Mean r with (16,  84)% credible interval from 500 simulations. We apply the three nominal component separation pipelines (plus extensions) to simulations with four foreground scenarios of increasing complexity. We assume a fiducial cosmology with r = 0 and A_lens = 1, inhomogeneous noise with goal sensitivity and optimistic 1/f noise component (dot markers), and inhomogeneous noise with baseline sensitivity and pessimistic 1/f noise component (cross markers). We note that the NILC results for Gaussian foregrounds are based on a smaller sky mask, see Appendix B.