rlxp-analyze-run

To analyze a completed reinforcement learning run, compare primary and guardrail metrics against an incumbent baseline to classify the run as accept, reject, inconclusive, or debug based on evidence.

Baseline comparison in reinforcement learning experiment validation evaluates primary, guardrail, and diagnostic metrics against the current incumbent to produce an evidence-based decision rather than relying on guesswork.

Diagnose failure modes in a finished RL training job by validating the active contract, extracting approved metrics, and recording whether the outcome is reject, inconclusive, or debug.

You need reward reflection for a reinforcement learning run when analyzing reward-focused studies to detect signals like saturation, domination, or suspected reward hacking that impact the final decision.

Yes, you can use this approach to update the run evidence trail after training by generating structured outputs including metrics summaries, analysis notes, and ledger updates for the study.

Automated run analysis for RL experiments requires contract validation and approved metric extraction before classifying outcomes, meaning it cannot produce valid decisions if these prerequisites are missing.