Constraint Sentinel and Constraint-Aligned Priming: Safeguarded Use of Large Language Models under the Epistemic Boundary Conditions

This record defines a formally constrained use of large language models as epistemic hygiene instruments under the Epistemic Boundary Conditions (EBC) framework. It contains three documents that together specify a complete, closed role system: • Constraint Sentinel — an orthogonal practice axis under EBC for detecting ontology smuggling, scope drift, definitional drift, and other violations at the meta-descriptive level.• Constraint-Aligned Priming (CAP) — a mandatory operational protocol that binds a language model to the Constraint Sentinel role, enforcing non-authority, non-generation, output restriction, and human responsibility preservation.• AI as Constraint Sentinel — a use-case paper demonstrating how this role and protocol allow LLMs to function as safeguarded epistemic infrastructure rather than as sources of knowledge or judgment. The system is tool-agnostic and model-agnostic. It does not align models to truth, values, or ontology; it fixes a strictly limited epistemic role relative to declared EBC constraints. All epistemic authority and responsibility remain with the human observer. This record should be cited as a single conceptual object.