Single-field inflation made one promise it cannot take back: modes that leave the horizon stop talking to each other. The Maldacena consistency relation, the squeezed-limit theorem of the simplest inflationary models, dictates that a curvature perturbation larger than the observable patch cannot couple observably to the small-scale power within it; any detected modulation of short-wavelength fluctuations by a long-wavelength mode would falsify the entire single-field class at a stroke. The CMB's large-angle behavior keeps flirting with exactly that signature: the hemispherical power modulation is, formally, a long-mode coupling to small-scale power; the low-multipole alignments and the dipole-linked statistics carry the same structure; and dedicated searches for mode coupling (clustering fossils, BipoSH off-diagonals, squeezed bispectrum limits) return the family of 2-to-3 sigma persistents that will not confirm and will not leave.
The model is therefore wedged: its flagship mechanism forbids the couplings its flagship dataset keeps hinting at. The escapes each surrender something: multi-field and curvaton models violate the consistency relation by construction, buying the coupling at the price of the relation's predictive content; superhorizon isocurvature gradients are bespoke; and the default position, that every coupling hint is a fluctuation, must hold simultaneously across the correlated anomaly family. A confirmed long-short mode coupling would be the single most consequential measurement in early-universe physics, which is precisely why the persistent hints carry strain disproportionate to their individual sigmas.
The standing is a theorem under siege by its own data: the consistency relation remains formally unviolated at threshold significance, the hints remain formally present at the same threshold, and polarization-era statistics (LiteBIRD, CMB-S4) will measure the squeezed sector cleanly enough to convict or acquit.