In Databricks, determining whether a materialized view is performing a full or incremental refresh typically depends on how the underlying table operations and refresh mechanisms are configured. From your observations, the execution times suggest that new rows may trigger incremental refreshes, while deletions might lead to full refreshes. Incremental refreshes usually depend on metadata tracking changes, while full refreshes rebuild the entire materialized view.

To confirm this behavior, you can monitor the query execution plan or logs during the refresh operation. Tools like the Databricks Query History or Apache Spark UI can help you inspect the operations. Check whether the materialized view refresh scans only updated rows or the entire table. Additionally, review the materialized view's settings or configurations, as some platforms offer explicit options to control the refresh type. If incremental refresh is not supported for delete operations, the system may default to a full refresh.

For further insights, you could enable logging for the underlying queries or run tests with verbose output to see exactly what data is being processed during the refresh. Comparing execution plans between different types of changes (insert, delete, update) can provide clarity.