Mock kinematic data and model comparisons. First: IDEAL mock IFU data of RXJ1131−1231, at the JWST NIRSpec resolution. Following the procedure in Y20, the data have been mocked up by means of the best-fitting lensing-only model and a set of random dynamical parameters. Second: Normalised residuals (in absolute values) of the best-fitting model without a mass-sheet component (λ_int ≈ 1.0), showing that the mock data are remarkably well recovered, mostly within the 1σ uncertainties. Third: Normalised residuals (in absolute values) of the best-fitting model with a (fixed and significant) mass-sheet component (λ_int ≈ 0.85), which is at the edge of the 3σ uncertainty bands. Fourth: Same as the two previous panels, but for the best-fitting model with a SPEMD. Various mock data points are not recovered, even within the 2–3σ uncertainties. The χ² difference between this model and the best-fitting COMPOSITE model without a sheet (second panel) is Δχ² ≈ 155.