Verifier
A program that scores whether an output is correct — the reward source in reinforcement learning from verifiable rewards.
A verifier is an automatic check that decides pass or fail for a model’s output: a unit test, a math-answer comparison, a schema validator. In RLVR it replaces a learned reward model, so training signal comes from ground truth rather than a gameable proxy. A sound verifier is critical — every false positive is a lie the model learns to exploit.
Worked example: a model or function that CHECKS whether a candidate answer is correct — running tests on generated code, checking a proof step, scoring against a rubric — separate from the generator that produced it. Gotcha: verification is often much easier than generation (checking a solution vs. finding one), the leverage behind generate-and-check loops and RL — but a weak or gameable verifier is worse than none, because the generator learns to satisfy the checker rather than the real goal (reward hacking).