Computational models of speeded decision-making, including the urgency-gating model and joint modelling approaches, assume that adaptation to speed-accuracy demands is homogeneous across individuals, an assumption rarely tested because model-based designs are not built to detect qualitative between-subject variation. To address this gap, this thesis introduces a non-model-based approach deriving per-participant behavioural summary features for clustering analysis, probing individual differences in urgency adaptation. This approach was applied to a probabilistic selection task with trial-wise speed-accuracy cues from Miletić et al. (2025; n = 37). A manipulation check confirmed the expected speed-accuracy trade-off: speed cues produced faster but less accurate responses, both effects supported by strong Bayesian evidence, alongside a positive correlation between RT and accuracy shift scores. The homogeneity assumption was then tested directly by deriving four per-participant features (RT shift, accuracy shift, cue sensitivity slope, and within-speed RT variability) and applying k-means clustering with bootstrap validation. Silhouette analysis favoured a two-cluster solution over a continuous structure, indicating that participants displayed distinguishable behavioural signatures of adaptation rather than varying smoothly along a single dimension of urgency. Bootstrap resampling confirmed stability only for the smaller cluster (n = 9); the larger cluster (n = 27) was unstable, limiting confidence in profile robustness. The stable cluster showed larger, transient urgency adaption; the unstable cluster showed smaller, sustained adaptation. These findings indicate a detectable but not fully robust departure from strict homogeneity, qualifying Miletić et al.'s (2025) latent urgency parameter and motivating further non-model-based work to characterize behavioural adaptation profiles.