Suspected BUG when using ALL with multiple non-ignored slicer values

Marco Russo said he will write about this and here it is! :smileyhappy:

https://www.sqlbi.com/articles/understanding-dax-auto-exist/

“The golden rule of data modeling is always the same: always use star schemas. If a column has to be used to slice and dice, then it needs to belong to a dimension. Numbers to aggregate, on the other hand, are stored in fact tables. Tabular lets a developer deviate from the regular star schema architecture. This does not mean that doing it is always a good idea. It seldom is.”

Thanks marcorusso ! :smileyhappy:

UPDATE:

Thanks AlbertoFerrari ! :smileyhappy:

Just noticed the above article was actually written by you.

Sean just getting up to speed with this thread. Given the answers already I may not have much to add, but let me digest first :)

I confirm this seems a bug in DAX.

In the following repro (execute the DAX code below connecting DAX Studio to the PBIX file), the first EVALUATE is the one generated by Power BI using SUMMARIZECOLUMNS.

The second one uses a CALCULATETABLE applying the same filters.

It could be similar to another bug I've seen last week and that MS is already fixing, but it could be also a different one.

I suggest you opening a support ticket with Microsoft to prioritize a fix for it. 

DEFINE
    VAR __DS0FilterTable =
        TREATAS ( { "B" }, 'Students'[Section] )
    VAR __DS0FilterTable2 =
        TREATAS ( { "Future", "Unenrolled" }, 'Students'[Status] )
    MEASURE Students[Count Students (Total All Sections)] =
        CALCULATE (
            CONCATENATEX ( DISTINCT ( Students[Student] ), Students[Student], "," ),
            ALL ( Students[Section] )
        )
EVALUATE
SUMMARIZECOLUMNS (
    __DS0FilterTable,
    __DS0FilterTable2,
    "Count_Students__Total_All_Sections_", IGNORE ( 'Students'[Count Students (Total All Sections)] )
)
EVALUATE
CALCULATETABLE (
    { [Count Students (Total All Sections)] },
    __DS0FilterTable,
    __DS0FilterTable2
)

 EDIT: this behavior is by design, so it will be not fixed, despite it is confusing. The reason is that manipulating column filters in the table that contains also the measures creates a counterintuive behavior. The column filters are applied to the table by SUMMARIZECOLUMNS (used by Power BI) which optimizes this by creating a filter over the existing filtered combinations of the columns. Once you remove the filter on a single column, SUMMARIZECOLUMNS no longer restores the rows that were initially filtered. This behavior affects only SUMMARIZECOLUMNS and not CALCULATETABLE (in my second example).

The best practice is moving the filters in other tables - once again, the star schema is the best design, because you should never modify the filters on the fact table.

I will write an article about SUMMARIZECOLUMNS describing these behaviors - there are also other side effects that can affect certain measures in more complex scenarios, I didn't realize in my initial answer that the issue was another side effect of the internal implementation of SUMMARIZECOLUMNS.

Looking at this a little more, here's the real issue, you disconnected your two slicers for Section and Status. So, when you select B, your available Status values actually goes to 1, which is Future. Your slicer does not display this because you disconnected the interaction between them. But, this filter is still there under the hood in terms of DAX.