Check also the behavior on both serverless and classic compute - it is catalog federation so it can work like that and needs setup as you read only metadata from catalog and later databricks is just reading data from that metadata location