AUTHREX pipeline applied to financial-sector AI assurance · FS AI RMF reference model · deterministic V&V trace
TRL 3–4 RESEARCH IDLE T+0.000s
Run
Scenario
Seed
Fault Inject
Compare
Validate
Benchmark
Swarm
Current Transaction — Raw vs Observed
AUTHREX 8-Stage Pipeline
SATA → ADARA → MAIVA → HMAA → FLAME → ERAM → CARA → AUDIT
HMAA Authority Tier
T3
Autonomous
T2
Supervised
T1
Advisory
T0
Human Ctrl
Event Log
Audit Ledger
V&V Checks
Monte Carlo
Swarm Review
Full 64-char SHA-256 hash-chained records over canonical JSON. Each record links the previous hash — tamper-evident.
Live pass/fail against defined acceptance criteria. Run "Self-Test" for the full deterministic suite.
Statistical validation over N seeded trials with Wilson 95% confidence intervals.
No campaign run yet.
Retrospective stigmergic ensemble review over the ledger look-back window. Deterministic. Surfaces coordinated rings the per-transaction fast path cleared. Run the stream (try scenario E), then click "Swarm Review".
No review run yet.
Scenario Briefing
GovernanceENABLED
Active faultsnone
MAIVA ensemble5 nodes, majority ≥3 (Byz-perturbed)
FLAME budget T3/T2/T1/T050 / 5k / 60k / 900k ms
Session Metrics
0 txns
Cleared
0
Review
0
to human tiers
Blocked
0
freeze/reject
Median latency
p50 decision
p99 latency
tail decision
FLAME breaches
0
over budget
Actionable-Risk Triage
actual actionable vs flagged
Flagged
Cleared
Actual Risk
0TP
0FN
Benign
0FP
0TN
Precision
Recall
F1
Metric is actionable-risk triage (fraud OR operationally-risky), not empirical fraud detection.
Reproducibility
Active seed42
Txns processed0
Ledger head (SHA-256)
Run checksum
Same seed + same config → identical checksum. This is a deterministic replay check, not a production certification claim.
SCOPE & LIMITATIONS (read before citing). Research simulator (TRL 3–4) using synthetic data and a hand-specified behavioral model — not a trained fraud classifier, not connected to any payment system. v2.2 adds schema validation, raw-vs-observed traceability, population-state coordination modeling, actionable-risk triage metrics, deterministic self-tests, Wilson 95% confidence intervals, and a full 64-char SHA-256 canonical-JSON hash-chained audit export. The ledger demonstrates browser-based tamper-evidence; it is not a hardware-backed ECDSA/HSM signing service. No real customer or transaction data is used. Metrics are simulator-internal and must not be cited as empirical real-world fraud performance. Academic research artifact; no commercial deployment. © 2026 Burak Oktenli · AUTHREX Systems · ORCID 0009-0001-8573-1667 · Washington, DC · CC BY 4.0.
BLADE-FINANCE Governance Simulator v2.2 · AUTHREX runtime-assurance pipeline · FS AI RMF reference model
Burak Oktenli · AUTHREX Systems · ORCID 0009-0001-8573-1667 · Washington, DC · CC BY 4.0