Dafny Program Verification on HumanEvalDafny (test)
97Verification Success Rate (NoDiff)SEVerA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SEVerAconstraints=NoDiff behavioral specification2026.03 | 97 | 97 | 0 | 18.2 | |
| DafnyBench baseline2026.03 | 86.9 | 87.9 | 4 | 16.1 | |
| SEVerA (w/o constraints)constraints=None2026.03 | 84.8 | 88.9 | 5.1 | 15.7 | |
| LLM (Claude Sonnet 4.5)Model=Claude Sonnet 4.52026.03 | 73.7 | 76.8 | 8.1 | 9.8 |