Finite-answer lead measurement on Quality-gated parser-clean replication dataset v1 (test)
100Parse Accuracyfinite-answer commitment
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| finite-answer commitmentCondition=Canonical, Sample Size (n)=962026.05 | 100 | 100 | 100 | 100 | 42.7 | 3.5 | |
| finite-answer commitmentCondition=Prompt shift, Sample Size (n)=962026.05 | 100 | 100 | 100 | 100 | 55.99 | 4.74 | |
| finite-answer commitmentCondition=Task-family shift, Sample Size (n)=642026.05 | 100 | 100 | 100 | 100 | 42.56 | 13.72 | |
| finite-answer commitmentCondition=Verbalizer shift, Sample Size (n)=962026.05 | 100 | 100 | 100 | 100 | 52.01 | 4.22 |