Elementary does support some anomaly detection based on SQL expressions, but with important limitations depending on the test type: - For grouping or filtering data, you can use SQL expressions in the

dimensions

and

where_expression

parameters. For example, you can group anomaly detection by a SQL expression (not just a simple column) using the

dimensions

parameter in tests like

all_columns_anomalies

,

column_anomalies

, and

dimension_anomalies

. This allows you to segment your monitoring by any valid SQL expression, not just a single column name. For filtering,

where_expression

accepts any valid SQL expression to filter the data being tested [dimensions documentation](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/dimensions), [where_expression documentation](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/where-expression). - However, for monitoring a metric or anomaly directly on a SQL expression (e.g., the difference between two columns, or a calculated field that does not exist as a physical column), Elementary generally requires that the expression be materialized as a column in a model or view. You cannot directly specify an arbitrary SQL expression as the target column for anomaly detection in

column_anomalies

or similar tests. This is confirmed in community discussions, where users are advised to create a temporary model or view for such cases [Slack discussion](https://slack.com/archives/C02CTC89LAX/p1736408683199129). - There is an open feature request to allow setting an SQL expression as a timestamp column, but as of now, this is not natively supported [GitHub issue](https://github.com/elementary-data/elementary/issues/897). Summary: You can use SQL expressions for grouping (

dimensions

) and filtering (

where_expression

), but not as the direct subject of anomaly detection (i.e., you cannot monitor an arbitrary SQL expression as if it were a column, unless you materialize it first). If you need to monitor a calculated value, you should create a model or view with that calculation as a column, and then apply Elementary tests to it.