Standard Skill: Interview
Interview is a Standard Skill: it fills the Semantic Layer with what only you know, such as what a word means, how a measure is counted, or which rows to leave out. The agent loads it after Schema Bootstrap or Model a Data Source left open questions, or when you want to teach it what your data means. To pick it yourself, type / and choose interview.
What the agent does
Section titled “What the agent does”- It gathers the open questions: those of the Conversation first, then gaps it finds in the Data Source’s entries, such as columns without a description, codes nobody explained or Rules that contradict each other. It doesn’t ask what the schema or a query can tell it; it looks that up. It tells you how many questions there are, and asks first what other answers depend on.
- It asks one decision at a time. Each question comes with the evidence it found and up to six answers, the one it recommends first, marked (recommended), with the reason. You can always answer in your own words. If you don’t know, it asks who would, notes it as an open question and goes on. Where a query can check your answer, it checks it and tells you if the data disagrees.
- It writes the entries: a Term for what a word means, a Rule for how the data must be read, a Benchmark for a known correct result, a Verified Query for a question and the query that answers it, and descriptions, keys, joins or measures in the Data Model. Before it writes a Verified Query, it runs the query and shows you the result, and writes the entry only after you confirm the result is right. It changes an entry that exists rather than adding a second one about the same thing.
- It proposes them together, after every few decisions or when you want to stop: one Context Change Request with one change per entry, at most 50, whose explanation says what was decided with whom and which questions are still open.