The Governed Knowledge Loop

This paper proposes the Governed Knowledge Loop (GKL) as a lifecycle for revising a semantic corpus without allowing discovery, machine fluency, storage, or repetition to acquire authority by themselves. The loop turns observations and structural analyses into source-bound proposals; crystallizes them into typed candidates; decides whether a candidate may affect corpus state; verifies the declared effect; promotes only eligible results; and uses contradictions, failures, omissions, and frontier questions to begin the next cycle. The current substrate name in the wider architecture is Governed Meaning Base (GMB), following the author-confirmed lineage CCF -> CMB -> GMB. GKL is the temporal mutation and learning lifecycle over a GMB, not a competing store. Hyperpedia is a curated projection, crystalline agent capsules are bounded context projections, and the Crystal Network is a federation surface. Each depends on GKL for revisable continuity while retaining its own standing and jurisdiction. The paper formalizes corpus state, candidate mutation, decision, effect, verification, promotion, and revision as different objects. It requires exact source custody, explicit authorship and authority, layered identity, lineage preservation, contradiction retention, non-escalation of standing, and receipts tied to observed effects. It also identifies failure modes including self-confirming loops, over-crystallization, graph-induced false certainty, authority laundering, orphaned effects, and destructive restructuring. GKL is therefore a candidate architecture and research program, not evidence that recursive semantic improvement has already been achieved. Implementation and empirical claims require an exact Instance Configuration, tests, negative cases, recovery evidence, and comparison with simpler alternatives.