SF evaluates position- dependent as for the boundaries of NNUE to "classical" and of course the nets themselves should give helpful information about the WDL- chances. I'd say it's mainly depth of search and pruning of lesser probable lines that makes it less "agggressive" now and then where engines with a yet often enough too optimistic evaluation give higher numeric evals.
What has been removed was only UCI- parameter of manually adjustable contempt as far as I know regards