Test Roadmap

Bets (B1–B7): a second ID scheme

The TEST-01–26 registry above covers data tests. The research repo's prediction ledger (PREDICTIONS.md) keeps a separate list of theory-level claims it calls “bets”: things that could distinguish the framework from standard physics if tested. They are not in the 24 or the 26. One of them appears in the site's six-refutation headline:

• B1: observer-relative Bell/CHSH (Clauser–Horne–Shimony–Holt). Can a local substrate, measured only through an observer's phase-lock, beat the classical CHSH limit S = 2 without signalling? Simulated 2026-06-21 to 07-06: local arm S = 1.98, nonlocal-grid arm S = 2.0, the framework's own density substrate S = 1.85. Refuted (a construction check, not data; see Honest Assessment).
• B2 RAR transition shape at γ = 2 (refuted under the registered point-level criterion, ΔBIC +184; ≈ +11 with the galaxy as the unit; +46 and +2.7 with each law's own global Υ: disfavoured at about 2σ, recount pending the project lead) · B3 consciousness threshold C ≈ 0.50 (untested; the one test ran on a different variable) · B4 compatibility scaling (heterogeneous case run 2026-09-25: refuted as stated; the surviving 1/〈C〉 holds by construction, so reparametrization) · B5 f(N) reconstruction rate (not yet derived) · B6 non-monogamous entanglement (gated on B1, which failed) · B7 vacuum Umklapp (Planck-suppressed).

Distinct predictions with no test ID

Two external-field predictions differ from MOND and appear on Relation to MOND, but neither has a test ID. Listed here so they are not described as tested elsewhere. Untestedⓘ

• Refraction of an external field. If the density-keyed law is solved as a field equation, ∇·[C(ρ)∇Φ] = 4πGρ, an external field bends wherever C varies across a system, as a field does in a dielectric. MOND's external-field effect has a different form. Rough sizes on file: about 0.3–0.37 g_ext beyond the knee in a halo globular cluster. In a disc it depends on the form of C: negligible for the floored form (C ≈ Ω_mbelow the knee, so ∇C ≈ 0), but of order R/h for the unfloored form, because there ∇ln C = ∇ln ρ however small C is. For satellites (explorer, 2026-10-21, Refracted Gravity's published parameters): the response is about (Δε/ε across the dwarf) × (g_ext/g_int), and the second factor is 10²–3×10³ for Milky Way dwarfs, so even a 7% gradient gives an order-unity effect. It is dipolar (m = 1, lopsided toward or away from the host), unlike Newtonian tides, which are quadrupolar. Being odd in g_ext, it does not shift σ at first order. Not confronted with data; the test is host-aligned lopsidedness in satellite member maps.
• Ambient density at fixed external acceleration. Two satellites at the same g_ext but in hosts with different gas density would differ here and not under MOND. TEST-05 did not run this: it used Mpc-scale Cosmicflows-4 density and controlled for neither g_ext nor the gas density at the satellite. A structural estimate (research archive, 2026-08-21) caps the lever: a satellite that survives its host's tides has ambient-to-internal density ratio at most about 0.1–0.2, and the framework's own baseline offset correlates with local density at r = −0.36, giving signal/systematics ≈ 4×10⁻³. That does not average down with sample size. Not run; estimated to be systematics-limited.

Why neither can move the confirmed count. Both are predictions of the density-keyed field equation ∇·[C(ρ)∇Φ] = 4πGρ, which is Refracted Gravity's (Matsakos & Diaferio 2016). The acceleration-keyed branch the galaxy fits use has MOND's own external-field effect, so it has nothing distinct here.

What “the density-keyed branch is excluded” covers, and what it does not. Excluded is this framework's permittivity, three ways:

• the tanh-log C(ρ) read algebraically (g = g_bar/C, unfloored, ρ = Σ/2h), which loses to the acceleration-keyed law on SPARC by ΔBIC +2843, surviving galaxy-level scoring;
• the same family under the solved field equation, at the framework's floor and at RG's elliptical floor, on a grid of knees: χ²/N 69–2700 against MOND's 21.2 (2026-09-09);
• any pointwise density-keyed G smoothed over ≥ 1 pc, by lunar laser ranging: 68 of 74 laws, including both of RG's published calibrations, on the non-covariant reading.

Not excluded by any of these: the RG class with its own fitted permittivity on discs, which nobody here has refit (Cesare et al. fit it to 30 DiskMass discs). RG's own record in this archive is mixed. Globular clusters pass. Its disc calibrations fail all 15 Milky Way dwarf spheroidals by ≥ 3σ, because in spherical symmetry Gauss's law caps the boost at 1/ε₀ exactly (Leo I, which RG keeps in equilibrium, is off by 9–15σ). Its elliptical calibration is in 2–4σ tension on the dwarfs that are in equilibrium under RG (explorer, 2026-10-21). Its published covariant completion (Sanna, Matsakos & Diaferio 2023) fails the Solar System outright on a site computation (explorer 2026-10-22; unpublished, unreviewed, and contrary to the authors' own presentation of the theory as Solar-System viable; the weak-field reduction is in the finding): its scalar equation is linear, so it is unscreened ω = −1 Brans–Dicke, with PPN γ = 0 (Cassini, ~4×10⁴σ), Nordtvedt η = 1 (lunar ranging, ~2×10³σ) and |Ġ/G| ≥ 0.99 H₀ from its own cosmology. Its weak field cannot vary the scalar by more than ~10⁻⁴ across a galaxy, so it does not contain RG's permittivity at all. RG therefore has a fitted galaxy-scale equation and no working covariant theory behind it. So the refraction effect above is RG's prediction, and prior art. If the framework adopted RG's permittivity it would inherit that test, and RG's dependence on flattening, but a result would score for RG, not for anything this framework adds. A positive result on the framework's own permittivity would contradict the exclusions above, and a null would add a further root to an existing refutation.