Gravitational Manipulation Theory v2.3 — Multi-Apparatus Information Sufficiency and Minimum-Record Robustness

Abstract

GMT v2.2 established, for a single Controlled Synthetic Apparatus (CSA), a minimum record for reconstructing a precision-force constraint: operator sufficiency at L*_operator = I2 and a minimum sufficient set I_min = I4. That result was representative-class and was explicitly not claimed to be universal. GMT v2.3 tests whether these two sufficiency boundaries survive when the apparatus class is deliberately changed. One frozen, pre-registered ablation protocol is applied unchanged to three synthetic apparatus classes: CSA-A, the v2.2 anchor; CSA-B, with altered fold number, radii, gap, and offset compensation; and CSA-C, a non-null axial-force observable at m = 2N. Geometry, fold number, compensation topology, and observable are varied across classes, while the tolerance, disclosure widths, sampling, estimator, and certification rule are held fixed. For the primary configuration ε = 0.30 dex and F4 = 1.8, all three classes return the same sufficiency boundaries: L*_operator = I2I_min = I4 The per-class geometry ranges are G1 = (4.964, 6.202, 4.783) dex for CSA-A, CSA-B, and CSA-C, respectively. Although these geometry responses are class-dependent, all exceed the frozen certification threshold by orders of magnitude. The resulting verdict is: Minimum record robust across the tested apparatus classes. The pre-registered hypothesis that the non-null CSA-C might remain operator-sufficient down to I1 is refuted. CSA-C instead exhibits G1 = 4.783 dex and D_H(I1) = 5.010 dex ≫ ε, giving the same L*_operator = I2 boundary as CSA-A and CSA-B. The two robust boundaries arise for different reasons. The operator boundary is geometry-driven: D_H(I2) = 0.228 dex < ε, whereas removal of absolute geometry produces multi-dex ambiguity at I1 for every tested class. The constraint boundary is disclosure-width-driven: D_C(I3) = 2 log10(F4), with no apparatus-geometry dependence at that level. Sweeping F4 over {1.3, 1.5, 1.8, 2.5} gives I_min = {I3, I4, I4, I4} identically for all three classes. A dedicated provenance audit confirms that the large G1 values are genuine geometry-driven spreads rather than accidental near-null artifacts. No hard near-nulls occur in the I1 samples. By contrast, the much larger G0 values are affected by accidental near-nulls and are not over-interpreted. CSA-A reproduces the v2.2 canonical baseline exactly. Two numerical qualifications are explicitly documented: v2.3.1a raises the CSA-C angular resolution from n_theta = 40 to 72 to remove harmonic aliasing, while v2.3.1b retains the frozen absolute-amplitude V3 result as NOT CERTIFIED at the 2% criterion and separately certifies the Stage-2 dex-range estimand through V3b (ΔG_grid = 0.0085 dex). All three apparatus are synthetic representative-class constructions; none is HUST hardware. The Yukawa-form kernel is used only as a synthetic response generator and is not re-entered as a physical constraint. The result is robust across the three tested apparatus classes only: a finite panel is not a proof of universality, and no universal minimum record is claimed. No experimentally demonstrated gravitational manipulation is claimed anywhere in this work. The accompanying supplement contains the frozen Master Spec, proposal, Stage-1 validation and Stage-2 result reports, source code, JSON numerical outputs, G1 provenance audit, and Figures 1–7 required to reproduce and audit the analysis.