The stochastic or non-stochastic nature of the tractography algorithm, and the SIFT(2) model, are reasonably independent. The search space available to those algorithms will be greatly reduced, as there will be far less variability between streamlines and therefore less unique vectors along which the cost function can be reduced. But I don’t think it’s a technically incorrect thing to do; or perhaps alternatively, the non-sensibility of running SIFT(2) on deterministic tractography data is maybe less than that of using deterministic tractography in the first place
I take your point on the deterministic tracking, I was more interested in using it when raising the FOD to a higher power (eg 3) I tend to find that the resulting streamlines are much nicer that way.