Orbital History Writes the Cosmic Clock on Dwarf Galaxies: A Pre-Registered, Modified-Gravity-Impossible Test of de Sitter-Unruh Modified Inertia

I pre-register a sharp, falsifiable prediction of the de Sitter-Unruh modified-inertia framework -- the proposal that a body's inertia is its nonlocal-in-time response to the cosmic-horizon Unruh bath, a bath carrying exactly one clock, the de Sitter rate H_Lambda, which fixes a0 = cH_Lambda/Z = 9.36e-11 m/s^2. Because inertia in this picture is a functional of the recent acceleration HISTORY rather than of the momentary acceleration, the bath's clock must imprint on the internal dynamics of any object whose external acceleration changes on a timescale comparable to its own internal orbital frequency. The cleanest accessible site is a diffuse Local-Group dwarf galaxy on a radial-plunge orbit: at fixed pericenter distance and fixed mass, a high-eccentricity plunger should run HOTTER (larger internal velocity dispersion sigma) than a circular-orbit dwarf of the same mass and same closest approach. The SIGN is a theorem (the Milgrom-1994-forced memory kernel theta is decreasing, so a plunge sheds adiabatic external loading and drops the dwarf deeper into the deep-MOND regime). The MAGNITUDE is now bath-constrained rather than free -- the de Sitter-Unruh correlator selects theta(0)=sqrt2 (the amplitude/excess-heat branch the framework already requires for the correct deep-MOND law) and forces a Lorentzian kernel shape (from the de Sitter Wightman function), with the dispersion rising as sigma ~ theta(y)^(-1/2) (external-field-effect regime); the discriminating axis is each dwarf's CURRENT y (present orbital phase via the ~0.4-0.5 Gyr memory kernel), not eccentricity or the pericenter peak. Pericenter-peak excess +90-130% for the named carriers, but both are currently near apocenter (present excess ~+7-8%), which explains the pilot null; the lone remaining modeling choice is the kernel's corner location (placed at the internal orbital frequency, quasi-statically licensed by Milgrom 1994, not derived -- honestly NOT claimed as fully forced). The discriminating physical axis is y = omega_external/omega_internal, gated by diffuseness rather than raw eccentricity; the two named carriers that reach the non-adiabatic band are Crater II (y = 3.28) and Antlia II (y = 2.55), against adiabatic controls such as Fornax and Sculptor (y < 0.3). Crucially, this prediction is MODIFIED-GRAVITY-IMPOSSIBLE: in any metric or field MOND (AQUAL, QUMOND, AeST) and in LambdaCDM, the external-field effect is instantaneous -- internal dynamics depend only on the momentary value of the external acceleration (Milgrom 2022) -- so those theories predict exactly zero correlation between internal sigma and orbital history at fixed pericenter, for any a0. Only modified inertia gives a nonzero, history-dependent sigma; this joins the Cassini bound as a clean modified-inertia-versus-modified-gravity discriminator. I report the honest current status straight: a pre-registered pilot on 24 real Milky-Way dwarfs (Pace, Erkal & Li 2022, Gaia EDR3 orbits) gives a partial Spearman correlation rho(sigma, eccentricity | pericenter, mass, half-light radius) = -0.196, p = 0.395 -- a NULL, slightly the wrong sign, robust to a tidal-heating control. This is NOT a falsification and NOT a hint: it is a null with almost no statistical power, because only two of 24 dwarfs reach the carrier band, eccentricity is a noisy surrogate for the real y-axis, and the per-object sigma errors are comparable to the predicted signal. Existing data cannot yet test this prediction. The decisive test is Gaia DR4 (December 2026) plus diffuse-carrier spectroscopy and resolved profiles, analysed as a carrier-versus-control y-contrast with explicit tidal modelling. This is a pre-registered, near-term, modified-gravity-impossible prediction -- NOT a theory of everything, and NOT a detection. It supplements and is consistent with the author's published one-parameter effective-theory papers (DOI 10.5281/zenodo.20935948, 10.5281/zenodo.20938891, 10.5281/zenodo.20947913). All numbers are reproducible from the public repository.