LLM Repair on LLM Repair Evaluation Set (n=200, seed=42) (test)
62.5Recovery RateWM-SAR
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| WM-SAR2026.07 | 62.5 | 1,424 | 1,389 | 5.4 | 60 | |
| Oracle-Region2026.07 | 62.5 | 475 | 402 | 5.8 | 60.1 | |
| LLMRepair-Full-Plan2026.07 | 62.5 | 10,857 | 10,333 | 5.8 | 62.6 | |
| TraceScan-Full-Point2026.07 | 62 | 21,714 | 20,623 | 0.8 | 47 | |
| Top-B-NodesK=32026.07 | 62 | 600 | 600 | 0.8 | 47 | |
| TraceScan-w4-PointWindow size=42026.07 | 37 | 4,800 | 4,800 | 0 | 34.4 | |
| Top-B-EdgesK=32026.07 | 32 | 600 | 600 | 0 | 28.7 | |
| PageRank-Subgraph2026.07 | 14 | 518 | 532 | 0.4 | 23.2 | |
| k-hop-k2k=22026.07 | 13.5 | 550 | 570 | 0 | 26 | |
| Uncertainty-Subgraph2026.07 | 4.5 | 317 | 328 | 0 | 6.4 | |
| TraceScan-w2-PointWindow size=22026.07 | 0 | 1,200 | — | 0 | 16.4 | |
| TraceScan-w1-PointWindow size=12026.07 | 0 | 275 | — | 0 | 7.1 | |
| LastError-Point2026.07 | 0 | 50 | — | 0 | 6.9 | |
| FirstFailedCall-Point2026.07 | 0 | 50 | — | 0 | 0 |