Answer out loud before looking back.
One. Why is an agent's report of success not evidence of success?
Two. What does a traceability matrix catch that a passing test suite does not?
Three. When is sampling across a data set the right call, and when is it a shortcut you should not take?
Four. What is the risk in a self-healing test, and what makes it visible?
Five. Why does requiring a citation constrain the model more than instructing it to stay on topic?