PayPal AI Trust Engine
ML anomaly screening — backend code, not this dashboard
The separate TrustGuard transaction path runs a scikit-learn IsolationForest (`trustguard-isolation-forest-synthetic-v1`) using a deterministic synthetic baseline dataset (`synthetic-normal-transaction-features-seed-20261104-v1`). Its anomaly score is not a calibrated fraud probability; this model is not production validated. `POST /v1/agent/authorize` only validates Skyfire tokens and does not invoke transaction scoring. This static page runs no model inference.
In the TrustGuard path, a high score or scoring/model failure places the already-authorized transaction on durable HOLD before provider dispatch. PayPal dispatch remains disabled by default.
Risk explanation — static visualization
This view only displays explanation fields already present in sample data or a locally selected screening API JSON response. It does not independently verify imported data, run model inference, request tokens, call an external provider, or override a backend decision.
No local demo report loaded. Imported files are displayed as unverified local evidence.
Transaction evidence
Demo statistics
Demo audit events
Source data timestamp: unavailable