The backend is Python/Flask with Numba for the heavy lifting.

The core math uses a modified Coefficient of Variation with an exponential decay function to normalize the score.

I'd love feedback on the statistical approach or edge cases I might have missed.