HYPERSIRO¶

HYPERSIRO extends SIRO from 3 dimensions (row/col/box) to 10 dimensions — adding band, stack, miniline, cross-band, cross-stack, and unconnected zone metrics. It also introduces Zone Position Hidden Singles, a new technique with zero overlap with any existing Sudoku technique.

The 10 Dimensions¶

#	Zone	Cells	Sum	What it measures
1	Row	9	45	Standard row rivals
2	Col	9	45	Standard column rivals
3	Box	9	45	Standard box rivals
4	Band	27	135	Rivals across 3-row band
5	Stack	27	135	Rivals across 3-column stack
6	Row-ML	3	—	Miniline: cells in same box & same row
7	Col-ML	3	—	Miniline: cells in same box & same col
8	Cross-band	2 boxes	—	Other boxes in same band
9	Cross-stack	2 boxes	—	Other boxes in same stack
10	Unconnected	4 boxes	—	Boxes sharing no row or col

HYPERSIRO Cascade¶

Standard SIRO predicts all cells equally. HYPERSIRO orders predictions by channeling — how many cross-zone boxes have the digit placed. The most channeled cells get predicted first, and propagation handles the rest.

from wsrf_zone_oracle import ZoneAnalyzer

analyzer = ZoneAnalyzer()
result = analyzer.hypersiro_cascade("980700600...", solution="985732641...")
print(f"Accuracy: {result['accuracy']:.1%}")

Results (50 Hardest Puzzles)¶

Method	Accuracy
Standard SIRO cascade	41.4%
HYPERSIRO (channeling order)	46.1%
HYPERSIRO + Zone Hidden Singles	57.6%

+16.2 percentage points over standard SIRO on the world's hardest puzzles.

Zone Position Hidden Singles¶

Discovered by Lars Rocha, March 2026.

The 9 box positions (TL/TC/TR/ML/MC/MR/BL/BC/BR) form 9 groups of 9 cells each. If a digit appears as a candidate in only ONE cell of its position group — that's a Zone Hidden Single.

Key Properties¶

Zero overlap with standard hidden singles (row/col/box)
50.5% accuracy when pinned (row or col rivals ≤ 1)
28.3% accuracy unpinned
12.1 per puzzle on average — lots of signal
A completely new constraint dimension in Sudoku

Why It Works¶

Standard hidden singles find digits unique in a row, column, or box. Zone hidden singles find digits unique across a positional pattern — all cells at the same relative position within their boxes. This captures structural relationships that cross row/col/box boundaries.

analyzer = ZoneAnalyzer()
zhs = analyzer.zone_hidden_singles("980700600...", solution="985732641...")

for z in zhs:
    if z['pinned']:
        print(f"R{z['row']+1}C{z['col']+1} d={z['digit']} {z['box_pos']} "
              f"confidence={z['confidence']:.1%}")

Cross-Zone Tension¶

Each candidate has a tension between its zone levels:

band_vs_row = band_ratio - row_ratio
stack_vs_col = stack_ratio - col_ratio

Negative tension means the digit is MORE constrained at the larger scale — the band/stack is channeling it. This correlates with oracle accuracy.

Closeness to 135¶

Each band sums to 135, each stack to 135, each row to 45. The residual (target - placed sum) constrains which digits can go where. HYPERSIRO tracks these residuals and uses them as additional prediction signals.

HTML Integration¶

The Larsdoku Research HTML (larsdoku_deploy_hypersiro.html) provides:

TGZ Button — Toggle Zone highlighting: click a cell, press TGZ, see all cells in the same box-position that share the same Likely #1 prediction
HYPERSIRO Cascade Button — Run the full cascade with channeling + ZHS boost
10-zone inspector — Every candidate shows band/stack rivals, ML confinement, cross-zone coverage, tension vectors, residuals, and band/stack harmonic distances
Zone data attributes — Every cell has data-band, data-stack, data-box, data-boxpos for CSS targeting

The Channeling Signal¶

Candidates with band_boxes_placed + stack_boxes_placed >= 2 are 2.2x more likely to be the oracle (37.5% vs 16.7%). This signal doesn't improve per-cell prediction directly, but it dramatically improves cascade ordering — predict the most channeled cells first.

ML Confinement¶

When a digit is confined to a single miniline within its box (row_ml_confined or col_ml_confined), that's a pointing pair signal. This correlates with oracle accuracy at 43.4% (above the 41% baseline).