A six-parameter model fit to one dataset should return the same six numbers from any sufficiently informative slice of that dataset. Planck fails this self-test. Addison et al. (2016, arXiv:1511.00055) showed that ΛCDM parameters derived from the low multipoles (ell below about 800 to 1000) disagree at 2 to 3 sigma with those derived from the high multipoles: the cold dark matter density omega_c differs at about 2.5 sigma, and the inferred Hubble constant from ell of 2 to 800 sits several km/s/Mpc above the value from ell of 801 to 2508. The same sky, the same instrument, the same model, two different universes depending on where you cut the spectrum.
The split is not random scatter; it has a known shape. The high-multipole range prefers more gravitational lensing smoothing than the best-fit model predicts, the same excess quantified by the A_lens parameter (1.180 +/- 0.065 in PR3), and when that extra smoothing is absorbed into the standard parameters it drags omega_c up and H_0 down in the high-ell fit specifically. Marginalizing over a free lensing amplitude largely reconciles the two halves, which localizes the problem without solving it: within ΛCDM there is nothing physical for A_lens to be, so the model must call the reconciling parameter a likelihood artifact. Explanations within the model reduce to systematics (PR4/NPIPE reprocessing does shrink the excess to 1 to 2 sigma) or to a statistical fluctuation unlucky enough to organize itself by angular scale.
The standing is a quieter cousin of the lensing-amplitude anomaly with the same unresolved core: ACT DR6 and SPT-3G show better internal consistency, supporting the systematics reading, yet the Planck split has never been traced to an identified instrumental cause, and the pattern that produced it, excess smoothing entering only where lensing matters, remains in the PR3 likelihoods that anchored a decade of cosmology. CMB-S4 will measure the damping tail well enough to close the question.