B-mode CMB power spectrum reconstruction after the E–B leakage correction with the application of the recycling method and the extensions presented in Sect. 3.1.1. On the left: Mean angular power spectrum over N_sims = 200 CMB-only simulations that include lensing and a primordial tensor signal with r = 0.01. Solid lines represent the power spectrum of exact B maps (input), which were reconstructed with a full-sky B decomposition of Q and U and then masked for power spectrum estimation; power spectra of leakage-corrected maps (output LC) are shown with circles and diamonds for SWIPE and SO-SAT, respectively; the dashed lines are the leakage residuals after the correction; and the dashed-dotted lines (if present) are the CMB B modes without any leakage correction (output noLC). On the right: Effective tensor-to-scalar ratio associated with the absolute difference between leakage-corrected and exact angular power spectra. The top panel shows the results when the recycling method was applied to CMB B-mode maps for the SWIPE (red) and SO-SAT (blue) footprints; the middle panel reports those obtained when an iterative B decomposition with no (red), one (green), two (cyan), or three (magenta) iterations were performed. In the bottom panel, we compare the performance of iterative decomposition with three iterations (magenta) and diffusive inpainting (orange) when correcting the residual leakage. The middle and bottom panels present results only for SWIPE. The adopted binning scheme is Δℓ = 3 for ℓ ≤ 4, Δℓ = 6 for 5 ≤ ℓ ≤ 28, and Δℓ = 15 for ℓ ≥ 29 for the spectra computed in the LSPE-SWIPE patch, while a constant Δℓ = 15 for those estimated in the SO footprint. The error bars highlight the uncertainty on the mean at 1σ, which was estimated from the dispersion of the angular power spectra of the simulations divided by . Readers can refer to the main text for details.