VirtualCompanySim: Explainable personality parameterization for multi-agent career exploration simulation

This paper presents VirtualCompanySim, a reusable research software package for exploratory multi-agent career simulation and role-awareness training. Its core contribution is explainable personality parameterization: MBTI dimensions are converted into inspectable action weights and multipliers rather than opaque prompt attributes. The software models a three-floor virtual company with 16 agents that make personality-conditioned decisions within a simplified organizational setting, incorporating a four-layer decision architecture with deterministic fallback, persistent social memory, adaptive task generation, and a dynamic event engine. Reproducible experiments yield a 98.0% MBTI behavioral consistency rate and a 65.33% loss in type distinctiveness upon removing the mapping. The package includes diagnostic APIs, fixed-seed scripts, and stored results for inspection and reuse.