Odd DISTINCT behavior from lakehouse SQL Analytics Endpoint

Hello - this is a bit of an involved issue (apologies for length!) that started in PowerBI but seems to be really a SQL Analytics Endpoint oddity. I was creating a very simple bar chart report with a very simple DirectQuery data model - 3 tables coming from our lakehouse SQL Analytics Endpoint - and found that, for the Y-axis, if I asked for a COUNT, it worked fine; but if I asked for a COUNT DISTINCT, I got a very nondescript "SQL command could not be executed" error - but only for certain selections in my "Keyword" slicer. I grabbed the diagnostic output, found the query that PowerBI had constructed, and popped it into a live SQL query against the endpoint, and got exactly the same result - "Msg 100000 SQL command could not be executed." I've put the query below, whittled down so it only has relevant fields, and noted a couple of things I found that got around this mystery error, but none of them help me with my COUNT DISTINCT reporting requirement. Hoping someone has some insight!

 SELECT
    basetable0.c55,
    (COUNT_BIG(DISTINCT a0) + MAX(CASE WHEN a0 IS NULL THEN 1 ELSE 0 END)) AS a0
FROM 
(
    (
        SELECT
            t4.PerformanceNo AS c43,
            t4.Facility AS c55,
            t3.SeatNo AS a0
        FROM 
        (
            (
                select
                    MyTable.PerformanceNo as PerformanceNo,
                    MyTable.SeatNo as SeatNo
                from
                    dbo.us_performance_seats as MyTable
            ) AS t3
            INNER JOIN 
            (
                select
                    MyTable.PerformanceNo as PerformanceNo,
                    MyTable.Facility as Facility
                from
                    dbo.us_performances as MyTable
            ) AS t4
            on t3.PerformanceNo = t4.PerformanceNo
        )
    ) AS basetable0
    INNER JOIN 
    (
        SELECT
            t4.PerformanceNo AS c43,
            t4.Facility AS c55
        FROM 
        (
            (
                select
                    MyTable.PerformanceNo as PerformanceNo,
                    MyTable.Keyword as Keyword
                from
                    dbo.us_performance_keywords as MyTable
            ) AS t0
            INNER JOIN 
            (   
                select
                    MyTable.PerformanceNo as PerformanceNo,
                    MyTable.Facility as Facility
                from
                    dbo.us_performances as MyTable
            ) AS t4
            on t0.PerformanceNo = t4.PerformanceNo
        )
        WHERE t0.Keyword = 'Keyword1'
        GROUP BY t4.PerformanceNo,t4.Facility
    ) AS semijoin1
    on 
    (semijoin1.c43 = basetable0.c43 OR semijoin1.c43 IS NULL AND basetable0.c43 IS NULL)
    AND 
    (semijoin1.c55 = basetable0.c55 OR semijoin1.c55 IS NULL AND basetable0.c55 IS NULL)
)

GROUP BY basetable0.c55

Things that made the error go away:

• removing the DISTINCT from the outer SELECT (same behavior as in PowerBI)
• changing the "Keyword1" filter to "Keyword2" (same behavior as in PowerBI) - I checked out the raw results prior to the GROUP BY, and Keyword1 doesn't have a huge number of rows more than Keyword2 - like 26K instead of 13K - so it doesn't feel like a data volume issue to me
• taking out the MAX(CASE) statement for [a0] in the outer SELECT - instantly returned expected results
• adding a TOP N clause to the SELECT for [basetable0] - tried numbers as big as 10M, didn't matter how large or small - as long as there was a TOP N, it was happy and returned the expected results immediately (adding TOP N to the SELECT for [semijoin1] had no effect)
• removing the very bottom null-join-check (OR semijoin1.c55 IS NULL and basetable0.c55 IS NULL) also produced instant correct results - no other parts of the join had any effect when removed - and note that c55, the Facility column, does not contain any nulls