Thanks @Ashwin_DSA , @amirabedhiafi for your swift responses.

I had high hopes when I saw lakeflowUtilityVersion_1_5() is queried, as I found the database user for the ingestion gateway connection (i.e. the @User parameter for both dbo.lakeflowSetupChangeTracking and dbo.lakeflowFixPermissions) DIDN'T have execute permission on that function (clearly an oversight in lakeflowFixPermissions which Databricks ought to address).

But after granting execute permission we are still getting exactly the same error.

And our setup meets all the other requirements you have mentioned between you.

The one thing that you both mention that doesn't make sense to me is the need for the trigger to "actually be writing rows" to the audit table. I realise that for DDL changes to be picked up this needs to be operational, but surely this can't be necessary for the ingestion to work at all. Isn't it perfectly normal/acceptable for the audit table to be empty? We are attempting to ingest a stable source that hasn't experienced any ALTER TABLE operations yet so the trigger won't have fired. Or am I missing something?