This is difficult to say, because we know virtually nothing.
First, does the problem arise from the use of subtly different data? Far too often, someone sees a difference between what they claim to be the identical mathematics in two different environments, but in fact, they truncated the data to only a few decimal digits when they moved the data. I know, you would NEVER have done that. Or would you?
Next, is this a question of the use of different integration methods. yes, I know people think the integral is the integral. BUT IT ISN'T! ANY numerical integration ends up being just an approximation at some level. Cumtrapz uses a cumulative trapezoidal rule. Did the python tool use a lower order integration (perhaps rectangle rule?) or a higher order integration (perhaps something like a Simpson's rule?)
We don't know any of this. And we are not given that information. The difference LOOKS like a method difference to my eyes, with a gradually growing bias. But if you truncated the accelerations to only a few digits, moving all the accelerations in essentially one direction, that too might have a similar impact.
And of course, we dont even have any data, So there is nothing we can even compare. If you want a better answer, then you need to provide your data. At the very least then we could test it out, changing the method, or truncating the numbers, to see if we could replicate that behavior. Until then, its just an educated guess.
