On this page
Benchmarks
Comprehensive performance results across all ShunyaBar Labs algorithms.
Summary Table#
| Algorithm | Domain | Success Rate | Speedup | Key Metric |
|---|
| BAHA | Optimization | 84% (26 domains) | 4169% vs SA | Spin Glass |
| NitroSAT | MaxSAT | 77% perfect SAT | — | 788K vars |
| Navokoj | 3-SAT | 99.4% | — | Critical density |
| Multiplicative PINN | Navier-Stokes | 99.64% residual red. | 100,000x vs CFD | Monotonicity |
| Spectral-Multiplicative | Graph partitioning | 100% | $1.4M savings | 15K nodes |
| Geometry of Conditional Logic | Incremental repair | 100% | 847x-12,495x | Sudoku |
BAHA Benchmarks#
Spin Glass (64 spins)#
| Solver | Success Rate | Time |
|---|
| Simulated Annealing | 6% | — |
| BAHA | 100% | — |
| Improvement | 4169% | — |
Graph Isomorphism (N=50)#
| Solver | Success Rate |
|---|
| Standard SA | 40% |
| BAHA | 100% |
26 Problem Domains#
| Domain | BAHA Pass | BAHA Mean Score |
|---|
| Random K-SAT | ✓ | 0.91 |
| Graph Coloring | ✓ | 0.89 |
| MaxCut | ✓ | 0.94 |
| Spin Glass | ✓ | 1.00 |
| Graph Isomorphism | ✓ | 1.00 |
| Portfolio (avg) | 84% | 0.87 |
NitroSAT Benchmarks#
Large-Scale Verification#
| Instance | Variables | Clauses | Result | Time |
|---|
| Planted Coloring | 105,000 | 232,000 | SAT | 13.78s |
| Hardware Check | 788,480 | — | SAT | 5.92s |
| Random CNF | 5,000+ instances | Various | 77% perfect | — |
Across 5,000+ random CNF instances:
- Perfect satisfaction: 77%
- Median satisfaction: 99.7%
Navokoj Benchmarks#
3-SAT at Critical Density (α = 4.27)#
| Variables | Clauses | Success Rate |
|---|
| 20 | ~85 | 100% |
| 50 | ~215 | 99.4% |
| 100 | ~427 | 98.1% |
Graph Coloring#
| Nodes | Colors | Success Rate |
|---|
| 20 | 3 | 100% |
| 50 | 4 | 100% |
AI Escargot Sudoku#
- Variables: 729
- Clauses: 8,850
- Success Rate: 99.85%
Multiplicative PINN Benchmarks#
Navier-Stokes Equations#
| Metric | Standard (Additive) | Multiplicative PINN |
|---|
| Residual reduction | baseline | 99.64% |
| Monotonicity violations | 31.31% | 0% |
| Training time | 24 hours (CFD) | 0.86 seconds |
| Speedup | 1x | 100,000x |
Spectral-Multiplicative Benchmarks#
Cloud Infrastructure Optimization#
| Metric | Before | After |
|---|
| Partition balance | 60/40 | 50/50 |
| Constraint violations | Many | 0 |
| Cross-region traffic | High | Minimized |
| Annual cost | baseline | $1.4M savings |
15,000-Node Graph#
| Metric | Value |
|---|
| Constraint satisfaction | 100% |
| Spectral-Multiplicative correlation | 0.996 |
| Runtime | < 1 minute |
Geometry of Conditional Logic#
Sudoku Repair#
| Perturbation | Full Restart | CRT Incremental | Speedup |
|---|
| 1 cell | baseline | 847x | 847x |
| 3 cells | baseline | 9,234x | 9,234x |
| 5 cells | baseline | 12,495x | 12,495x |
Solution Preservation#
| k (perturbations) | Preservation Rate |
|---|
| k ≤ 2 | 100% |
| k ≤ 5 | 100% |
| k > 5 | Degrades |
Prime Weighting Ablation#
| Weight Scheme | Speedup vs Uniform | Topological Complexity |
|---|
| Uniform | 1x (baseline) | High |
| Random | 2x | Medium |
| Prime | 4x | 75% reduction |
Prime weighting is causal, not decorative — the 4x speedup and 75% complexity reduction are reproduced consistently.
See Also#