No idea.

now the test suite checks the error. should be easier

It is not easier, Marco. The test suite is filled with noise, and there is no tip on what method to use. Moreover, the last test case goes against what available data provides (which is always monotonically increasing). Even with machine learning, we would not guess the next value using partial and noisy data. If we are trying to classify an animal, but all data is noisy and comes from the animal's tail, it is almost impossible to guess what the animal is.
For any new challengers, know that it is a polynomial that you are looking for and it is NOT a line. All test cases have noise, but they all refer to the same polynomial (an even function). The results were taken from the noiseless polynomial, so there is no need to find an exact number (unless you can somehow discover the polynomial).