So I'm working through the learning modules online, and in particular was just watching "Additional Filtering Topics" which is part of the Visual Analytics topic. I notice (which is what they want us to see) in their example called ">$10k Sales", which puts SUM(Sales) on the Columns shelf and CustomerID on the Rows shelf, sorting Customer ID in descending order, that if I filter on an aggregated measure, say "SUM(Sales)" and filter for SUM(Sales) >= $10000 instead of filtering on CustomerID and using a "Condition" by field setting Sales Sum >= 10,000 and then if I add Category to the color shelf, that the viz decides to break a Customer ID's Sum of Sales down by Category, and so the filter I end up with is Sum of Sales for each category for each Customer ID. They do this to show us that if what we want is to see a customer's sales greater than $10,000 for all categories, the appropriate way to design the filter is to filter on CustomerID and use the condition by field setting Sales Sum >= 10,000.
MY QUESTION IS: how does Tableau decide what marks to show in the viz? Is it a set calculation or based mostly on a proprietary algorithm to "make things look nice based on best practices"? Because in the first example of setting the filter on Sum(Sales), if instead I use the exact same setup but instead filter at a lesser sum of sales amount (say, Sum(Sales) >= $3000), I can get the viz to show me Customer IDs with Sum(Sales)>$3000 in either a single category or by combining two categories! If I reduce the filter to Sum(Sales)>=$1000 then the viz even shows me CustomerIDs with Sum(Sales)>$1000 in either one or combining two even up to three categories! What is going on here?... what exactly is Tableau showing me, because adding Category to the color shelf is NOT showing me CustomerIDs with Sum(Sales) > x in a single category like the learning video claims. The video says, "Tableau ran each mark past the filter" -- that makes sense to me, but how do I know what marks Tableau would be creating prior to running them past the filter, since there's not consistency here -- is this an arcane algorithm that is unknowable except to those who work at Tableau, or is there a simple computation I'm missing?