data-analytics-engineering

Define semantic models in YAML with entities, measures, and dimensions, then declare metrics as simple, derived, cumulative, or ratio types using MetricFlow syntax. Assign owners, apply semantic versioning, and certify metrics after validating values against a source of truth.

Choose dbt for its large ecosystem and dbt Cloud features, SQLMesh for stateful builds and compile-time validation on large DAGs, and Coalesce for visual development with enterprise governance. The Skill's comparison reference provides a full decision matrix for 2026.

Start with built-in dbt tests (unique, not_null, accepted_values, relationships) in YAML, add dbt-expectations for statistical checks, and layer Elementary for ML-based volume and freshness anomaly detection. Configure severity levels so critical failures block pipelines while minor issues only warn.

Use Cube when you need sub-second dashboard latency, pre-aggregation caching, or API-first embedded analytics across multiple data sources. dbt Semantic Layer fits better when you already use dbt and want metric definitions living next to transformation code.

This happens when metrics are defined independently in BI tools or ad-hoc SQL without a canonical definition. Fix it by centralizing the metric in a semantic layer, running side-by-side comparison tests, and deprecating duplicate definitions through a governed workflow.

dbt relies on manifests and incremental flags rather than built-in state tracking, so schema changes can trigger full refreshes and Jinja errors surface only at runtime. SQLMesh addresses this with Terraform-like state management and virtual dev environments.