Taxonomic relationships come up a lot in games.  Dogs and cats are kinds of animals, Collie and Labrador are kinds of dogs:

graph TD
n0("Animal") --> n1("Dog")
n0 --> n2("Cat")
n1 --> n3("Labrador")
n1 --> n4("Collie")

We can express those relationships using rules:

# Try: [Animal lassie]
[predicate]
Animal ?a: [Dog ?a]
Animal ?a: [Cat ?a]

[predicate]
Dog ?d: [Labrador ?d]
Dog ?d: [Collie ?d]

[predicate]
Collie lassie.
[predicate]
Labrador bruce.

Now suppose we run [Animal lassie], i.e. we ask “is Lassie an animal?” The system has to choose an Animal method, then based on that method, either a Dog method of a Cat method.  If a Dog method, then it has to choose either a Collie method or a Labrador method.  We assume it always makes good choices, and the of the first Animal method (dog) followed by the second Dog method (collie) works.  In so doing it's basically concluded that Lassie is an Animal because:

• Lassie is a collie
• Collies are dogs
• Dogs are animals

We could instead have run [Animal ?x], meaning “who’s an animal?” Then it would find an animal.  Since the predicates aren't tagged with [randomly], it would always choose the first methods, and so say bruce.