The trusted engine —
and why we test it.

The Volterra process has a variance that grows over time. The correct way to normalise the model uses the discrete variance the scheme actually produces — call it v_i. An earlier engine used the continuum formula t^(2H) instead, which is subtly wrong for a discretised path: it assumes the process has unit variance, when the discretised path’s true variance is smaller — around 15% smaller at H=0.10 — a gap that quietly biases everything built on it. The paths still looked fine. Flip the bug on below, and run the test.