Good QC for RL data

Reinforcement learning training data requires standardized QC evaluation across performance-cost-latency tradeoffs, as most vendors fail multiple quality gates.