IQRAX: Deterministic Authority, Remediation at Source, and an Eight-Property Controllability Conformance Profile for Enterprise AI

Current capability benchmarks measure what a model can do. Boundary behaviour evaluations also exist, notably for tool-using agents under adversarial prompting, but we are not aware of an instrument that states, as conjoint normative requirements, that authority must be invariant under capability change, that acceptance and authorisation must be separable and separately evidenced, and that a repair must be shown to cover a predeclared population. We propose an 8-property controllability conformance profile that an enterprise, a laboratory, or a standards body can run against any deployed system, stating each property as a normative requirement with a declared conformance target and a yes-or-no test. We then report a working estate built to satisfy it. IQRAX separates qualification, authority, and acceptance from the proposing intelligence: an enterprise-controlled persistent estate combines Qualified Intelligence, Deterministic Authority, and, where an approved sealed process applies, Deterministic Processes that produce evidence for independent compliance verification. Defects are classified, located, and repaired at source rather than at the instance, and every repair is challenged and externally re-accepted. We evaluate the controlled system through a longitudinal engineering case study using operational records, designed challenges, and dated intervention cases. In the mature operational window, 116 of 2,036 response groups were refused, with 126 findings across 24 control codes. Across 4 phases, live group refusal rates declined from 33.94% to 5.70% as the control set was armed. Eleven selected cases document behavioural correction. The findings support specified boundary containment, recorded correction, and governed improvement of the controlled system; they are an observational association, not isolated causation and not a model-weight claim.