Unlock: Reward Models and Verifiers

How preference reward models, outcome verifiers, process reward models, executable checks, and ensembles provide different training signals, and where Goodhart pressure enters.