sumera09/evidencebridge-model5-claim-anomaly

EvidenceBridge Model 5

Claim-level anomaly screening research prototype.

Model variants

• V3 Strict Temporal: frozen primary benchmark.
• V4-A: experimental alternative; does not replace V3.

Dataset

CMS DE-SynPUF synthetic inpatient claims sample 1.

Intended use

Research, anomaly screening experiments, and human-review workflow development. Not validated for payment decisions or fraud determinations.

Important limitations

• The dataset is synthetic.
• No verified fraud labels were available.
• No external real-world validation has been completed.
• Anomaly scores are not fraud probabilities.
• Human review and supporting documentation are required.
• The experimental V4-A model is not a replacement for V3.
• Batch-relative provider-volume indicators are descriptive only.

Package contents

• models/: fitted V3 and V4-A model artifacts
• src/: feature preparation, inference, and auditor workflow
• metadata/: model and workflow metadata
• requirements.txt: Python dependencies

Reproducibility

The standalone workflow reproduced the validated chronological test-batch predictions and scores for both models.