TrajectoryEvaluationRunner

Models & Types · AiDotNet.Agentic.SelfImproving

Runs an ITrajectoryEvaluator over trajectories, annotating each with its Reward and producing an aggregate EvaluationReport. This is the continuous-evaluation step that turns raw captured runs into a measurable quality signal.

For Beginners

Hand it a grader and a batch of recorded runs; it grades each one, writes the score back onto the run, and hands you a summary report. Run it before and after a change to measure whether the agents are improving.

Constructors

Constructor Summary
TrajectoryEvaluationRunner(ITrajectoryEvaluator evaluator) Initializes a new runner.

Methods

Method Summary
Task<EvaluationReport> EvaluateAsync(IReadOnlyList<AgentTrajectory> trajectories, double passThreshold, CancellationToken cancellationToken) Scores each trajectory (writing Reward) and returns an aggregate report.
Task<EvaluationReport> EvaluateStoreAsync(ITrajectoryStore store, double passThreshold, CancellationToken cancellationToken) Scores every trajectory in a store and returns an aggregate report.