Thank you Keith for providing further explanation. So in my understanding, in the original query, if column v1 is NOT a PRIMARY KEY, then the floating point result 9223372036854775000.0 will not be converted back to an integer for table lookup, thus the comparison 9223372036854775807 >= 9223372036854775000.0 is accurate and always evaluates True. The loss of precision is only happening when v1 is PRIMARY KEY and the 9223372036854775000.0 would converted back to INTEGER. Also, thank you for the fix and the fix comment.