Repository-level Program Repair on SWE-bench Lite (50 tasks)
41Validation Success CountContextSniper
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ContextSniperHost=Claude Code, Runs=50, Patch=502026.07 | 41 | 15 | 30 | 52.55 | 10.97 | 19.2 | 522 | -38.9 | |
| BaselineHost=Claude Code, Runs=50, Patch=502026.07 | 39 | 16 | 32 | 85.97 | 15.09 | 29.9 | 791 | — | |
| ContextSniperHost=OpenClaw, Runs=50, Patch=502026.07 | 33 | 12 | 24 | 32.91 | 0.404 | 17.2 | 461 | -51.5 | |
| BaselineHost=OpenClaw, Runs=50, Patch=502026.07 | 26 | 13 | 26 | 67.8 | 0.635 | 32.1 | 572 | — |