So,

Google's AI mode suggested using the Table.Buffer command for the source queries and I did this for the source queries in my flow and it seems to have worked.

I remain unimpressed - my queries were not (IMHO) complex and there was not a huge amount of data involved either (<3,000 rows and ~30 fields).

Buffer the Table (Advanced Workaround): In some advanced scenarios where sorting at the end doesn't work, using Table.Buffer in the M language can force Power Query to load the entire table into memory and maintain order, but this has performance implications. For most cases, a final explicit Sort step is sufficient.