Additive route value
An edge score summed along a walk: what was done, not only where you arrived.
Role. Target. Graph node route.
Discounted potential-based reward shaping, F(s,a,s′)=γΦ(s′)−Φ(s), preserves optimal policies under its stated assumptions [Ng, Harada, and Russell 1999]; that invariance is not identity of edge values, trajectory returns, or every bounded-rational learning process.
— Frames draft §4.1
Route value is a function of transitions. It is the parent of the exact-potential summand (endpoint-only) and the cycle residue (round trips matter).
A matched-endpoint pair — learning versus rewiring a judge to the same later verdict — is invisible to a state score that stores only the verdict, and visible here if the two edges are labeled differently.