Multiple Choice

What is the only valid conclusion when every ML pipeline component is at human-level but the overall pipeline is significantly below human-level?