Thanks :D
The commutativity issue still persists. Either the description needs to be updated to specify the form of the result or all cases accepted.

Hi there I'm still failing on some random tests even tough the code should work. The proposed "correct" result can't be correct.
Example: diff (cos (* 52 x)) --> Expected: (* 3 (* 2 x)) or (* 6 x) which is incorrect

Edit: commutativity of multiplication is also still a problem
Edit2: Otherwise a really great Kata

If I create a buffer variable in global scope it sometimes gets carried over to the next assert statement and sometime does not.

lol

What other tags would this belong to?

Edit: How do you remove comments? :')

fixed that bug

Runge-Kutta3 Solver. Maybe this should be the new baseline for solutions.

Try incrementing the positions first then the velocities.
Edit: Either way works for me with your code (as it should with sufficiently small step sizes)

looking into it

I think the only thing left is adding a Runge Rutta 3 solution or something comparable and checking if this will work as well

ok I know I added something that should work yet i'm suprised it did

G is arbitrary
Choosing the mean (to be correct something similar with weights resulting by comparison with a tayler expansion) is used in different approaches. If both solvers are stable and use sufficiently small timesteps they will produce extremely close results. I just wanted to use the simplest method possible. By that philosophy I should probably increment the position first then the velocity (this would also be technically more correct)
can you check if the instability problems still persist?