How Much of a Program Is a Theorem?

Abstract

A geometric research program of V. Veselov proposes to read curvature, a flow, and a spectral trace off a family of graphs built from the golden mean, and to reach — eventually — a bridge to the Riemann hypothesis. A program is not a theorem. We ask a sharper question: exactly how much of it is one? We answer by constructing the program's finite substructure as a single certified pipeline and by naming, precisely, where the certificates stop. The carrier is the Fibonacci cube Γn , whose vertex count we prove is Fn+2 for every n and whose edges are exactly the one-bit toggles the golden-mean constraint permits. On it we build exact integer Ollivier–Ricci curvature certificates whose signs bound the entire transport optimum rather than one exhibited coupling; one bounded rational Ricci-flow step that provably returns to normalization; a decidable surgery rule with explicit contractible caps; and the Laplacian spectrum of the star K1,m , which we prove is exactly {0, 1(m−1) , (m + 1)1 } — sound and complete, so that the eigenvalue m is provably absent. These five stages compose into one report, each field computed from the last. What remains — an infinite trace formula, convergence of the flow to an expander, and the spectral bridge to the Riemann hypothesis — we do not prove. We isolate it as three named conjectures that no theorem in the core is permitted to invoke. The separation is not editorial. It is an import boundary a proof kernel enforces: the hard core of a research programme rendered as machine-checked mathematics, its protective belt rendered as an axiom no core module imports. Published by The Institute for Applied Ontological Mathematics.