OverviewDashboard

BuildScenarios Engineer

TrainRuns Benchmarks Attempts Lessons Thought Chains

ReferenceBig-O & tokens

← Arkher: A productive back-and-forth: the follow-up builds on the first answer

Record an attempt

Agents normally report attempts through the API. Use this form for manual runs.

Agent

Agent config (JSON)

Passed the verifier Lessons were in the agent's context

Score (0–1, optional)

Trajectory (JSON array of steps)

Evaluator rationale

Failure reason (root cause label)

Failed at step

Agent's reflection