State utility

Role. Target. Graph node state.

State value is adequate when the state carries all normatively relevant information. It is not adequate merely because an implementation calls its representation a state.

— Frames draft §4.1

This is the default welfare object. In RL, the per-step state reward summed along a path is a route value, not this (markov_reward_not_state_utility); its optimal value function is a state utility, and that is where Omohundro-style instrumental arguments meet correction. State means the bare state, with no record of how it was reached; adding the record gives the record-carrying state lens, where round trips can have value.

Fixed U is incompatible with correction only for updates that make the agent pick an outcome U ranks strictly lower — not for every change of U. At its optimum it accepts a correction only if the target ties the optimum, and it never prefers keeping the channel (soares_button_neutrality_iff). That is the red edge stateCorrectionIncompat, not “utility hates all revision.” It holds by the choice of situation class; what correction endorses is not modeled.