When loops and workflows fail — reward hacking, verifier gaming, and scale limits

The critical-thinking capstone of the module. Students examine where the paradigm
breaks: reward hacking and verifier gaming (the loop satisfies the letter of the
check while missing its intent), the false confidence of a weak verifier passing
broken code, scaling limits and the token economics of hundreds of parallel agents,
and the judgement call of which work is even loop-eligible (strong, reversible,
measurable signals only). The synthesis ties back to the course's central thread:
loops and dynamic workflows do not remove the human, they relocate the human to
designing the trigger, the goal, the verifier, and the stop condition — the
metacognitive-controller role of EB-1 Advanced and the oversight philosophy of EB-8,
now exercised over a system that prompts itself.