Audit-relevance Classification on benchmark_1000 n=993 (train)
79.6AccuracyPipeline current
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Pipeline currentType=Production pipeline2026.06 | 79.6 | 50 | 0 | |
| Always-approve (degenerate)Type=Degenerate reference2026.06 | 74.6 | 0 | 0 | |
| Pre-deployment pure LLMMode=Pure LLM2026.06 | 72.7 | — | 100 | |
| Opus + skills_v8 promptPrompt=skills_v82026.06 | 69.8 | 17 | 100 | |
| Opus zero-shot (B1b)Mode=Zero-shot, Variant=B1b2026.06 | 43.8 | 32 | 100 |