Calibration
Decision Index Language → When2Call (Decision Index)
When2Call (Decision Index) is estimated from Decision Index Language with a Michaelis–Menten curve fitted on 11 models measured on both: y = 1.7269·x / (0.60203 + x), R² = 0.52, cross-validated error 9.4 pp. It is used for 1 estimate.
| Estimated model | Decision Index Language | When2Call (Decision Index) | Source |
|---|---|---|---|
| Tev1-0.8B-experimental | 14.4% | 33.3% | estimated ± 9.4 pp, low confidence |
Fitted on
| Model | Decision Index Language | When2Call (Decision Index) |
|---|---|---|
| Clef | 47.8% | 71.0% |
| Clef-Flash | 40.5% | 64.4% |
| Clef-Omni | 40.7% | 62.1% |
| Drex Max | 62.1% | 94.0% |
| Drex v1.5 | 40.8% | 84.8% |
| Drex v2.0 | 52.7% | 86.2% |
| elephant-1 | 62.1% | 94.8% |
| GLiDE v1.2 | 66.8% | 94.4% |
| Jev 1.14 | 58.0% | 80.8% |
| Perplexity Decider Flash v1.2 4B | 62.2% | 74.4% |
| Perplexity Decider v1.2 27B | 65.5% | 87.7% |