What we learned building a quality measurement system for a multi-agent AI, drawing on practitioner wisdom from Hamel Husain, Eugene Yan, Braintrust, and applied-llms.org.