The true capabilities of a model come from moving the precision-recall curve upwards by improving both precision and recall. This allows for more effective trade-offs between false positives and false negatives.