FeasibleAssignments

The vehicle-to-task matchings available at this instant, as something a policy can enumerate and search rather than a predicate it applies after guessing.

The distinction is not cosmetic. A rule like "send the nearest vehicle" only ever needs to score the candidates it thought of; a policy that scores every available action and takes the best needs the action set itself to be an object it can walk. That is the shape a cost-function or value-function policy has, and it is the shape a decision epoch has in the sequential decision-making sense — state in, feasible actions enumerated, one chosen. Providing it here is what keeps a dispatcher's wake adoptable as a decision epoch later without the interface changing.

Feasibility is reachability, and nothing more. A pairing is feasible when the vehicle can get to the task's pickup along the guide path. Everything else a modeller might mean by "feasible" — enough charge, the right attachment, a shift that has begun — belongs to a bidding rule or a scoring function, because those are judgements about desirability that vary by model, while reachability is a fact about the space that does not.

Cheap to construct and lazily evaluated: candidates is a sequence, so a policy that wants the first acceptable pairing does not pay for the rest. Distances are computed on demand rather than precomputed, since a policy scoring by something other than distance should not pay for a matrix it never reads.