# Methods


---
<!-- trackio-cell
{"type": "markdown", "id": "cell_23974ed49895", "created_at": "2026-07-17T04:04:03+00:00", "title": "The official reward functions are translated from the pinned repository. Unifor…"}
-->
The official reward functions are translated from the pinned repository. Uniform-backward exact state flows produce ingredient policies whose terminal law is exactly R/Z. Forward rollout and an independent edge-flow conservation audit verify the mixture.