A hypothetical device that examines arguments, and returns to the FreeStore (i.e. deletes from the text of the argument) any premise or conclusion that isn't supported (directly or indirectly) by the root set (here, a set of facts or assumptions which are known or assumed to be true). Like any good GarbageCollector, this device is capable of detecting and removing cycles in the logical dependency graph. Useful for untying CircularReasoning and its more troublesome cousin, GordianReasoning. Of course this device assumes that we reason with TwoValuedLogic. With FuzzyLogic, which derives results with associated confidence, we would need a FuzzyGarbageCollector, which is not a well defined concept (at least not yet; I can imagine, that such a thing could be useful for KnowledgeBases which can forget ''unimportant'' facts).