From the Witness to Witnesses: Observability Conditions for Non-Drifting Recursive Inquiry

A previous paper extended the DIKW hierarchy by a transcendence operator T : W → D′ and specified a Witness node, defined by prohibition, as its mechanism. That paper named the Witness a single point of failure. This note states why a single witness is insufficient for recursive inquiry and proposes a multi-axis witness architecture to replace it. Drift is treated as unobservability: it proceeds in the directions of change that no independent witness records. We define a witness axis by one generating requirement — a transition is permitted; a violation is a transition not disclosed in the manner fixed for its axis — and give seven candidate axes: basis (A_B), world (A_M), currency (A_H), question (A_Q), identity (A_Id), lineage (A_L) and regime (A_R). Five are axes of disclosure over transitions made by the system; two are axes of correspondence with the world: truth enters the system through A_M and ages through A_H. We distinguish coverage across axes from redundancy within an axis, replace a hierarchy of judges by a network without apex, state an accompanying-witness law tied to the firing of T, name eight pathologies of the architecture, state a parsimony criterion for admitting new axes — under which drift of the yardstick and reflexive contamination of the world are closed by rules rather than by new axes — and specify a double-dissociation protocol by which each axis can be retained or refuted. All claims about recursive systems are labelled [MODEL] or [HYPOTHESIS]; none is yet established empirically. A structural toy run of the double-dissociation protocol on synthetic learner journals is reported in a companion paper on learning.