Glossary · Evaluator mechanisms
Value model
Estimates expected future return from a state or prefix, conditional on a policy. Rollout-labeled "step quality" usually behaves like one.
Defined in
Handbook · InstrumentSignals and resolutionVerdict, score, preference or critique; about a step, action, artifact, trajectory or population. Why PRMs, value models and margins are different objects.Where it comes up
- Vocabulary without category errorsHandbook · Foundations
- Mechanism familiesHandbook · Instrument
- Evaluators inside training loopsHandbook · Regime
- E: Reasoning-model trainingHandbook · Case studies
- Notation and formula cardHandbook · Reference