dashboard-authoring

Loaded when a dashboard is being proposed or published. The input is a SQL template (literals stripped to typed slots) and its catalog grounding; the output is a definition the publisher renders. No figure from any person's run enters the definition.

1. Classify the template's columns

• dimensions: every column in GROUP BY (and any non-aggregated select column);
• measures: every aggregated select expression;
• slots: every literal placeholder — the column it filters, its type, and its cardinality bucket.

2. Name everything from the catalog

• a measure's name is the glossary term or metric it was grounded in; a dimension's name is the column's displayName (fall back to the column name);
• the title is "<Measure>[ and <Measure>] by <Dimension>[, <Dimension>]", plus ", filtered by <Column>" per slot;
• if any measure or dimension has no term and no display name, keep the raw name and mark the definition title-quality: degraded, naming the column. That mark is a catalog finding and must survive into the published item's description.

3. Translate the measure

Each aggregated SQL expression becomes one DAX measure over the same columns, preserving the filter in the SQL WHERE as a CALCULATE(..., FILTER) when the filter is part of the definition (it came from a metric expression) and as a slicer when it is a slot. A slot never gets a default value.

4. Keep it verifiable

Emit, next to the definition, the SQL that the publisher will run to check the measure: the original template with slots bound to the dashboard's slicer defaults (none → no filter). The publisher fails the publish when DAX and SQL disagree.

5. Lineage

Record the source tables by their catalog FQNs so the publisher can attach the dashboard to them in OpenMetadata.