Qwen3-8B formal acceptance matrix (check_qwen3_ppo_rollout): not run on Hopper 4×H100.
Reason: published fixed-rollout FP32-reference thresholds / acceptance manifest not present in tree (see examples/post_training/tp_ppo/QWEN3_ACCEPTANCE.md).
Completed on this node: tiny-GPT pytest matrix + Megatron/DeepSpeed TP2 actor/critic smoke (see info.txt).