Interpretation and conclusion

The three-separation contrast is not an architecture-independent gravitational witness. For Gaussian preparation it cancels the normalized pre-split coherence template of both hypotheses. The mathematical nonquadratic contrast in En01-B remains valid for its attenuation model, but that model is not automatically the dominant signal-generation mechanism of the reference protocol.

A physically selected nuisance changes the analysis. White support displacement couples through potential curvature. It cannot imitate persistent flat-stage localization by changing one shared amplitude. The complete split record supplies a concrete way to exploit that difference: four free times remove shared preparation second moments and retain flat-stage diffusion, even with finite sensing and quadratic measurement backaction.

The retained observable measures total diffusion. Attribution to the selected gravity model still requires calibrated bounds on ordinary flat-stage sources. Its bounded-statistic margin also requires uniform moments, implementation errors, and a valid loss model. The results therefore provide both a warning about signal cancellation and a conditional constructive alternative, rather than an unsupported experimental exclusion.

No compiler run or numerical execution has been performed or claimed in this source. The mathematical results are complete as stated. An apparatus-specific global histogram gap, validated continuum propagation, cooling reproduction, and execution provenance remain separate empirical and computational requirements.