rlxp-validate-candidate

Reinforcement learning candidate validation prevents unsafe launches by verifying schema formats, contract scope, and safety constraints. It checks task identity, allowed files, metric invariance, and budget limits to ensure experiment safety before any launch decision is made.

Launch gating is a pre-launch review mechanism that rejects invalid or out-of-scope reinforcement learning candidates. It enforces lifecycle status, budget, and command resolution checks, blocking candidates that need additional probes or approved rollout conditions from proceeding.

Checking contract compliance for domain randomization changes requires validating candidates against active contract scope and safety rules. This verifies allowed-file checks, metric invariance, and task identity matching to prevent contract violations during reinforcement learning experiment runs.

Reinforcement learning candidates fail schema checks when they do not match the required format or correct task and study identifiers. Validation enforces these identity matches alongside budget and command resolution rules to prevent out-of-scope or invalid experiment launches.

Automated validation applies scope and safety rules to reward and curriculum changes in reinforcement learning experiments. The system verifies changed files, lifecycle status, and metric invariance to gate launches safely without manual contract review for every single iteration.