Incorrect data online vs desktop for larger resultsets

[cmorrow]

I have a report that displays some calculated measures in a matrix and a column chart.  The report works fine in the desktop.  The measures and totals calculate correctly.  However, when I publish the report to the service, the report displays incorrect results for the same filters.  If I narrow down the date range (and therefore the resultset) enough, the results look good.  However, for larger resultsets the results online are extremely off.  I've compared the query being executed in the database and it's the same for both versions.  It's almost like the service is truncating/limiting the data being returned.  Is that possible? 

The report uses a combination of import and direct query.  We have a gateway configured, and there are no issues with the dataset refresh.

[GilbertQ]

Hi there

If the reports are different, almost every time there is either a filter online that is being applied.

Or possibly there is a change made when importing the data that could affect it.

What I would suggest is to create a new page where there are NO filters online and see if you can get the same results as in Power BI Desktop.

If you can, then at least you know that there is a filter causing the numbers to appear to be incorrect.

[cmorrow]

Gilbert -

Thank you for your reply, but I have checked the filters.  In fact, I set up the filters in my desktop report, save it, publish it to the service as a new report and run it in both locations.  I would expect the reports to be exactly the same at that point, would you agree? 

I ran a series of tests, changing nothing but the date range report filter (using the relative date, last X days).  For each test, I recorded the total value of the matrix.  I'm including the results here.  As you can see, the results sync up exactly, and then they diverge.  Again, I'm not touching any other filters.

[GilbertQ]

Hi there

That is indeed strange, in all the datasets that I have worked with over time I have never seen the dataset being wrong.

It is sometimes persistent filters that have been applied.
Or DAX measures
Or Date tables that are different
Or underlying data.