Calibration
React Native Evals → AA-SciCode
AA-SciCode is estimated from React Native Evals with a Michaelis–Menten + offset curve fitted on 6 models measured on both: y = 0.0000 + 2.0000·x / (2.43313 + x), R² = 0.33, cross-validated error 8.6 pp. It is used for 4 estimates.
| Estimated model | React Native Evals | AA-SciCode | Source |
|---|---|---|---|
| Composer 2 | 96.1% | 56.6% | estimated ± 8.6 pp, low confidence |
| Composer 2 Fast | 94.9% | 56.1% | estimated ± 8.6 pp, low confidence |
| DeepSeek V3.2 | 71.5% | 45.4% | estimated ± 8.6 pp, low confidence |
| Grok 4 | 72.6% | 46.0% | estimated ± 8.6 pp, low confidence |