What died: the first positive result of the program — a flat U(1) bundle with φ-scaled holonomy whose low-lying spectrum was compressed and rigid far beyond random and value-matched shuffle nulls (CDF separation 21–23σ at K=3).
What was tried [E]: the original construction prescribed holonomy on an H₁
basis built from a spanning tree, with loop "scale" assigned by fundamental-cycle
length. The cocycle re-build (2026-09-24) verified the arithmetic was exact
(face flux 0 to 1e-16, holonomies exact) — the math was right. But the basis was
an arbitrary spanning-tree artifact. Re-testing with the canonical bar cocycle
(no spanning tree, no length proxy) showed the K=3 "φ effect" was an artifact of
that arbitrary basis: at the gap metric, φ is unremarkable
(phi-not-special).
Lesson recorded [F]: a numerically exact computation over a gauge-dependent object can be both correct and meaningless. Basis-independence checks are a precondition for any spectral claim about flux.
Surviving residue: scale-correlated phase assignment does leave a real
signature in spacing statistics — see symmetry-sector-spacing-mechanism. What
did not survive is any φ-specific role.