Best-action selection on Cross-Domain n=60
73.4Macro F1PIWM (full)
Evaluation Results
| Method | Links | |
|---|---|---|
| PIWM (full)State Conditioning=Oracle2026.06 | 73.4 | |
| GPT-5.5Interface=Codex CLI, Protocol=diagnostic, State Conditioning=Oracle2026.06 | 64.8 | |
| Gemini 2.5 FlashState Conditioning=Oracle2026.06 | 63.8 | |
| Claude Sonnet 4.6State Conditioning=Oracle2026.06 | 56.9 | |
| GPT-4oState Conditioning=Oracle2026.06 | 50.3 | |
| Random BaselineState Conditioning=Oracle2026.06 | 46.6 | |
| State-Outcome ModelState Conditioning=Oracle2026.06 | 25.9 | |
| Qwen2.5-VL-7BProtocol=zero-shot, State Conditioning=Oracle2026.06 | 14.2 |