Larsdoku-ZS¶
Pure logic Sudoku solver. Zero guessing. Every step proven. Sudoku is solved.
Larsdoku-ZS solves the hardest Sudoku puzzles ever created using only logical deduction — no backtracking, no trial-and-error. 100% pure logic on all 48,765 puzzles in the forum hardest set (SE 11.0+), the hardest puzzle collection in existence.
At a Glance¶
| Metric | Value |
|---|---|
| Forum Hardest (SE 11.0+) | 48,765/48,765 (100%) |
| Top1465 solve rate | 1465/1465 (100%) |
| Average solve time | 0.019s (Top1465) / 0.6s (Forum Hardest) |
| Total techniques | 35 across 7 levels |
| WSRF inventions | 5 (FPC, FPCE, D2B, DeepResonance, FPF) |
| Three independent paths to 100% | DeepResonance, FPF+GF(2), Recursive FN |
| Dependencies | NumPy, Numba |
Install¶
Solve a puzzle in 3 lines¶
from larsdoku_zs import solve
result = solve("4...3.......6..8..........1....5..9..8....6...7.2........1.27..5.3....4.9........")
print(f"Solved: {result['success']} in {result['n_steps']} steps")
Or from the command line¶
larsdoku "800000000003600000070090200050007000000045700000100030001000068008500010090000400" --board --no-oracle
Status: SOLVED
Steps: 63
Time: 26.0ms
┌───────┬───────┬───────┐
│ 4 6 8 │ 9 3 1 │ 5 2 7 │
│ 7 5 1 │ 6 2 4 │ 8 3 9 │
│ 3 9 2 │ 5 7 8 │ 4 6 1 │
├───────┼───────┼───────┤
│ 1 3 4 │ 7 5 6 │ 2 9 8 │
│ 2 8 9 │ 4 1 3 │ 6 7 5 │
│ 6 7 5 │ 2 8 9 │ 3 1 4 │
├───────┼───────┼───────┤
│ 8 4 6 │ 1 9 2 │ 7 5 3 │
│ 5 1 3 │ 8 6 7 │ 9 4 2 │
│ 9 2 7 │ 3 4 5 │ 1 8 6 │
└───────┴───────┴───────┘
What makes Larsdoku different?¶
Most Sudoku solvers fall back to guessing (backtracking) when they get stuck. Larsdoku never guesses. Instead, it escalates through increasingly powerful logical techniques until every cell is proven.
The solver pipeline:
Each level fires only when everything above it has stalled. Four of these techniques — FPC, FPCE, D2B, and FPF — are original WSRF inventions not found in traditional Sudoku literature.
What is GF(2)?
GF(2) is Galois Field arithmetic (binary math). Larsdoku encodes Sudoku constraints as a system of linear equations over GF(2) and solves them using Block Lanczos decomposition. This finds forced digits that classical techniques miss — pure algebra, no search.