A Witness Is Not a Verdict: Sealed Simulation Tests of Instruments for Gravity-Mediated Entanglement, and a Coherence Fingerprint of Classical Gravity

The Bose-Marletto-Vedral (BMV) proposal infers that gravity is quantum if two masses, each in a spatialsuperposition, become entangled through gravity alone. No such experiment has been performed. Its reading has twoweak points: non-gravitational interactions (Casimir-Polder, electric dipoles) also entangle, and a classicalgravitational channel can imprint the Newtonian phase without entangling. We simulated the two-spin experiment(10^-14 kg nanodiamonds, 250 um superpositions, 2.5 s, six separations from 125 to 800 um, nine-settingtomography) under quantum gravity, classical channels and no gravitational phase, with unknown Casimir strength(x0.2-20), dipoles, terms of the wrong power, calibration and mass errors, dephasing and finite shots, and measuredeach instrument's false-alarm rate before trusting it. In sealed runs a lone witness called classical-gravity worldsquantum in 10.7% of cases (data-tuned bases 15.4%), and reading only the 1/r shape of the phase did so in 68.3% ofclassical-channel worlds. A three-valued instrument (W1) that demands entanglement beyond what non-gravity alonecan produce together with the Newtonian coefficient, and that needs an independent dipole calibration, never called aclassical world quantum (0 of 16,500 over four seals). A first claim that a null result can read "classical" rested on aclassical channel whose phase had the wrong sign; corrected, the channel hid behind its own decoherence and theclaim became Unresolved. Classical channels must pay coherence for the phase, and that cost grows as the massesapproach, so the dependence of coherence on separation is a fingerprint. Fixing its shape as phi^2 failed a seal(Falsified: 25 of 1022), although the signal was present in 1020 of them; with the power left free the final seal gave Ron all five claims: a time-sliced classical channel never seen before was called classical in 996 of 996 worlds,quantum worlds never (0 of 1500), and classical worlds overall in 57.0%. All results are simulated. ※ Data and code. All scripts (AGPL-3.0-or-later, GNU Affero General Public License v3.0 or later), the four seal documents with their SHA-256 files, develop logs andresult files are provided as Bridge_Paper_A_Witness_Is_Not_a_Verdict_scripts_v1.0.0.zip; every world isregenerated from its seed. ※ AI use. Designs, code, analyses and text were produced by Kurado (Claude Opus 5.5,Anthropic) working with M. Watabe, who chose every direction.