Modelling results for D|$_{\rm 1}$| (source = |$s_{\rm 1}$|⁠, |$P_\mathrm{ f}$| = 12 per cent, i  = 90 deg; left) and D|$_{\rm 8}$| (source = |$s_{\rm 2}$|⁠, |$P_\mathrm{ f}$|  = 3 per cent, i = 45 deg; right) in the lowest frequency channel (i.e the channel with a frequency of 2 GHz), as an example (note that we simultaneously model all channels). Panels contain the ground truth (left column), recovered (middle column), and residuals (right column) for the RM (upper row), the projected line-of-sight magnetic field |${\boldsymbol {B}}_\mathrm{LOS}$| (middle row) and electron number density |${\boldsymbol {n_{\rm e}}}$| (lower row), averaged along the line of sight. In the residual column, we quote the fractional RMS difference between the input and recovered quantities, computed within the region where the lensed images are at least 20 per cent of the peak surface brightness (black contours). This illustrates the inherent degeneracies between |${\boldsymbol {B}}_\mathrm{LOS}$| and |${\boldsymbol {n_{\rm e}}}$|⁠. The RM is recovered to within a few per cent, but the uncertainties on |${\boldsymbol {B}}_\mathrm{LOS}$| and |${\boldsymbol {n_{\rm e}}}$| individually are much higher.