I did see your other comment and I do see what you are saying of "why" it happens to work that way, no argument there, but I am not accepting that that is what "should" happen. My contention is simply that either it must be so guaranteed in the documentation to work that way, now and forever, OR, pragma_reverse_unordered_selects must catch this case and reverse it also. We are very much using the reverse_unordered_selects as part of pre-production testing because I thought that if it works with that pragma enabled, then it works correctly, no exceptions (and I do not think my belief is misplaced). Now we are seeing a case where that simply may not be, a query order may change in future for a query that is going into production and we have no way of knowing now if all our SQL is good and ordered and will remain so for days to come. I mean, if it only works "sometimes", what is the point of having that pragma?