Home / Measured, not merely claimed.
Measured, not merely claimed.
Open benchmark results with sample sizes, models, dates and blocked releases. Mechanical rates published, legal correctness pending practitioner grading.
Citation integrity ≠ legal correctness. Mechanical rates below; practitioner grading for legal correctness is pending.
| Benchmark | Result | Status |
|---|---|---|
| Grounding-100 (n=95, 2026-09-09) | Recall 100%; zero dangling 100%; exact-quote 72.5% (87/120) | VERIFIED INTERNAL |
| Frozen P0 full (n=100) | Exact-quote 95.8% PASS; entailment 66.7% FAIL; unsupported 81.0% FAIL | BLOCKED |
| Adversarial v1 live (n=12) | Accurate 50%; hallucinated 0%; incomplete 50% | VERIFIED INTERNAL |
| Stanford-style (n=24) | Accurate 58.3%; hallucinated 0% | VERIFIED INTERNAL |
Method: verification methodology. Sources: cross-bench report, P0 full report, ablation reports.