I think you have to distinguish dependent variables from independent. If all variables are independent, then the number of test cases is the max of the number

I think you have to distinguish dependent variables from independent. If all variables are independent, then the number of test cases is the max of the number of (relevant) values for each variable (i.e., you can combine cases). When a variable depends on another, you have combine them. But even then you can quantize the outcomes (i.e., which combinations result in different behaviors) and then partition the values to produce those combinations. If you don't know if variables are dependent, you have to do some stochastic testing and find associations to investigate for causality, distinguishing confounders, etc. -- assuming that's cheaper than just running a full gamut of combinations. Sometimes you can use grey-box indicators from the system itself to quantize behaviors, or define coverage as traversal through such indicated system states. So, yeah: great start, but more to it...