Capabilities · Classification
Fraud Detection
HealthyIn-context predictionEndpoint fraud-detectionProduction v7Challenger v8
Fraud Detection
v7· Binary classificationIN PRODUCTION
Quality
- ROC AUC
- .947
- Precision
- 91%
- Recall
- 79%
- False positive rate
- 0.7%
- CalibrationECE 0.011
- Healthy
Evaluation
- In-context examplesauto-selected
- 4,096
- Evaluation examplesheld out, last 9 weeks
- 200K
- Temporal holdout
- ✓
- Customer-specific trainingpretrained models only
- None
Estimated production
- P50 latency
- 18ms
- P95 latency
- 42ms
- Cost / 1K predictions
- $0.18
Candidates
Pretrained foundation models, scored on your held-out examples and selected automatically under your production constraints
| Candidate | AUC | P95 | Cost / 1K | Status | Note |
|---|---|---|---|---|---|
| Managed A | .947 | 42ms | $0.18 | Production | Best quality within constraints |
| Managed B | .951 | 71ms | $0.41 | Challenger | +.004 AUC at 2.3× cost |
| Managed C | .932 | 15ms | $0.08 | Candidate | Fastest, needs no examples |