malloy-phrase-detection

Write search_text as a short semantic description of the concept, classify it as dimension, measure, view, or source, and use one target per distinct concept. Time references and numeric ranges become dimension targets; aggregation metrics become measures.

Target the dimension the value lives on, such as the subscription tier or city dimension, then run execute_query to list that dimension's distinct values. Confirm the exact stored string before filtering, since data may store "Premium", "PREMIUM", or "NYC".

Both are dimension targets. Time references map to the relevant date field, and numeric ranges map to the numeric dimension; the comparison itself is applied later in the query, not matched as search text.

Create one target of each plausible type: one dimension target and one measure target. Ambiguous concepts could be either a categorical attribute or an aggregation metric, so covering both types ensures the match succeeds.

Use 3-8 words naming the entity and its business process, like "customer order history and line items". Exclude filter values, time ranges, aggregations, and the user's brand or vertical, since source summaries describe data structure.

There is no value-search target in v1; get_context searches model metadata, not stored categorical values. Literal values must be resolved by targeting the parent dimension and querying its distinct values with execute_query.