Yeah, dbt snapshots do a row hash and update if anything in the row changes. We had a source table that had a bunch of daily changing timestamps, e.g. "load date", that we needed to ignore, and focus on a business key. Dbt was an utter torment to try get this going. Ended up building a more elegant framework without it.

So apparently the check_cols parameter didn't help either?

I'm trying to learn about the critical pain points of dbt, and this case seems interesting.