So that's not removing randomness, it's just grade inflation. I suppose it's also secret range narrowing, since an 3% improvement from 92% to 95% is actually no improvement at all at 1.1 * (random) capped at 100%.