000c45fb232

A larger population is not sufficient

Missing
55
log |Ω|
61.5
Population
200

Unique 0/3 · Prime 0/3 · MO 0/3

Mostly missing size?

Yes, partly. The solvable Diabolical case has only 49 missing cells and log |Ω| 50.3; the two failed Medium cases have 52/55 missing and log |Ω| 55.8/61.5.

Generation?

About one third of unsolved Exchange runs still improve after generation 400, but the median late gain is zero. More generations may rescue a minority; most runs have already plateaued.

Next controlled test

Match puzzles by missing count and block-size profile first; then compare population P/2, P and 2P at equal evaluations to isolate landscape structure from search budget.

Human labels reflect solving techniques—not permutation landscape geometryClaim limited to the present 12 Exchange fixtures
MO ROW × COLUMNPAPER-106 · UNION OF 3 SEEDS

The scalar winner is not the whole frontier.

Row duplicate excess ↓Column duplicate excess ↓
NB-COINCNB-COINHNE-COINCNE-COINSNE-COINNB-COIN/WT
Top six algorithms by matched aggregate residualdo not compare raw MO fitness with scalar evaluators
CARE × EVALUATORSEED 42 · MATCHED 24 PUZZLES

The evaluator changes the route—not the certificate.

Unique-count (paper)

162 - unique rows/columns

24/2429.17% solved
Median residual
2
Median runtime
133.0 s
Median eval. to solve
13,350

Prime-product hierarchical

duplicate excess → prime-product invalid groups → squared excess → worst group

24/2429.17% solved
Median residual
2
Median runtime
103.7 s
Median eval. to solve
8,129

MO row x column

row duplicate excess x column duplicate excess

24/2412.50% solved
Median residual
16
Median runtime
91.1 s
Median eval. to solve
1,272
Exact solution is always paper penalty = 0obvious-fill disabled · 10 workers
CARE × EVALUATORCONVERGENCE · SEED 42

Same solved count; Legacy reaches its successful runs earlier.

Generation 1 → 800Median paper penalty ↓
Prime/Paper and Legacy both solve 7/24 at this checkpoint. Among solved puzzles, median evaluations-to-solution are 13,350 versus 8,129. This is one seed—not yet a superiority claim.
Curves aggregate median paper residual across puzzlesMO curve appears after corrected rerun
CARE × EVALUATORMO ROW × COLUMN

A frontier exposes which direction still carries conflict.

Row duplicate excess ↓Column duplicate excess ↓paper-106 · 8 frontier points

Why two objectives?

A scalar total can hide whether progress comes from rows or columns. The archive preserves candidates that trade one direction against the other.

What counts as solved?

Only the shared point (0,0). A nondominated score of zero is not an exact Sudoku certificate—the corrected rerun now enforces this distinction.

Pareto archive + crowdinginvalid pilot excluded from analysis
CARE × EVALUATORFIVE MATCHED SEEDS · 120 RUNS EACH

Legacy changes the route; MO changes the problem.

Prime/Paper

39/12032.50% solved
Median residual
2
Median runtime
97.4 s
Median eval. to solve
21,834

Legacy hierarchical

39/12032.50% solved
Median residual
2
Median runtime
100.0 s
Median eval. to solve
15,287

MO row x column

15/12012.50% solved
Median residual
16
Median runtime
101.0 s
Median eval. to solve
866
Generation 1 → 800Median paper penalty ↓
Interpretation: Prime and Legacy tie at 39/120 solved, but Legacy reaches successful runs earlier (15,287 vs 21,834 median evaluations). MO solves 15/120; its 866-evaluation median reflects a much smaller, easier subset and must not be read as globally faster.
5 seeds × 24 puzzles × 3 evaluators = 360 runsobvious-fill disabled · exact certificate shared
UNSOLVED-CELL FOLLOW-UPPOPULATION 240 · GENERATION 400 · SEED 2030

A larger population opens one previously closed puzzle.

Exact breakthroughs2 / 26

previously unsolved puzzle × objective cells

Runs completed
416/416
Errors
0
Newly solved puzzles
1
PuzzleObjectiveAlgorithmResult
Exchange Diabolical
0001d2888928		Unique-countCNB-COINSOLVED
MO row × columnCNB-COINSOLVED
Remaining cells24 puzzle × objective cells0 exact
What changed?

The larger population produced two exact hits, both from conditioned node-position learning. Prime-product hierarchical did not gain an exact solution.

What can we claim?

Population 240 is promising but not a general cure: it converts only 2/26 cells. Because generations also changed from 800 to 400 and only one new seed was used, this is a follow-up signal—not a clean population-size effect.

Only baseline 0/3 cells rerun · 16 algorithms per cellobvious-fill disabled · original results retained
SECOND UNSOLVED-CELL FOLLOW-UPPOPULATION 280 · GENERATION 400 · SEED 2031

More diversity opens a different objective.

New exact breakthroughs1 / 24

cells remaining after the population-240 run

Runs completed
384/384
Errors
0
Cumulative cells opened
3/26
PopulationPuzzleObjectiveAlgorithm
240Diabolical
0001d2888928		Unique-countCNB-COIN
240Diabolical
0001d2888928		MO row × columnCNB-COIN
280Medium
0000c45fb232		Prime-product hierarchicalNB-COIN
Representation signal

All three breakthroughs come from node-position learning: conditioned NB at population 240, then plain NB at population 280. No edge-only or ROSE exact breakthrough appears in these follow-ups.

Budget reading

The gain remains sparse—1/24 new cells at population 280 and 3/26 cumulatively. Different seeds and the 400-generation cap prevent a causal claim that 280 is better than 240.