When Does Agent History Matter? A Critical Methodological Audit of IMR-20 to IMR-47 and a Falsifiable Framework for Functional Self-Relation

When Does Agent History Matter? A Critical Methodological Audit of IMR-20 to IMR-47 and a Falsifiable Framework for Functional Self-Relation Author: Karel HrubecDOI: 10.5281/zenodo.23892221 This paper presents a retrospective methodological synthesis of a sustained programme of synthetic AI-agent experiments (IMR-20 to IMR-47), investigating the relationship between intelligent adaptation, causal self-modeling, individual developmental history, and the possibility of historically constituted functional individuality. The central question is whether an agent's relationship with its own history and environment can become a causally significant source of future decisions, rather than merely supporting more efficient task execution. The experimental programme combines custom synthetic environments with instrumented evaluations in Crafter. Selected results demonstrate the benefits of improved predictive models, causal representation, and historically acquired capabilities. However, explicit self-monitoring, relational architectures, and internally maintained commitments generally failed to establish advantages over strong, simpler control systems. The paper introduces a methodological distinction between three independent dimensions: the causal presence of a mechanism, its incremental functional utility, and its explanatory distinctiveness. It further develops six operational criteria and a falsifiable experimental framework for investigating historically grounded functional self-relation. The accompanying open-science package provides archived experimental code, original results, research protocols, an evidence matrix, integrity manifests, and scripts for verifying archived data and reproducing selected statistical comparisons. Rather than claiming to demonstrate artificial consciousness or autonomous subjectivity, the study identifies the evidence required to distinguish competent adaptive behavior from stronger forms of functional self-relation. Its contribution lies in integrating positive and negative experimental findings into a transparent methodological framework that may support future research on AI agency, developmental autonomy, and the emergence of individual functional organization.