Calibration
MCP Atlas → AA AutomationBench
AA AutomationBench is estimated from MCP Atlas with a linear curve fitted on 5 models measured on both: y = 4.0854·x + -2.9312, R² = 0.88, cross-validated error 13.0 pp. It is used for 3 estimates.
| Estimated model | MCP Atlas | AA AutomationBench | Source |
|---|---|---|---|
| GLM-5 | 31.1% | 0.0% | estimated ± 13.0 pp, low confidence |
| LLaDA2.2-flash | 46.2% | 0.0% | estimated ± 13.0 pp, low confidence |
| Solar Open 2 | 58.2% | 0.0% | estimated ± 13.0 pp, low confidence |