Predictive Emotional Selfhood (PESAM): A Unified Variational Account of Diachronic Self-Coherence under Realistic Scoring in Artificial Agents

This record contains the camera-ready manuscript (v1.1.1) of“Predictive Emotional Selfhood (PESAM): A Unified Variational Account ofDiachronic Self-Coherence under Realistic Scoring in Artificial Agents.” This version reflects a reproducibility-aligned rebuild of the PESAM frameworkand supersedes earlier camera-ready releases by incorporating a realisticscoring protocol and fully audited results. Key contributions of this version include:- A unified variational formulation of emotional selfhood as diachronic self-coherence under affective volatility.- Formal specification of three computational mechanisms: Affective Precision Control (APC), Self-as-Hyperprior (SaH), and Affective Homeostatic Objectives (AHO).- Introduction of a realistic scoring regime in which policy selection is belief-based, while internal stress dynamics are updated against environment-grounded contingencies, preventing belief-consistent bookkeeping.- Systematic lesion studies demonstrating that SaH and AHO are non-negotiable for stable self-regulation, while APC induces a principled calibration–performance trade-off.- Paired statistical evaluation with fixed random seeds, permutation tests, and bootstrap confidence intervals. All figures, tables, and quantitative claims in this manuscript are alignedto the PESAM v1.1.0 audit archive, which provides fixed configuration files,archived random seeds, episode-level outcomes, and SHA256 integrity logs forfull auditability and deterministic reproducibility. Supporting data, code, and audit materials are openly available at:https://doi.org/10.5281/zenodo.17999558