Behavioral
Six detectors run on every agent authorization against the agent's own fingerprint: confidence inflation, alternatives collapse, reasoning-length anomaly, vocabulary shift, amount escalation and MCC drift.
Mandate Labs began as the fraud layer for agentic commerce, and that stack is Ward's agent engine today. It identifies the agent, checks its intent against its mandate, authorizes on card and x402 rails and keeps a record you can audit.
Seven decision gates and two system steps sit in the authorization path. Every settlement and dispute outcome comes back as a label.
Registry, KYA Score, trust zones and a certified trust report an agent can present anywhere.
GREEN · AMBER · RED · CRITICAL 02 · authorizeThe gates on card and x402 rails, with advisory risk data in ISO 8583 fields.
APPROVE · STEP_UP · DECLINE 03 · recordHash-chained decision records with attested intent: audit trail, chargeback evidence, SAR/STR export.
chain verified · evidence pack 04 · learnSettlement and dispute outcomes label every decision, and the behavioral models retrain on them.
SETTLED → label attachedAn agent can pass every limit check and still make a bad economic decision. The Economic Decision Quality Score is our research answer to that, and it comes in two parts with different status.
Six detectors run on every agent authorization against the agent's own fingerprint: confidence inflation, alternatives collapse, reasoning-length anomaly, vocabulary shift, amount escalation and MCC drift.
The six-dimension rubric from the research paper: alternative analysis, confidence calibration, constraint alignment, temporal efficiency, value optimization and risk-adjusted reasoning. Not yet in production.
type: intent or type: agent to /decide, or let Ward route a payment to the agent engine when an agent initiated it.Bring a statement, a declines export or the case your team is stuck on.