rosatisoft/axiomatic-criterion-engine: Axiomatic Criterion Engine (ACE) — Ontological Discernment Engine

Zenodo-ready Metadata Abstract Large language models (LLMs) operate primarily as probabilistic systems optimized for linguistic prediction. While effective for pattern completion, this paradigm lacks an intrinsic mechanism to evaluate ontological validity, leading to semantic entropy, hallucinations, and inefficient computational allocation under ambiguous or adversarial inputs. This work introduces Axiomatic Ontological Discernment, implemented through the Axiomatic Criterion Engine (ACE), a deterministic pre-inference layer grounded in The Axiom of the Absolute. ACE evaluates inputs according to ontological coherence, contextual domain, and evidential grounding prior to probabilistic generation. Rather than filtering content or enforcing policy constraints, the system establishes whether an input is eligible for inference based on its alignment with objective reality, explicit fictional context, symbolic language, or speculative reasoning. The proposed framework formalizes a clear distinction between truth claims and non-factual domains, preventing category collapse while preserving creative and fictional expression. A documented case study demonstrates how fictional narratives are validated as meaningful symbolic constructs without being misclassified as factual assertions, thereby reducing hallucination without suppressing imagination. Architecturally, ACE reorders AI pipelines by introducing judgment before generation. Empirical deployment indicates significant reductions in token usage on entropic inputs and improved semantic stability across models, independent of model size or training regime. This work positions axiomatic discernment as a foundational requirement for scalable, coherent artificial intelligence systems, enabling efficiency, safety, and interpretability through ontological structure rather than probabilistic correction.