Just take a look on the field-values itself within a table-box. A chart with calculations isn't suitable to find reasons behind unexpected results. Very helpful is also to include recno() and rowno() within the loads to show any duplicates in the data-set as well as being able to track source- and target-records.
Comparing both approaches against each other in this way should show the differences and hinting to the causes. As reason in your case I could imagine a different interpretation of the values. Qlik uses no data-types else an interpretation as numbers, strings or a mix of them in regard to the interpretation-variables and the order in which the field-values are loaded. This may result that some or all values are now strings and not numbers and/or that thousands/decimal-delimiters aren't wrongly interpreted or similar stuff.
In the most scenarios this worked smoothly and if not there are various resolving-approaches which will be depending on the data. Very common is to apply an own interpretation with (max.) something like this:
num(num#(text(MyField), 'FormatPattern'), 'FormatPattern') as MyField
Another way would be to pre-load the wanted interpretation like:
