semantic-view-patterns

Add a metric with the correct aggregate expression, such as COUNT(DISTINCT customer_id), using the metric optimization pattern. Apply it through semantic_view_set.py with a create operation, then validate that generated SQL uses the new metric.

First verify at least one table has a primary key on the join columns, then validate the join with relationship_creation.py. If it succeeds, apply the relationship via semantic_view_set.py with the generated name, columns, cardinality, and join type.

No. Dimensions and facts are physical database columns and cannot be added through this Skill. You can only enhance their metadata with descriptions, synonyms, sample values, and the unique flag.

Use custom instructions only as a last resort after dimensions, metrics, relationships, and filters cannot express the logic. Keep them generalized rather than query-specific, and prefer module_custom_instructions when neither type exists yet.

Validation often fails when neither table has a primary key, producing a rejected many_to_many relationship, or when reflect_semantic_model rejects the join. If validation fails, remove the relationship and use custom instructions instead.