Engram: Execution‑Realized Cognition in Modern AI Systems - A Unified Framework for the Execution Substrate and Its Falsification Criteria

This unified edition consolidates two complementary research papers on execution‑realized cognition in large language models: Engram: A Framework for the Execution Substrate of Modern AI Models [13] and A Falsification Framework for Execution‑Realized Cognition in LLMs [13]. Together, these works articulate the Engram layer—the execution substrate through which trained functions become realized functions during inference—and formalize the scientific conditions under which execution‑dependent cognitive effects can be confirmed or falsified.The first component introduces Engram as a conceptual framework describing how precision formats, memory hierarchy, kernel scheduling, numerical stability, and hardware topology shape the realized behavior of modern AI systems [1–12]. The second component provides a neutral, operational falsification framework that refines seven core claims of the Engram hypothesis into testable scientific statements, each with explicit disconfirming conditions and minimal experimental protocols [14–24].This consolidated edition offers a single archival reference for the conceptual foundations and falsification criteria of execution‑realized cognition. It preserves the original scope and intent of the two papers while harmonizing their structure and terminology for long‑term citation and reproducibility.This Zenodo record provides the unified archival edition of the Engram framework and its falsification criteria