How do we measure whether this is working?

Adoption metrics answer whether people use the system, not whether it should be trusted. The measures that matter are whether answers are supported by retrieved sources, whether refusals fall in the categories policy intended, whether repeated questions get consistent answers, and what each answer costs.

All four come from the same surface in Centralpoint. The Interaction Log records what was retrieved, which skills governed the response and the token accounting; the governed cache makes variance and repeat cost directly observable. Pre-deployment measurement and ongoing monitoring use one instrument rather than two.