Calibration
HealthBench Hard → HealthBench Professional
HealthBench Professional is estimated from HealthBench Hard with a offset logistic curve fitted on 8 models measured on both: y = 0.6063 + (0.0000 − 0.6063) / (1 + exp(−200.00·(x − 0.4077))), R² = 0.69, cross-validated error 6.7 pp. It is used for 0 estimates.