Multi-agent attribution on Synthetic Additive Benchmark
0.855MAEWho&When
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Who&Whentype=LLM-as-Judge, variant=prompt-based scorer2026.05 | 0.855 | 17.7 | 3.664 | 4 | |
| MASTtype=LLM-as-Judge, variant=prompt-based scorer2026.05 | 0.882 | 8 | 3.811 | 8 | |
| Oursmethod=path integral, K=302026.05 | 1.15 | 100 | 1.45 | 100 | |
| LOOdescription=leave-one-out2026.05 | 1.15 | 100 | 2.33 | 100 | |
| Sampled Shapleypermutation samples=2002026.05 | 1.15 | 100 | 6.79 | 100 | |
| Banzhafcoalition samples=2002026.05 | 1.15 | 100 | 6.89 | 100 |