Revision-cycle prediction on FIRSTPASS n=318 (test)
78.2F1 MacroFIRSTPASS
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| FIRSTPASSloss_masking=true, adaptation=LoRA2026.06 | 78.2 | 71 | — | 80.5 | |
| Qwen2.5-7Bprompting_strategy=zero-shot2026.06 | 73.3 | 63.3 | 0.185 | 77 | |
| Gemini-3.1-flash-lite-previewprompting_strategy=zero-shot2026.06 | 56.3 | 31.7 | 0.001 | 70.1 | |
| Qwen2.5-7B + LoRAloss_masking=false, adaptation=LoRA2026.06 | 56.2 | 40.4 | 0.001 | 62 | |
| Llama-3-8Bprompting_strategy=zero-shot2026.06 | 48.3 | 16.4 | 0.001 | 67.9 | |
| Majority baselinestrategy=constant prediction (STANDARD)2026.06 | 39.5 | 0 | 0.001 | 65.4 | |
| Qwen2.5-7Bprompting_strategy=5-shot2026.06 | 39.5 | 0 | 0.001 | 65.4 | |
| DeepSeek-R1-7Bprompting_strategy=zero-shot2026.06 | 39.5 | 0 | 0.001 | 65.4 |