Put a mass in a spatial superposition; modulate its energy distribution at frequency ω; sweep ω. Three families of theory give three curves that cannot be confused:
| Theory | What the sweep sees |
|---|---|
| Quantized general relativity | No knee-shaped intrinsic signature. (Graviton-induced decoherence master equations exist and predict their own parameter-dependent spectra — the honest comparison is model-by-model, not against a silhouette.) |
| Collapse models (Diósi–Penrose, GRW; white-noise variants) | A rest floor. (Colored-noise CSL variants carry frequency structure of their own — again: compare to defined models with parameters, not silhouettes.) |
| The audited vacuum | Zero at rest, an ω² rise, a Lorentzian knee at γ, then more than the collapse scale. |
And this gate owns everything: the framework predicts the shape, not yet an absolute rate: a null excludes the amplitudes the experiment reaches, nothing more. TOTAL falsification by null therefore requires the absolute-rate prediction (Γ = F(m, ΔE, Δx, ω) in s⁻¹) that the unbuilt units bridge owes — named here as the falsifiability gap, the third of the three priority deliverables. Until it is closed, the knee is the framework's best discriminator, not yet its executioner. Closed at class B (n127 staked → n128 measured): the bridge constant is the monitor's two-channel ratio — Γdyn/Ωstatic = ζQ with ζ = 1/4 exact (0.4834 retracted as a partial-cycle artifact, audit §11), slope 1.0000 over four decades in one model (the family sweep ran, n196: the Q¹ exponent survives — slopes 1.000 across the family — while ζ itself is demoted to a geometry-dependent parameter, ×28.6 spread across kernels; order-unity only for compact sources). Assembled: Γ = ¼·Q·(G/c⁵)·P²/(ħγ)·γ²/(γ²+ω²) (coefficient corrected from the artifact 0.48) with the electron-anchor Q = 4.2×10⁴² — L(ω) is Lorentzian to within 11% at the knee in the n128 model. The tensorial functional Γ[TAμν−TBμν] is now derived at model level (n180): the monitor bills time derivatives of local k, incoherently per junction — and its own pre-registered kill test fired. The coherence cell is derived, not fitted: ℓc = c/γ = ħ/(αmec) = the Bohr radius, exactly (the anchor identity rewritten — zero new parameters). Coherent P² holds only inside one cell; an XFEL beam is ~10¹³ cells, and the form factor 1/Ncells moves the required power to kW class. The milliwatt-of-total-beam-power target is retracted (the old claim, the failure, and the replacement are recorded in n180). The gate then ran as frozen (n181: the four rows exact, mass-preserving scalar histories, translation invariance at machine precision — rotation tested only as a lattice axis swap — positivity manifest, no post-hoc tuning, all discretization and benchmark choices declared) — and failed its own interval: the preregistered benchmarks come out at g = 3.5×10⁻³ (breathing sphere) and 1.9×10⁻² (rotating dumbbell) against the declared [0.1, 10]. Per the frozen rule: both absolute targets — the milliwatt and the 3 pN single-ion figure — are retracted; no absolute laboratory target is currently armed. (The single-ion g = 1 idealization was itself an unregistered geometry assumption — the class of rescue the gate forbids. The coherence-cell scale ℓc = c/γ is a named assumption, not a derivation.) What the gate delivered instead: the functional as a rate calculator for explicitly specified designs, and one benchmark-specific number: in the selected 64³ benchmark the dumbbell's functional is 5.27× the sphere's — geometry-dependent (the reviewer's sweep spans ~2–10 across reasonable geometries), not yet a universal BMV prediction; promotion requires a registered geometry sweep and an operational definition of equal modulation. The gate, executed → · n180 · n128
Three constraint sources sharing no data agree on the band: the pulsar-friction floor (5.2×10¹⁴ s⁻¹), the GRB-dispersion ceiling (~7×10³⁴ s⁻¹), and the hierarchy demand, which lands at 10¹⁹–10²² s⁻¹ — inside the cage. The window then collapsed to a formula when G cancelled: γ = α·mc²/ħ, the fine-structure rate of whichever mass the vacuum counts in. The window's width is exactly mp/me. (Bookkeeping note: the 10¹⁹–10²² band was the order-of-magnitude bracket of the original arithmetic; once G cancelled, the exact anchor window is 5.67×10¹⁸ ≤ γ ≤ 1.04×10²² s⁻¹ — electron edge ħγ = 3.729 keV, proton edge 6.847 MeV — and the electron anchor sits at the refined lower edge, not outside it.) The anchor itself is a hypothesis, not a derivation: that the electron — the bound-saturating dressing — is the vacuum's counting unit is the registered bet, and the knee experiment adjudicates it. Under that hypothesis: ħγ ≈ 3.7 keV (5.7×10¹⁸ s⁻¹). The practical reading: T₀₀ modulation in the attosecond/X-ray regime, not the optomechanics bands originally imagined — mass-distribution dynamics at soft-X-ray rates are where the vacuum's audit should become visible.
Through the hierarchy lock (γknee/Ωgrav = Q), a measured knee frequency plus the saturated decoherence rate pin the vacuum's monitor completely: G follows through a one-unknown consistency relation (closable via the α anchor), the hierarchy becomes an output, and prediction 14's rigidity test (knee vs γPPN, one parameter in two observables) turns on. The same measurement decides which mass the vacuum counts in — settling, as a side effect, what the mp/me window was for.
1. The modulated-superposition test (flagship). A levitated mass in spatial superposition, held at rest, then driven at ω swept across decades. Predicted: exact zero at rest; under drive, Lorentzian ∝ (dT₀₀/dt)²/(γ²+ω²). The kill is asymmetric, and honesty requires saying so: decoherence at rest kills unconditionally (prediction 3 — that is collapse-model territory, not ours); a detected modulated signal with the wrong shape or onset exponent kills; but a flat modulated null excludes only the amplitudes reached — the falsifiability gap above. No rival theory occupies this signature space: shape, if seen, is unambiguous.
2. The anchor test. Repeat test 1 with masses spanning 3+ decades. Framework-level claim: the knee is a property of the vacuum, not the probe — it must sit at the same frequency for every test mass. Knee scaling with test mass kills the framework. The 3.7 keV location is the electron-anchor bet: absence of a break in the keV band kills the bet, not the framework (the window stays open).
3. The clock–accelerometer split. Co-locate an optical clock and an accelerometer in a strongly varying potential (elliptic-orbit satellite; moving kiloton mass). Prediction 5: any EP violation nature ever shows is an acceleration anomaly, never a clock anomaly — ω = E/ħ reads the potential exactly. Kill: one confirmed clock-EP violation with intact force-EP. Cleanly separates this framework from scalar–tensor and dilaton theories, which generically leak into clocks.
4. The dwarf-galaxy archival fit. No hardware: fit existing dwarf and ultra-diffuse rotation curves against the winding sector's standing content (prediction 10, repaired): effective collisionlessness is the prediction — the r⁻³·⁵ shape survives only at the irrelevant amplitude σ/m ≤ 3×10⁻¹² cm²/g, so the fit is a consistency check, not a discovery channel. Kill: any confirmed self-interaction at σ/m ≥ 10⁻³ cm²/g. Cheapest test on the list.
5. The collapse-frontier audits. Archival, corollary grade — strong kill, weak confirmation: LIGO's noise budget must close with known physics, and cold neutron stars must keep getting colder (no Diósi–Penrose rest floor, ever). A confirmed rest floor kills this framework and confirms its collapse-model rivals in one stroke; continued nulls merely leave both us and general relativity standing.
The instrument concept exists: the Auditometer — an XFEL-driven coherent T₀₀ modulator (the pump, sweepable through the soft-X-ray band) beside a matter-wave interferometer (the probe), discriminating on the one property no conventional channel shares: visibility loss that tracks the Lorentzian in frequency at fixed power and penetrates every shield. Poor man's version on existing tech: the knee ladder — hyperfine → optical → Th-229 → XFEL → Mössbauer → gamma isomers, straddling the predicted knee with weights 10⁻¹⁶ → 1.
The shape claims are dimensionless and pre-registered: exact zero at rest (no floor), onset exponent 2.00, Lorentzian form — with the trunk audit's correction carried: the Lorentzian was the single-pole monitor's transfer function — and is now a DERIVED approximation (n211): the joint-model relaxation spectrum is multi-pole (graph eigenvalues), and for local modulation the derived shape is Lorentzian to <0.1% on the test graph; multi-pole/memory monitors remain distinguishable at the n48b precisions (8.4% / 104%); n48's registered gates P1/P2/knee failed as written — unit-convention calibration, recorded in the prereg itself — and the 0.3% Lorentzian match is a flagged post-hoc comparison against the assumed single-pole transfer function; the n128 mediator model deviates up to 11% at the knee, saturation above the knee. The knee's absolute frequency carries the stated anchor bet with its declared convention — a null at any γ is a real exclusion, and a knee anywhere is the discovery. Everything is public and runs in minutes: the registration · the script · the window arithmetic · the anchor argument.