I understand. Really - I do. I've had these "fights" with the CFO. The problem is - there's no such thing as "precision agnostic" in floating-point arithmetics: not dealing with the issue of precision means that those less aware than we think they're dealing in accurate numbers when we're not.

I am also observing the "differences" you describe - but I expect them since I was told they'd be there. They change because of the continuous "mini" rounding operations happening to certain numbers along the way. The rounding noise will trigger a little differently each time.

I'm merely addressing that you can't expect different behavior from what the definition tells you it's going to be.

The only way you will ever make the financial-types happy is to deal in "precise" data types: integers (all of them), and by extension - decimal/money (which are AFAIK stored as integers, with an extra piec of data stating where the decimal point is).