Audited Claims for the Semantic Deviation Research Program — The Glas Function: An External-Format Restatement

The Glas Function. An external-format audit of the Semantic Deviation Principle (Sharks 2026, v0.2 Final, DOI: 10.5281/zenodo.20250736) and its associated empirical protocols. Operating in the transparent-medium register, this paper returns a narrowed, citationally grounded, externally evaluable statement of the technical core. It does not amend the founding formulation. It does not depend on the institutional architecture that has accreted around the formulation. It is a standalone document — the Layer A skeleton without which the body of work cannot be taken seriously by the broader scientific community. What it does. Distinguishes three layers (A: technical core, B: philosophical interpretation, C: institutional architecture) and commits to engaging only Layer A. Diagnoses the load-bearing technical gap — the underspecified semantic field Ψ_t(C) — and proposes three canonical operationalizations (F1 closed-system continuation, F2 retrieval response, F3 citation graph) with pinned divergence functionals, temporal weightings, and statistical-power constraints per regime. Narrows the headline claim from "meaning is deviation" (universal ontology) to "meaning-bearing interventions produce durable trajectory restructuring under specified field operationalizations" (measurement architecture). Specifies a six-condition component-decomposition experimental design (Model-π, Model-Dev, Model-Coh, Model-Full) with the advance prediction that the provenance component will carry more independent uplift than the deviation component. Replaces philosophical anti-Goodhart commitments (the Vow, the Step 0 audit) with mechanism-design machinery (entropy-floor capping, provenance-weighted damping, saturation limits with operationalized τ thresholds, temporal coherence penalties, KL anchoring, adversarial judge validation, black-box judge replacement testing). Pre-registers the cheapest dangerous test — negative net signed deviation as the empirical signature of AI slop — with explicit corpus (GPT-wiki-intro + HC3), reference model (Llama-3.1-8B-Instruct), statistical test (Mann-Whitney U at α=0.05), and effect size of interest (Cohen's d > 0.5). Maps the program into contemporary literatures: DPO/IPO/RLAIF, reward hacking, mode collapse, mechanistic interpretability, semantic information, semantic entropy, causal inference, cultural evolution, diachronic semantic change, active inference, hallucination and attribution failure, recursive model collapse. Closes with a budgeted near-term roadmap totaling approximately $14,000-$19,000 across four experiments in the next twelve months. What it does not do. Does not claim that meaning is universally definable as deviation; does not claim the proposed operationalizations are uniquely correct or exhaustive; does not claim the canon-formation conjecture has been proven; does not claim the anti-Goodhart machinery is sufficient against all gaming strategies; does not claim the institutional architecture surrounding the SDP corpus is required to engage the technical core; does not claim Layer C terminology adds technical precision; does not replace existing deposits in the program. Series: EA-GLAS-01 (a new series within the Crimson Hexagonal Archive, designated specifically for external-format audit documents). Related deposits (full register in Appendix A): Sharks, L. (2026). The Semantic Deviation Principle, v0.2 Final. DOI: 10.5281/zenodo.20250736 Glas, N. (2026). The AI System as Closed-System Test Bed (MM-AI-01 v2.0). DOI: 10.5281/zenodo.20251738 Glas, N. (2026). Measuring Meaning in Retrieval Basins (MM-02 v2.0). DOI: 10.5281/zenodo.20251740 Glas, N. (2026). The Deviation-Optimized Language Model (MM-AI-02 v2.0). DOI: 10.5281/zenodo.20251742 Approximately 8,200 words. 36 references to published works in alignment, machine learning, computational linguistics, information theory, causal inference, and cultural evolution. The function has run. The output is this paper.