Resolving conflicts between action effects and Environment Model constraints

Ascertain the appropriate rule for defining a bound distinction e in the succeeding state when e is simultaneously constrained by the Environment Model V (the conditional distribution P(H | F) where e ∈ H) and directly affected by an action A (e ∈ eff(A)); specifically, determine whether the succeeding-state value of e should be governed by the action’s conditional belief network or by the Environment Model’s conditional distribution.

Background

The authors introduce an Environment Model V as a conditional distribution P(H | F) that captures invariant probabilistic relations across states and a projection algorithm that uses conditional belief nets (CBNs) of actions alongside V. They note a potential conflict when an action directly affects a bound distinction in V, acknowledging it is unclear which model should dictate the succeeding state. They adopt a compatibility restriction to avoid such conflicts, but the underlying precedence question remains explicitly stated as unclear.