device and network signal model, so you collect context once and reuse it across calls. For the conceptual differences, see Identity intelligence.
The exact signals, thresholds, and weights behind these products are intentionally not published — that’s what keeps them hard to game. Build against the documented outputs (
trustScore, decision, anomalies, reasonCodes); treat the scoring internals as opaque.The shared payload
Each call accepts the customer’smobileNumber, your metadataId, a sessionId, and two signal objects:
Trust — a score for the decision
trustScore (0–100), a confidence band, a recommendation (ALLOW/REVIEW/BLOCK), and a high-level breakdown. Trust scores an identity from its history with you, so it’s strongest on returning customers; drop trustScore into the rules you already run, or branch on recommendation directly.
See POST /v1/api/trust/score.
Shield — a decision for the action
decision — ALLOW, CHALLENGE, REVIEW, or BLOCK — with a riskScore, reasonCodes, and consistency signals (deviceConsistencyLevel, operatorConsistencyLevel, sdkVerified). Wire the decision straight into your gate:
POST /v1/api/shield/assess.
Detect — patterns over time
anomalies (categories like retry bursts, one device or installation linked to many numbers, IP concentration, and operator/country failure spikes), an anomalyScore, and a recommendedAction (ALLOW/CHALLENGE/REVIEW/BLOCK). Detect is ideal for fraud-ops dashboards and asynchronous review rather than blocking a single request.
See POST /v1/api/detect/analyze.
A common composition
Call Shield inline for the gate, and run Detect continuously to catch slow-building abuse your inline checks won’t see in a single request.Each product must be enabled for your account. A disabled product returns
400 Bad Request with a message. Contact connect@authmatech.com to enable Trust, Shield, or Detect.
