The point about gpus and workflow is interesting. In practice I would keep the first test simple, record what worked, and only add extra complexity if it solves a specific limitation. The original point — "For GPUs, it helps to think in terms of inputs, constraints, verification, and iteration...