Comparison of the un-normalized modulation power with various combinations of frequencies. The middle plot shows the results used for our main τ_NL result, since it removes the significant largely-quadrupolar signal from anisotropic noise misestimation seen in the two other plots. The noticeable difference in the odd octopole signal between channels indicates that the residual signal in 143 × 217 GHz is unlikely to be physical, but we cannot currently identify its origin.