CNC - UAT Holonomy Witness

An audit-ready pipeline to test whether a network of atomic clock comparisons admits a single global time coordinate (up to affine gauge) by checking for non-trivial loop holonomy in the inferred clock-to-clock transition maps. The method consumes record-level time-transfer series (phase/time-difference or paired timestamps), infers per-edge affine fits on certified overlaps, and computes loop holonomy defects (scale and offset) on fundamental cycles. Uncertainty is propagated from raw records using preregistered non-asymptotic concentration bounds (Hoeffding / empirical Bernstein / self-normalized martingale options) and a fixed multiplicity rule to return a single decision label: support_UAT, reject_UAT, or inconclusive. This record contains the frozen prereg pack (protocol PDFs), deterministic runner code, JSON input/output schemas, example inputs/outputs, and SHA-256 fixity manifest. It is intended as a CNC-style “no silent switching” witness: the bridge/overlap definitions, tolerances, and decision rule are locked before analysis. The work is method-level; interpretation of rejection outcomes requires separate modeling of systematics (link calibration, reference-clock common-mode noise, relativistic corrections). This is a classical/metrological analogue of CNC’s ‘chart choice as gauge’ principle, expressed as integrability (trivial holonomy) across clock charts.