dbt Analytics Engineering Certification: Leveraging the dbt State Practice Questions

State comparison requires a baseline manifest.json in the directory specified by --state. If that directory is empty or contains a stale/missing manifest, dbt cannot compare state and defaults to selecting all models. A dbt version mismatch can cause hash differences but typically surfaces as mismatches rather than full selection. --defer is separate from state selection and its absence does not cause all models to be selected. dbt compile does produce a valid manifest.json that can be used for state comparison — the artifact format is the same.

The + prefix on a model selector means 'include all upstream ancestors.' Running +fct_orders selects fct_orders plus every model it depends on, traversing up the DAG. A suffix + (e.g., fct_orders+) selects downstream descendants. Selecting only fct_orders would use no + at all. The --select flag does not filter by directory location by default.

TRUE — ref('dim_product') resolves to the latest version by default; consumers can pin a specific version with ref('dim_product', version=1). TRUE — deprecation_date is a first-class property on versioned models that communicates to downstream consumers when a version will be removed, surfaced in dbt docs and logs. TRUE — each model version is an independent node in the DAG with its own compiled SQL, tests, and run results. FALSE — model versions can have different materializations; there is no such constraint. FALSE — contracts are attached at the versioned model level and can differ, but versions are expected to share the contract defined on the parent model config; they do not arbitrarily have completely independent schemas by default (though advanced setups allow variation). FALSE about versions being restricted to table/incremental — all materializations are supported.

The correct approach is to use manifest.json → nodes.<unique_id>.depends_on.nodes to get the list of upstream dependencies for each model, and run_results.json → results[].unique_id as the matching key to look up each node's execution status. unique_id is the stable identifier shared across both artifacts, making it the correct join key. compiled_sql is query text, not an identifier. execution_time is a metric, not a join key. path is a file path that is not guaranteed to be unique across packages and is not the standard join key.

run_results.json captures runtime telemetry for each executed node. The execution_time field (in seconds) under each result object records how long that model took to run — exactly what is needed for performance analysis. materialized describes the materialization strategy, not runtime. rows_affected is a database response count, not execution duration. compiled_sql is the rendered SQL query, not a timing metric.