Where The Governance Lives

Where the Governance Lives is a companion paper to Decision Justification Engineering (DJE), Executive Summary by Sandy Fitzburgh of Truth-Machine (October 2026). The paper examines Fitzburgh's empirical findings from a governed AI process used to predict residential real estate sale prices in a Florida community, where a general-purpose AI with structured evidence and human oversight achieved 1.22% mean absolute percentage error retrospectively and 0.80% error prospectively, outperforming dedicated automated valuation model providers such as HouseCanary (2.7% median error). The paper agrees with Fitzburgh's core findings and extends them by identifying the structural gaps his work exposes: the absence of a governed boundary for evidence representation (demonstrated by a 26.2% execution-exclusion rate under unstructured inputs versus 2.2% under structured inputs), the need for a third governance state beyond binary pass/fail qualification, the requirement for tiered persistence channels separating raw observation from confirmed outcomes, structural drift detection that produces governance signals rather than metrics alone, distributed accountability across distinct operational roles, and candidate-level governance operating independently from system-level controls. The paper identifies that the operator's chair between human authority and machine execution remains unspecified in the current framework and addresses what that specification requires. The paper further identifies that Fitzburgh's execution integrity findings carry safety-critical implications for physical systems, where post-process exclusion gates do not exist and execution failures cannot be recalled, particularly in environments where children are present. All claims are independently verified. No proprietary terminology is used.