Terms we use

A semantic layer is a written map from the words your business uses to the tables, columns and filters that compute them.

A metric contract is one metric's written definition, its numerator, denominator, grain, filters and time window, which an AI analyst reads before it computes anything.

A golden set, also called a golden dataset, is a list of questions with verified answers that you run your AI analyst against to get a real accuracy number.

An AI judge is a model that grades another model's output against a written rubric, so you can score work with no single right number.

Agentic analytics means an AI analyst runs the full loop of an analysis, from framing the question to querying, checking and writing it up, while people own the definitions and the decision.

Context engineering is deciding what an AI analyst reads before it works, writing it down, and testing whether each piece changes the answers.

An analysis trace is the record of how a number was produced: the query, tables, filters, definitions and assumptions, so someone else can rerun it.

The reliability spread is the range of answers you get when you ask an AI analyst the same question more than once.