NILC residuals for SWIPE+Planck dataset with the d0s0 Galactic model. On the left: Mean angular power spectra of NILC foregrounds (solid) and noise (dashed) residuals over 200 simulations. On the right: Posterior distributions of an effective tensor-to-scalar ratio fitted on the foregrounds residuals in the case of half de-lensing (A_L = 0.5). For the estimation of the posterior a binning scheme of Δℓ = 15 has been used to make the angular power spectrum Gaussianly distributed (see Sect. 4.2). On the top: Blue and red lines represent the results when E − B leakage is corrected with, respectively, the iterative and inpainting recycling methods (rit-NILC and rin-NILC). On the bottom: Results when the ZB- (blue) and hybrid ZB-NILC (red) methods are applied. The adopted binning scheme is Δℓ = 6 for 5 ≤ ℓ ≤ 28 and Δℓ = 15 for ℓ ≥ 29 on the top left, while Δℓ = 6 for 2 ≤ ℓ ≤ 19 and Δℓ = 15 for ℓ ≥ 20 on the bottom left. In both cases the chosen binning strategy is just for visualisation purposes. Everywhere the grey region represents the primordial tensor BB angular power spectrum for r ∈ [0.015, 0.03]. The lower bound represents the targeted upper limit at 95% CL by LSPE in case of no detection, the upper one the targeted detection at 99.7% CL. The spectra have been estimated without any masking of the regions most contaminated by foregrounds. The final f_sky is 32% in the case on top, because 4% of the pixels closest to the borders are masked to avoid residual E–B leakage effects, and 28% in the case on the bottom due to the needed apodisation of the mask.