Notes. With reference to Table 2, the bias:SOM column gives the SOM trained against KiDS-Bright to infer δ_OR (Eq. (5.3)) and inform the density-field bias applied to 30 FLASK realisations of KiDS-Bright-like data. The recovery:SOM column gives the SOM trained against those biased FLASK mocks to produce organised randoms (see Sect. 5.2 for details). On average over the 30 FLASK realisations, the Biased and Recovered columns then give the absolute deviation from the unbiased clustering signal | w_{biased/recovered} − w_unbiased | in units of the uncertainty in the unbiased signals σ, averaged over the intermediate scales 7 − 100 arcmin. Non-uniformity of biased column values with self-similar bias:SOM setups arises stochastically due to our probabilistic application of biases to FLASK realisations, and all variations are at < 0.1σ. The 800Ares2 configuration illustrates a failure mode of our organised randoms methods, where the resolution and scale sensitivity of the setup are too high, resulting in over-fitting to cosmic structure and a consequent over-correction of the clustering signal.