rlxp-dr-probe

To test reinforcement learning policy robustness, you plan controlled domain-randomization probes that break the parameter space into one-factor-at-a-time jobs to isolate which randomized settings degrade nominal performance.

A controlled domain-randomization probe is a targeted evaluation that systematically varies simulation parameters one at a time to map safe operating ranges for an incumbent reinforcement-learning policy before deployment.

Yes, you need local evaluation evidence, approved probe bounds, guardrail thresholds, and dry-run helper workflows, because this Skill plans probes conservatively and keeps planning separate from launching.

Your policy likely lacks robustness to randomized simulation variations, requiring a controlled probe to identify which specific noise parameters exceed your policy's tolerance and narrow the safe operating ranges.

The best way to isolate harmful randomization parameters is executing one-parameter-at-a-time probe jobs, which evaluate each domain randomization factor independently to pinpoint exact failure thresholds.

You should not use controlled domain-randomization probes if nominal performance is unacceptable, because these probes are designed for task-scoped robustness checks only after baseline performance is verified.