Abstract Long-running human–machine reasoning systems frequently suffer from epistemic discontinuity: unresolved concepts, questions, and dependencies are lost at session boundaries, leading to repeated rediscovery, implicit assumptions, and false convergence. This paper introduces the notion of Dangling Cognates (CM), a first-class construct within the Cognitive Memoisation (CM) framework, designed to explicitly capture and preserve unresolved conceptual artefacts. By treating acknowledged incompleteness as a durable and governable state, Dangling Cognates provide an anti-amnesia mechanism that enables cumulative reasoning across sessions without forcing premature resolution.