Each row is returned in a hash reference. The columns are ordered according to how their names hash, which pseudo-random.

I am no DBI expert, but I suspect there will be another way to retrieve the results, which preserves the column order.