benchgap
Calibration

Harvey LAB → AutomationBench

AutomationBench is estimated from Harvey LAB with a linear curve fitted on 8 models measured on both: y = 4.9568·x + -4.0288, R² = 0.77, cross-validated error 13.0 pp. It is used for 2 estimates.

Estimated modelHarvey LABAutomationBenchSource
Claude Fable 5.1 (xhigh with fallback)93.3%59.6%estimated ± 13.0 pp, low confidence
Claude Fable 5.1 (high with fallback)93.0%58.1%estimated ± 13.0 pp, low confidence