Small spheres are base facts (A–D, toggleable) and derived facts; cubes are IF-THEN rules connecting them. In forward chaining, any rule whose input facts are all true fires and lights up its output fact — the wave spreads outward generation by generation. In backward chaining, the engine instead starts from a goal fact and walks backward, only checking the facts actually needed to prove it.
Forward: fire rule R if ∀ input ∈ R, value(input) = true
CF_combined = CF1 + CF2·(1 − CF1) (certainty-factor combination, MYCIN-style)
- Fact toggles A–D — set which base facts are known true; this determines which rules can ever fire.
- Forward / Backward — switches the inference strategy: data-driven wave outward, or goal-driven trace backward to a target conclusion.
- Rule count — how many IF-THEN rules populate the knowledge base graph.
- Play / Step — auto-run the inference engine continuously, or advance exactly one reasoning step.
Forward chaining powers real-time monitoring expert systems (e.g. industrial alarm diagnosis); backward chaining powers goal-driven ones like the classic MYCIN medical-diagnosis system.