This should not pass.

atleast python: no random tests

python new test framework is required. updated in this fork

The new testing framework should be used.

Now for some randomly generated B and T. These tests are designed to pass if percentages are correct to within 0.1%.

This form of testing should really be done to fixed tests as well since the normalization constant for the probability (aka the sum of all the probabilities) is related to the Jacobi theta function by a constant, which is not easily computable and at best we're estimating its actual value by summating enough states. The normalization constant will always be under-estimated and hence getting a larger result than the actual value.

You could try to make the time step adaptive, i.e. ask the user to figure it out under the condition that output be within 1% of the correct answer.Obviously, you'd have to write the tests in a way that they accept answers within a particular range. Is it possible to do that?

for python translation need more random tests.

The description has to mention that the strings should be joined and then filtered, but not filtered and then joined.

I feel the reference to the Euler method for integrating ordinary differential equations is potentially confusing and somewhat irrelevant. The Wikipedia page leads people down a whole procedure and proof based around ODEs. As we're just numerically approximating the integral of a polynomial, I would just say we're using the rectangle method or midpoint method, drop the mention of Euler entirely, and only keep the second link.

I suggest to clarify in the description that only the y = ... function will be tested in the kata.

Furthermore, you could increase the difficulty of the kata a little by adding a start input.