A miniature MYCIN: a 1970s expert system where the intelligence is human knowledge typed in by hand as IF-THEN rules, and a small inference engine that walks them. No data, no training, nothing learned. It still belongs to the Classical AI layer, and it has a superpower modern models lack: it can show you exactly which rule produced every conclusion. Answer the questions about your sick houseplant.
every symptom is asked at most once, and a rule is dropped the moment one of its symptoms is false, so most of the 15 questions never get asked.
Every rule in the knowledge base, and what happened to it. This transparency is the whole selling point of expert systems.
The knowledge base is a list of rules, each a set of required symptoms plus a diagnosis and a treatment, and the question map gives every symptom one plain-language yes/no question. The engine knows nothing about plants: it forward-chains through the rules in order, and for each one asks about the symptoms it needs, testing already-answered symptoms first because they are free. The moment any symptom comes back "no" the rule is eliminated and its remaining questions are never asked, which is what made these systems feel intelligent rather than clerical. Answers go into a working-memory cache, so a symptom shared by five rules costs one question. Every rule that survives fires, because a plant can be overwatered and infested at the same time, and a diagnostician that stops at the first plausible story is a bad one.