A Two-Configuration Cavendish Test That Separates Low-Acceleration Gravity Models: A Proposal for G Metrology Groups (Working Note)

Working note, version 1.0 (2026-10-15). AI-assisted research programme; not peer reviewed. We propose an inexpensive laboratory measurement that sharply separates low-acceleration gravity models: the fractional change delta in the measured G when the source masses' acceleration at the test masses is changed by a factor of four on the same torsion-balance apparatus. A source-field-boost model (Klein's interpolation as used by Scarpa, Falomo and Treves 2026) predicts about 340 ppm between half and twice the Quinn et al. source acceleration; a dark-energy-tied framework screened at laboratory scales predicts |delta| below 1e-6 (model value of order 1e-20); Newtonian gravity predicts zero. Separating the boost prediction needs about 30-50 ppm on the difference, within reach of leading G metrology groups. We describe configurations (source density swap preferred; distance change; combined), interleaved and blinded measurement, null checks, a precision budget, and the interpretation of each outcome. Caveat: the Quinn geometry and Scarpa et al. numbers are approximate; groups should recompute predictions for their own apparatus. AI assistance: Anthropic's Claude.