We do believe in a hybrid approach where a fast/deterministic representation is saved - but think there is a more seamless way were the framework itself is high level and manages these details by caching the underlying actions that can run