I faced such a problem: after a reload of a model numbers in my table are changing (without changing a script or initial data). Each reload provides different numbers, but deviation is in range +- 0.05 (seems like something happens with round()).
I am loading data from 2 tables on SQL server and Qlik is joining them on 1 field:
SQL
select Key, debit, credit, ...., .... from X;
SQL
select Key, mapping and configuration fields, ... from Y;
Debit and credit are in decimal(n, 2) format. Number of records is about 10 mil.
What could be the problem? I am really going mad how predefined algorithm on a same data provides different result.