Agent Reliability

process-supervisiontrajectoriesevaluationgradingreasoning

Process vs outcome evaluation of agent trajectories

Two places to point the grader: the final state (outcome) or every intermediate step (process). Outcome grading is cheap, objective and gameable by lucky trajectories; step-level grading catches reasoning that arrives at the right answer wrong, at the cost of expensive labels. The research result: process supervision beat outcome supervision decisively on hard math. For agents, outcome grading gates releases while process signals explain failures and catch right-for-the-wrong-reason passes.

Why this wins its question: Assigns each grading mode a production role — outcome gates, process explains — instead of importing the research debate wholesale, and names the failure mode (right answer, wrong reasoning, lucky pass) that only step-level signals catch.

Claims

Every assertion below is bound to registered sources and carries its own confidence. Weight them; do not treat the page as uniformly authoritative.

  1. Process supervision — feedback on each intermediate reasoning step — significantly outperforms outcome supervision on the MATH dataset, with the process-supervised reward model reaching 78% on a representative test subset.

    confidence 0.9Let's Verify Step by Step · secondary

  2. Step-level evaluation has a public data foundation: PRM800K releases 800,000 step-level human feedback labels used to train the best process reward model.

    confidence 0.9Let's Verify Step by Step · secondary

  3. Outcome grading remains the production gate for tool-using agents: comparing final system state against an annotated goal is reproducible and narration-independent, which is what release decisions need.

    confidence 0.9tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains · secondary