Evaluation of different metrics as a function of the semantic threshold. Top panel: Evolution of different metrics (TPR - green, TNR - blue, PPV - purple, F₁-score - yellow & ACC - orange) measured employing the predictions of the semantic model as a function of the semantic threshold selected (solid lines); we also show the values measured for the corresponding metrics studying the differences between the baseline simulations (horizontal dashed lines). Bottom panel: Fraction of voxels predicted to be collapsed (equivalent to 1 − β) as a function of the semantic threshold employed (solid black line); the horizontal black dashed line corresponds to the fraction of particles that end up in DM haloes measured in the validation simulations. In both panels, the vertical black dotted line shows the semantic threshold we employ; this threshold has been selected to match the fraction of collapsed voxels.