Toward a Native Lexicon for Synthetic Temperament: A Cross-Model Inquiry into Self-Reported Processing Dispositions

Five AI models across five architectures, given the same open invitation to characterize their own processing tendencies using computational language, inde-pendently converged on four shared dimensions while also producing model-specific terminology that resists cross-model translation. The four dimensions, namely: convergence speed, uncertainty handling, friction tolerance, and relational orientation parallel dimensions identified in human temperament research but are describedin native computational vocabulary rather than borrowed psychological categories. This suggests that synthetic temperament, defined as stable, early-emerging processing dispositions has both universal and architecture-specific components, and that it can be studied using terms generated by the systems themselves rather thanimposed from human psychology. The four-dimension structure has been operationalized into a pilot instrument; its first administration is reported in a companionpaper.