verify-environment

Validate a training environment by running pre-launch checks on validation rollouts, reward behavior, and baseline health to catch reward bugs and tool issues early before GPU runs. This confirms the setup is safe to launch.

A flat or erroring reward signal on a training scorecard indicates the reward function is not discriminating properly. You must analyze reward variance and error signals to determine if the environment baseline is weak or broken.

You can check rollout budgets and retrieval behavior for a RAG environment before launch by running a validation baseline. This pre-GPU evaluation confirms whether reward scoring and retrieval setups are correct.

Run environment validation immediately after editing training data, provider settings, or an environment. This pre-launch check ensures the updated baseline is healthy and ready before starting expensive GPU runs.

Baseline validation works with scaffolded agent tasks, RAG setups, trace-based environments, and custom training workflows. It interprets scorecards and rollout budgets to confirm any of these setups are safe to launch.