A pre-registered quadratic photon-dispersion coefficient from an integer automaton, decidable by UHECR composition

We present a falsifiable prediction extracted from a candidate discrete substrate for physics: an integer, bit-exactly reversible Feistel automaton whose massless linear branch carries an isotropic, quadratically Planck-suppressed modification of the photon dispersion relation. The isotropic coefficient is an exact rational derived from the lattice geometry, ξ = 187/7680 = 0.024349 (3D spherical average of the trace dispersion), and has been measured on the live integer engine to 0.20% (ξ_iso = 0.024299 ± 1.1×10−4, subluminal sign resolved at 221 numerical standard errors — a property of the automaton's own dispersion law, not a measurement in nature). In the convention of Lang, Martínez-Huerta & de Souza, the prediction reads δ_γ,2 = −2ξ/E_Pl² = −3.27×10−⁴⁰ GeV−² at lattice spacing a = ℓ_P (an assumption; all δ scale as (a/ℓ_P)²). The sign is subluminal — precisely the branch constrained by the Pierre Auger Observatory's 2022 propagation analysis, whose conditional bound δ_γ,2 > −10−⁴⁰ GeV−² (valid only for a composition scenario with a subdominant proton component up to 10²⁰ eV) places the prediction in ~3.3× tension. The tension is not an exclusion: it is decidable by ultra-high-energy cosmic-ray composition measurements, and current data increasingly disfavor the proton scenario while the decisive per-event composition measurement (AugerPrime) is still pending. We pre-register an explicit decision table mapping composition outcomes to the fate of the candidate, committed to a public repository before the literature review that informed this paper, and we state in advance which evasions we would regard as inadmissible. Either outcome — falsification of this candidate (and of any substrate sharing its isotropic n = 2 coefficient) at a = ℓ_P, or survival into the AugerPrime era — is a well-defined result.