Bar chart comparing the run-time of a single evaluation of KiDS-SBI (above) versus a single evaluation of a simulation based on Joachimi et al. (2021) (below), both on a single core (N_shells = 19, N_tomo = 5 and N_side = 1024). Both suites of simulations use CAMB (Lewis et al. 2000; Lewis & Challinor 2002; Howlett et al. 2012) to compute the 3D matter power spectrum. For the reference simulations, we use the non-Limber projection built into CAMB with limber_phi_lmin = 1200 rather than Levin with ℓ_max, nL = 1200. We run FLASK (Xavier et al. 2016) rather than GLASS (Tessore et al. 2023) to compute the underlying matter and convergence fields of each of the 19 shells. Subsequently, we sample galaxies using SALMO in both cases, and then calculate the spatial two-point correlation functions, ξ_±(θ), rather than calculating the angular power spectra, $\stackrel{\sim }{C}(\ell )$. To calculate ξ_±(θ) in the reference simulations, we use TreeCorr (Jarvis et al. 2004).