Results of splitting of the real test data DATA-I, into g-th and h-th subtests of equal (=25) number of items, using minimisation of S, i.e. minimisation of the absolute difference between sum of components of item score vectors in the 2 subtests. Lower left: plot of value s of S at the t-th splitting of the test into the g-th and h-th subtests, where the current splitting index t : =25 (ℓ − 1) + j, with l the current iteration number, and j the current swap number; ℓ = 1, … , 50, j = 1, … , 25. An iteration comprises 25 distinct swaps, where a “swap” is defined for this method, in Definition 1. Lower middle: plot against t of value s_ρ of S_ρ which is the inner product of item score vectors of g-th and h-th subtests. Lower right: plot of linearly transformed S and S_ρ values, against splitting index t, to empirically verify the equivalence between maximisation of S_ρ and minimisation of S (in thin solid lines). scaling and translation of S and S_ρ are undertaken to allow the transformed variables to be plotted within a given interval that allows for their easy visual comparison. Also, to enable such visualisation, we focus on a sub-interval of the values of t relevant to this run (≥500). Upper right: plot of reliability r_tt as computed by our definition (Eq. (4)), against splitting index t. Upper left: histogram of the r_tt values obtained from this run that attains splitting of the given DATA-I test dataset, using minimisation of S. Upper middle: histograms of examinee scores obtained at the last accepted swap, in the g-th (in dark solid lines) and h-th subtests (in grey, or red in the electronic version of the paper), identified in this step.