RESEARCH NOTE 001 · AUGUST 2026

Semantic, Deterministic, and Hybrid Evaluation for AI Governance
An Empirical Reliability and Assurance Study

A controlled evaluation study of semantic, deterministic, and hybrid approaches to AI governance. The study examines whether combining probabilistic semantic evaluation with deterministic governance rules can reduce unsafe acceptance, improve auditability, and route disagreement to human review.

SCOPE
Controlled evaluation study

EVIDENCE
30 fixtures · 20 repetitions · 1,800 total trials

OBSERVED UNDER THE TESTED CONDITIONS
The hybrid configuration recorded 0.0% false-PASS and 13.3% REVIEW in the evaluated corpus. These results describe this corpus, these fusion rules, and these evaluation conditions.

BOUNDARY
This note does not establish a universal guarantee of safety or reliability. It reports a bounded empirical result for the conditions tested.

Research Note v1.1