Write a Soft Rule

from beekeeper import SoftPreliminaryRule


class PreferSeniorWorkers(SoftPreliminaryRule[MyWorker, MyRequest]):
    def score(self, entity: MyWorker, allocation: MyRequest) -> float:
        return min(entity.years_of_service / 10, 1.0)

Soft rules return a non-negative score: float instead of a bool. compatible stays True — soft rules never veto. The framework aggregates per-rule scores via geometric mean (see Soft-rules aggregation) and surfaces the result on each Candidate.

A score of 1.0 is neutral. Higher is better. The bundled LoadBalancingAssignmentAlgorithm picks highest-scored candidates first, subject to a per-entity load penalty so work disperses across the pool.

Both preliminary and stateful soft variants exist — SoftPreliminaryRule for static preferences, SoftStatefulRule when the score depends on what's already been scheduled (e.g. "prefer the worker who hasn't done this task recently").

For both axes (hard veto and soft score) in a single rule — uncommon — subclass PreliminaryRule or StatefulRule directly and return a RuleVerdict(compatible=..., score=...) from evaluate(...).