A stability-first framework for replicable game personas

Player persona construction in game design is often criticized for limited reproducibility and sensitivity to analytic degrees of freedom, leading to unstable segmentations. This paper proposes a stability-first methodology for robust profiling from heterogeneous measurement blocks. The framework formalizes how inputs are integrated into a common profiling space, including measurement-consistent block representations and block balancing, and replaces single-index optimization with a multi-criterion selection strategy that prioritizes resampling stability and segment separation. The framework is demonstrated in a case study of players (N = 175) modeled in a joint space of motivation (Bartle) and immersive tendencies (ITQ). Model-based clustering with Gaussian mixtures quantifies assignment uncertainty. It enables uncertainty-aware reporting by distinguishing high-confidence core members from ambiguous boundary cases, thereby supporting more cautious interpretation in the presence of overlap. The resulting persona structure is then examined using held-out external criteria excluded from profiling and model selection, enabling a non-circular post hoc assessment of whether the derived personas retain interpretable differentiation beyond the profiling variables. Overall, the paper presents a transparent, reproducible user-research pipeline that emphasizes structural robustness, explicit uncertainty reporting, and post hoc assessment without circularity.