F = (D^I) × A
Constitutional AI decision model — formal specification of the four-variable equation governing all RCT Ecosystem outputs.
The FDIA Equation (F = (D^I) × A) is a constitutional AI governance model that makes human authorization a hard mathematical gate rather than a soft preference. This second-version specification formalizes the four variables, proves the kill-switch invariant (A=0→F=0), and documents the operational semantics used by the RCT Ecosystem in production.
F (Future) is the constructed output — not a prediction, but a deliberately designed result. D (Data) is the quality-weighted knowledge base and serves as the exponential base, so higher data quality produces exponentially better outcomes. I (Intent) is the exponent: explicit human specification that determines how powerfully data is amplified. A (Architect) is the human-in-the-loop multiplier — the constitutional gate that gives AI outputs their meaning and accountability.
When A = 0, the equation collapses to F = 0 regardless of data quality or intent strength. This is not a preference or a soft guideline — it is a hard mathematical invariant. The RCT Ecosystem enforces this with deterministic replay: given the same inputs and A=0, the output is always zero. This property makes the system auditable and safe by construction.
In production, the FDIA pipeline evaluates each request against all four variables before invoking any model. Data quality scores are computed from provenance metadata and retrieval confidence. Intent strength is assessed from specification completeness and goal disambiguation. Architect authorization is a runtime token that can be revoked at any time. The pipeline is fully observable: all four variable scores are returned in the audit trail.
Version 2 formalizes the 8-Dimension scoring model for D and I variables, introduces the geopolitical balance constraint used by SignedAI (3W/3E/1R), and adds the PDPA-compliant provenance dimension (dimension 8) to the Data variable. The kill-switch invariant remains unchanged from v1.