Planck measures f_NL local = -0.9 +/- 5.1 (PR4: -0.1 +/- 5.0): a null three orders of magnitude above single-field inflation's 0.01 target. The parameter built to divide the primordial model space currently falsifies nothing, and if the truth sits at the single-field level no planned experiment will ever see it. The headline certifies the early universe was very Gaussian without explaining why.
ΛCDM delegates the prediction to an unspecified inflaton sector: single-field gives 0.01, multi-field gives nearly anything, so the model has no committed number, only a family of targets spanning the measurable range. Whatever is measured, some variant claims it, which is the opposite of a test.
SCT replaces the hot-dense-center with a cascade of order 10⁴ collision events (P36, P37, P38), and the Gaussianity of the primordial field becomes bookkeeping: summing that many independent deposits invokes the central limit theorem, leaving residual non-Gaussianity of order 1/√N_coll, so |f_NL| of about 10⁻². The universe is very Gaussian because it was built by very many events, and slightly non-Gaussian because the number was finite. The value is structural: no potential to tune, no model family to span, one number locked to the same N_coll that fixes the spectral tilt n_s = 1 - 1/L through the cascade depth.
The prediction lands inside the Planck null and below near-term sensitivity, which SCT shares with single-field inflation; the difference is falsifiability from above. Inflation's broader family absorbs any detection; SCT cannot. A measured |f_NL| above 5 forces N_coll below 25, while the observed tilt requires the cascade to be thousands of events deep: the two numbers cannot both survive, so any unambiguous detection at current sensitivity kills the framework outright. Meanwhile the companion signatures stay testable: mild scale dependence of the residual non-Gaussianity from the stage structure, suppressed isocurvature, and no tensor background above r of 10⁻⁵.
There is no need to invoke an inflaton, and no refuge if the data ever disagree.
The registered kill is two-sided. Detection of |f_NL| above 5 at 3σ or better contradicts the cascade event count demanded by n_s and falsifies SCT's primordial sector. Conversely, if CMB-S4, SPHEREx, and 21-cm surveys eventually reach sensitivity near 10⁻² and find f_NL exactly zero, the finite-count residual the cascade must leave is absent and the counting argument fails.