Formal Verification Repair on VeruSAGE-Bench
79.2Success Rate (IronKV)Claude Sonnet 4
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Claude Sonnet 4type=hands-off agent2026.07 | 79.2 | 66.7 | 70 | 46.3 | 46.2 | 59.4 | |
| GPT-5type=hands-off agent2026.07 | 75 | 55.6 | 90 | 31.7 | 23.1 | 50 | |
| o4-mini (STRACE-enhanced)enhancement=STRACE2026.07 | 62.5 | 88.9 | 100 | 31.7 | 61.5 | 58.5 |