Redshift evolution of the radial acceleration scale: a composite horizon law tested against 2025–2026 rotation-curve data

Recent kinematic surveys indicate that the characteristic acceleration scale a0 of the radial acceleration relation (RAR) increases with redshift. We confront this trend with four zero-parameter laws for a0(z), each tied to a distinct physical anchor: (i) a constant scale (particle-physics anchor, as in superfluid dark matter, or a pure-Lambda vacuum anchor); (ii) the instantaneous horizon, a0 proportional to cH(z) (Gibbons-Hawking/Cai-Kim temperature); (iii) the dynamical Kodama-Hayward horizon temperature; (iv) a matter-density anchor. Using the published MUSE-DARK III measurements, the MIGHTEE-HI local value, the SPARC calibration, and an independent re-analysis of the public DARK data release (88 galaxies, 440 reconstructed acceleration points), we find that the constant and Kodama-Hayward laws are excluded under every representation of the data (>= 4 sigma), the matter anchor is disfavoured, and a0 proportional to cH(z) is the only surviving shape. A one-parameter composite law, whose form follows from a cosmological consistency argument, reconciles the residual normalisation tension between surveys; we obtain the first measurements of the slaving fraction nu = 0.46 (MIGHTEE anchor) to nu = 0.86 (SPARC anchor), with an anchor-free track-level refit preferring evolution over a constant scale at 2.8-4.6 sigma across a grid of declared systematics. The dominant degeneracy - cold gas unaccounted in the baryonic acceleration - is quantified and constrained by three independent arguments; the residual loophole is directly closable by LADUMA. The law makes dated, falsifiable predictions for ALMA [CII] discs at z ~ 4.5, for the low-z slope measurable within a single HI pipeline, for the baryonic Tully-Fisher zero point, and for stacked weak lensing in lens-redshift bins. We summarise a covariant embedding within an aether-scalar framework, in which a linear-order equivalence theorem transfers the CMB and gravitational-wave-speed successes of the chassis, and a first-pass stability analysis returns an internal bound on the slaving fraction whose numerical value is compressed into a single geometric parameter, with nu predicted in closed form for every value of that parameter. The decoupled case, equivalent to a pure-number coupling relation, fixes nu = 0.745 and removes all shape freedom from the law. All assumptions are declared; all analysis inputs are public.