Simulation-validated, software-enforced authority-arbitration reference architecture for financial-sector AI decision systems, aligned to the U.S. Treasury Financial Services AI Risk Management Framework (FS AI RMF) under the implementation framing of Executive Order 14179. An eight-stage AUTHREX pipeline (VALIDATE, SATA, ADARA, MAIVA, HMAA, FLAME, ERAM, CARA) with a four-tier HMAA authority model, a population-state coordination model across account, device, payee, and IP-cluster history, a SHA-256 canonical-form evidence chain, and a retrospective stigmergic swarm-review module that recovers coordinated rings the per-transaction path clears. Reference authority node: 36 hardware components, 33 electrical connections, 32 mechanical connections, approximately $9,228 BOM. TRL 3-4 (simulation) / TRL 2 (hardware); synthetic data only; not deployed in any financial institution. The reported recall is an actionable-risk triage measure, not an empirical fraud-detection rate. Published as fundamental research under CC BY 4.0.