But that's the inner query. Does the outer query compensate, so that the actual results are in a consistent order? Or are the actual results nondeterministic as observable by running the program without inspecting the query SQL?

Actually whole column-alias pairs are shuffled.

So they are in different order each time but each column gets mapped to same x* number`

Outer query is always in fixed order x3...x93 which works since the column-alias mapping is preserved despite shuffling as I just found out.

The problem is the generated query looks different which is a problem for observability tools that try to aggregate them

Ok. That's a lot less surprising. I wasn't sure if you found an edge case where Slick is badly broken!

That would be good to improve. Would you be interested in doing a pair programming session, or perhaps would your company have interest in sponsoring work on this?

I'll try to find out what's possible although this is non-blocker so pretty low priority at the moment and might take a while