Skip to main content
SourceMedium supports multiple “levels” of querying, depending on what you’re trying to do:
  • Report tables (rpt_*): fastest for common dashboard-style questions
  • One Big Tables (obt_*): flexible analysis with pre-modeled joins and business logic
  • Semantic metrics: a single “metric definition layer” so the same metrics power dashboards and saved queries
This page focuses on the semantic metric layer and how it connects to:

​
Metric definitions (single source of truth)

Metric definitions are documented in: In the dbt project, semantic layer definitions live under models/semantic/ (semantic models + metrics + saved queries).

​
When to use semantic metrics vs raw tables

Use semantic metrics when you need:
  • Consistent metric logic across tools (dashboards, ad-hoc analysis, exports)
  • Standardized naming (no “two versions” of CAC/ROAS/net revenue)
  • The ability to group metrics by dimensions without re-writing business logic
Use raw tables (OBT/RPT) when you need:
  • Custom fields not covered by metrics
  • Deep row-level inspection and debugging
The semantic layer includes a set of saved queries designed to match common “dashboard tile” questions. If you’re using the Query Library, these saved queries also map cleanly to sections/recipes.
“Dashboard tile” mappings above are intended as practical starting points, not hard guarantees. Not every brand has every integration, and coverage can vary by source system.

​
How this relates to the SQL Query Library

If you prefer writing SQL directly in BigQuery, the Query Library recipes are pre-built around the same core concepts:
  • Valid order filtering (is_order_sm_valid = TRUE)
  • Stable “analysis tables” (obt_* and rpt_*)
  • Lowercased categorical groupings for safe aggregation
Start here:

​
(Optional) Querying via dbt Semantic Layer

If you have dbt Semantic Layer tooling enabled (e.g., MetricFlow), you can query metrics using a CLI interface instead of writing SQL.
If you don’t have Semantic Layer tooling enabled, use the SQL Query Library + OBT/RPT tables instead.