Spreading activation is a popular technique for retrieving and ranking indirectly related information by activating query items and spreading their activation along relatedness links. Almost every use of the technique is accompanied by its own set of restrictions on the dynamics, though, and the usual motivation is a reduced computational demand or an improved t to specific types of data. We show that in linear, constraint-free scenarios spreading activation would actually yield query-independent results, so that applications crucially depend on the imposed restrictions. To avoid this undesirable behavior, we study natural modifications that ensure query-dependent results even without heuristic restrictions and provide experimental evidence for their effectiveness.