Does Planck 2015 polarization data favor high redshift reionization

We study the relationship between signatures of high redshift ionization in large-angle CMB polarization power spectra and features in the Planck 2015 data. Using a principal component (PC) ionization basis that is complete to the cosmic variance limit out to $z_{\mathrm{max}}=30,40,50$, we find a robust $>95\%$ CL preference for ionization at $z>15$ with no preference for $z>40$. This robustness originates from the $\ell \sim 10$ region of the data which show high power relative to $\ell \le 8$ and result in a poor fit to a steplike model of reionization. Instead by allowing for high redshift reionization, the PCs provide a better fit by $2\Delta \mathrm{ln}\mathcal{L} = 5-6$. Due to a degeneracy in the ionization redshift response, this improved fit is due to a single aspect of the model: the ability to accommodate $z>10$ component to the ionization but does not constitute a highly significant detection on its own. For models that accommodate such a component, its presence is allowed and even favored; for models that do not, their poor fit reflects statistical or systematic fluctuations. These possibilities produce very different and testable predictions at $\ell \sim 15-20$, as well as small but detectable differences at $\ell>30$ that can further restrict the high redshift limit of reionization.