The local-type non-Gaussianity amplitude f_NL is the most consequential single number in primordial physics after the spectral index, because it cleanly divides the model space: single-field slow-roll inflation predicts f_NL of order (1 - n_s), about 0.01, through the Maldacena consistency relation, while multi-field scenarios, curvatons, and most alternatives to inflation generically predict |f_NL| of order unity or larger. Planck's final measurements give f_NL local = -0.9 +/- 5.1 (2018) and -0.1 +/- 5.0 (PR4 reanalysis): consistent with zero, consistent with single-field inflation, and consistent with dozens of alternatives, because the error bar sits three orders of magnitude above the single-field target.
The tension is epistemic: the parameter that was supposed to falsify inflation's competitors cannot currently falsify anything. Reaching sigma(f_NL) near 1, the threshold separating single-field from multi-field scenarios, requires the next generation: CMB-S4, SPHEREx, and galaxy surveys exploiting scale-dependent bias, where local non-Gaussianity imprints a distinctive 1/k^2 signature on the clustering of biased tracers. Until then, every claimed constraint on the primordial universe from f_NL is a constraint on models that were already generous enough to span the measurable range, and the headline number mostly certifies that the early universe was very Gaussian without saying why.
The standing is a sensitivity race with a buried discriminant: if the true value is at the single-field level of 0.01, no planned experiment will detect it, and the null will persist indefinitely, indistinguishable from exactly zero. A framework that predicts a specific small value for a specific structural reason, rather than a potential-dependent range, is the only kind that stays falsifiable in that scenario.