Epistemic Gating for Text-Only Language Models: An Experimental Instantiation of Computable Embodied Constructor Theory

This record contains a minimal, reproducible experimental artifact demonstrating how the epistemic dynamics of Computable Embodied Constructor Theory (cECT) can be instantiated as an external epistemic gating layer for text-only language model interaction. The experiment implements a language-specific discursive sensor (English heuristics) that supplies epistemic inconsistency signals to a fixed cECT-style precision update mechanism. Language model outputs are treated as hypotheses with revisable epistemic precision (ω), which is updated without reinforcement learning, reward optimization, inductive training, or access to internal model states. A scripted dialogue scenario induces sustained contradiction and resistance to revision, allowing direct comparison between an adaptive-precision condition and a fixed-precision control. The artifact does not aim to assess factual correctness, safety, or normative alignment; rather, it isolates epistemic alignment as a falsifiable dimension of system behavior. This work does not modify or extend the underlying cECT theory. The original cECT framework and core implementation are published separately and are referenced as related identifiers. The present record should be understood as a language-specific experimental instantiation, with language-agnostic sensors and live language model integration identified as future work.