Accuracy of FORESTFLOW across the input parameter space estimated via leave-simulation-out (top panels) and leave-redshift-out tests (bottom panels). Top panels. Each leave-simulation-out test involves training one independent emulator with measurements from 29 distinct simulations, and then using the measurements from the remaining simulation as the validation set. Lines and shaded areas show the average and standard deviation of 30 leave-simulation-out tests, and each panel shows the results for a different redshift. Bottom panels. Leave-redshift-out tests require optimizing one emulator with all measurements but the ones at a particular redshift, and then using measurements from this redshift as validation. Each panel shows the results of a different test.