Does the MOND Acceleration Scale Evolve? A Pre-Registered Decisive Measurement: the Deep-MOND Tully-Fisher Zero Point of One Lensed Rotator at z ~ 2.5, with a Two-Stage JWST/ALMA Funnel

The de Sitter-MOND framework ties the galactic acceleration scale to the dark-energy density, a0 = κ c √(G ρΛ), and therefore predicts that a0 does not evolve while Λ does not: the deep-MOND baryonic Tully-Fisher zero point is the same at z = 2.5 as at z = 0 to better than 1%. The ΛCDM-native alternative, an emergent acceleration scale set by halo structure, rises with redshift by a factor 2.1 at z = 2.5, a zero-point displacement of 0.33 dex in baryonic mass at fixed velocity or 20% in velocity at fixed mass. We show, from a twelve-constraint joint likelihood over every published high-redshift Tully-Fisher zero point and a 21-object ledger of candidate deep-MOND targets, that the existing archives cannot decide this: every published sample above z = 0.5 except one lensed-dwarf sample sits at 1.7-6.4 a0, where the zero point keeps only 7-18% of the a0 lever and is degenerate with disc-size evolution, and no known object passes the deep-MOND, rotation and lens-quality gates simultaneously. We therefore pre-register the measurement that would: one strongly lensed, rotationally supported galaxy at 2.3 < z < 2.9 with gbar(Rout) < 0.3 a0 on both a0 footings and at the upper end of its baryonic-mass uncertainty, screened for rotation with the JWST NIRSpec IFU (G235H/F170LP places Hα and [O III] in one setting for 2.32 < z < 3.83) before ALMA time is spent, then given CO(3-2) in ALMA Band 3 (1.98 < z < 3.12) for an independent gas mass and a second dynamical tracer. The decision statistic ΔBTFR = log Mb − 4 log Vf − C0, with C0 frozen from the local calibration, is scored against the two pre-specified values, 0.00 and +0.33 dex, at a required total uncertainty of 0.13 dex (20:1 discrimination). No drift nuisance may be introduced after the target is observed; a result inconsistent with both values counts against both. Every number is produced by a committed script in the public repository. AI-assisted research draft; not peer reviewed.