If you change the vector ordering you need to change back the answer ordering though 😅

We take the output of canonicalize(Tunable(), ...) and the solve calls through the usual probmap and probpmap, which will order accordingly, and we AD through.

    du_ = [1.0, 1.0, 1.0, 0.0]
    du = [du_ for _ in sol[[D(x), x, y, z]]]

the order of this is not based on the tunable order, it's based on the sort order of u. That's why it broke. I am saying if you fix the order by checking it symbolically, then you should change the du_ being tested to the order of the symbolic index.

I had pushed 470880e to fix that.

There seems to be a new failure in SciMLSensitivity/Core1 that I haven't seen earlier. But that is independent of this PR, since this only touches different files, from a different test set

@ChrisRackauckas solution_interface.jl and modelingtoolkit_remake.jl seem to have non AD related changes, so I was curious if there were any missing changes in this PR that you had in mind.

@AayushSabharwal can take it separately from here.