noul, choice, score) in sub-5ms on commodity CPU using deterministic Mandelbrot escape dynamics. Zero weight files, zero GPU memory allocations, zero external cloud token bills.
| Benchmark Axis | v1.2 (Active at Submission) | v1.3.0 (Chance-Corrected) | Reference LLM Cloud Baselines |
|---|---|---|---|
| Easy Tier (48 items) | 87.50% (42 / 48) | ECE: 0.1270 | 87.50% (42 / 48) | ECE: 0.1270 | 75.0% – 85.4% |
| Original Tier (72 items) | 90.28% (65 / 72) | ECE: 0.0816 | 90.28% (65 / 72) | ECE: 0.0816 | 80.5% – 88.9% |
| Hard Public (111 items) | 48.65% (54 / 111) | ECE: 0.2966 | 48.65% (54 / 111) | ECE: 0.2966 | 45.0% – 62.0% |
| Intelligence Score | 72.39% | 58.59% (Chance-Adjusted) | 65.0% – 75.0% |
| Weighted ECE / Calibration | ECE: 0.1800 | Score: 64.00 | ECE: 0.1800 | Score: 64.00 | Score: 50.0 – 62.0 |
| Speed Score (Latency) | 100.0 (P50: 3.02 ms CPU) | 100.0 (P50: 3.02 ms CPU) | 55.0 – 75.0 (650–1500 ms) |
| Cost Score | 100.0 ($0.00 / 1k calls) | 100.0 ($0.00 / 1k calls) | 65.0 – 80.0 ($0.04 - $1.00/1k) |
| Composite JevScore | 82.50 (All-Time Peak) | 78.25 | 72.0 – 78.5 |
| Benchmark Suite | WERR Performance | Baseline Competitor | Speedup / Margin |
|---|---|---|---|
| WindTunnel WebMCP (49 tasks) | 49 / 49 (100.00% Success) | 1.81 ms | Production Web Agents | 100% Deterministic Air-Gapped |
| Snake AI Autonomous Reflex | 411.9 moves/s | 1.32 ms | Laya-MLX 421M (74.5 moves/s) | 5.4x Faster (0 VRAM vs 943 MiB) |
| Jevenator 2 Video Tracking | 20.04 ms/frame | 1746.4 dec/s | Maisa djev Gemma (761.8 ms) | 38.0x Speedup (0 FP Control) |
| Tau-Bench Agentic Tools | 10 / 10 Passed (100% Tool Adherence) | Commercial Agent Workflows | 0 Drift / 0 Hallucinations |
81a33e723dea04ddd40d38b056ff376cd54d2206c0920bb064af03de0bae7c5f06b134dea501216c8888aa5a3cd13e1b68b15beb31987e2c8df6872c4caeffc430111404aac815366afc93b1091f8f07c318f78ded7e56dd61fa18482f02986b333814952ad1f8f1b2c25c7749b658dd729503e48dfbea2e9142c9a4f6af5963