A telescope cloud-cover desk can either sit with a few carefully chosen sky plates or run a tool that produces a large number of variations. How should those two ways of performing adversarial testing be contrasted?

Sitting with a few carefully chosen sky plates versus running a tool that produces a large number of variations is manual crafting of specific adversarial examples versus automated generation of many variations. That is tester method only, no payload. Relabeling it as white-box versus black-box, as drift types, or as A/B versus back-to-back misses that contrast.