Fermion Generations and the Koide Relation as Geometric Features of the $\mathcal{M}$-$\mathcal{A}$ Embedding in the Absolute Frame Theory

Within the Absolute Frame Theory (AFT), in which the observable four-dimensional manifold $\mathcal{M}$ is continuously embedded in an $N=10$ Riemannian substratum $\mathcal{A}$ via a map $X:\mathcal{M} \to \mathcal{A}$, we develop a geometric framework for the family structure of the Standard Model fermions. We identify the fluctuation (Jacobi) operator of the embedding as the spectral object whose bound states count and label the families, and show that a minimal, a posteriori-calibrated background --- a single localized deformation --- produces a Pöschl--Teller well admitting exactly three bound states, thereby accounting for the number of charged-lepton generations as a counting statement that is falsifiable by the discovery of a fourth sequential generation. We state explicitly, and verify by direct calculation, that this minimal model does not fix the mass ratios: the observed hierarchy is steeper than the well's bound-state spacing by more than an order of magnitude. We then examine the conjecture that the spin degree of freedom contributes to the mass through the extrinsic "twist" of the embedding (the transverse Dirac operator built from the second fundamental form). Combined with a cyclic $\mathbb{Z}_3$ phase structure acting on the three generations, this reproduces the Koide relation $\sum_i m_i=\tfrac{2}{3}(\sum_i\sqrt{m_i})^2$ for the charged leptons exactly, through the amplitude ansatz $\sqrt{m_n}=M_0(1+\sqrt2\cos(\theta_0+2\pi n/3))$. We show that this generation $\mathbb{Z}_3$ has no geometric realization in the $N=10=4+6$ embedding (the six normal directions are colour; the colour centre cannot label generations of identical colour content), and we therefore adopt it as an emergent $S_3$ flavour symmetry of the three democratic radial copies --- a status shared with all current frameworks, none of which derive the family number from first principles --- while proposing its single phase $\theta_0$ to be a Gödelian ($\mathcal{A}$-internal) input. We also establish that physical mass is the amplitude with which a radial mode couples to the background, not the bare fluctuation eigenvalue (the standard localized-fermion mechanism), and that the democratic part of that amplitude is exactly common across generations by the constant modulus of the Helmholtz pilot wave. The amplitude $\sqrt{2}$ is analyzed at length: we prove it cannot be a rotation/holonomy overlap (such overlaps are bounded by unity), identify it instead as the crest factor of a sinusoidal twist, and trace its origin to the equal ($50/50$) partition of the quantized action between two chiral channels. That equal partition is derived, not assumed: the Information Bottleneck optimum of the embedding channel is symmetric under $L\!\leftrightarrow\!R$ and concave in the allocation, hence its unique maximum is the symmetric one; the relevant variable is forced to be the mass, the unique continuous Casimir invariant of the Poincaré group; and since that Casimir is the pole mass, AFT predicts the relation for the pole masses --- which is precisely where it holds ($Q=2/3$ to $2\times10^{-6}$ for the pole masses, against a $\sim10^{-3}$ scheme gap to the running masses). We are explicit throughout about what is rigorous, what is conjecture, and what is open: the count of three is structural; for the charged leptons the Koide relation becomes a prediction, conditional on the emergent generation $\mathbb{Z}_3$, the Gödelian phase $\theta_0$, and the bottleneck preserving the mass alone among continuous invariants (which the mass-Casimir uniqueness nearly secures); the absolute mass ratios remain underived; and the quark and neutrino sectors are open. The framework yields one sharp falsifiable prediction (no fourth sequential generation).