Nutpie math over different dimensions in samples

I created a model with random variables that have various combinations of three coordinates. In one step I do a subtraction.
In any prior sampling or sampling with nut_sampler=‘pymc’ the values make sense. (e.g., the aircraft’s tanks drain in sequential order over the duration of the flight).

When I use nut_sampler=‘nutpie’… the subtraction is along the wrong dimension (e.g., for 5 input flights of quad-tank-aircraft, the 4 tanks of the first aircraft and the first of the second drain first )

Can you provide a short example illustrating the issue you are seeing?