Capabilities · Classification
Fraud Detection
HealthyIn-context predictionEndpoint fraud-detectionProduction v7Challenger v8
Production constraints
Constraints drive candidate selection and deployment decisions.
P95 latency
<
ms
Availability
>
%
False positive rate
<
%
Cost / 1K
<
$
Challenger v8: satisfies all constraints
Candidate eligibility preview
✓Managed AAUC .947 · 42ms · $0.18Selected
✓Managed BAUC .951 · 71ms · $0.41Eligible
✓Managed CAUC .932 · 15ms · $0.08Eligible
Capability
- Endpoint
- fraud-detection
- Task
- Binary classification
- Objective
- Detect fraudulent transactions
- Mode
- In-context prediction
- Backend
- Managed by Zero Grad
- Examples
- transactions.parquet
- Available history
- 8.4M observations
- In-context examples
- 4,096
- Target
- is_fraud
- Production version
- v7
By default you manage capabilities and versions, not models. Turn this on to see which engine family backs each candidate.
Optimization mode
Preview
Optimize for
Allowed actions
When a better model appears, Autopilot evaluates it against your data, shadows it, and rolls it out within your constraints — your application changes zero code.
Where should this capability run?
Enterprise controls
Data residency
Pin data and inference to a region (US, EU)
Private networking
PrivateLink / Private Service Connect endpoints
Customer-managed encryption keys
Bring your own KMS keys
Audit logs
Every deployment, promotion and config change, exportable
SSO / SAML
Okta, Google Workspace, Azure AD
RBAC
Separate who can deploy, promote and change guardrails
Dedicated endpoints
Isolated capacity with reserved throughput
Custom retention
Configure how long inputs and predictions are kept
Compliance documentation is shared directly with design partners on request.