facts-extraction-normalization

To extract and normalize patient facts from clinical visit transcripts, this Skill processes current visit evidence to generate candidate facts, deduplicates entries, and normalizes them into a consistent longitudinal view for human review.

Longitudinal fact deduplication in healthcare data workflows identifies and removes duplicate patient facts across multiple visits. This Skill normalizes extracted facts like meds, goals, and incidents to ensure deterministic comparisons and auditability.

Yes, you can use extracted clinical facts to guard against RAG hallucinations. This Skill persists approved updates to the patient facts table and exposes active facts to the RAG context pack, discouraging memory-based hallucinations.

To manage fact provenance to OCR regions and transcript segments, this Skill links extracted candidate facts directly to today's transcript and OCR outputs. This traceability enforces a write-back workflow requiring human-in-review confirmation.

High-stakes, medium-stakes, and low-stakes clinical facts are handled through a controlled write-back workflow. The Skill proposes insert, update, or resolve actions for each fact type, requiring human-in-review confirmation before persisting changes.

The best way to maintain a reviewable facts table across patient visits is to formalize extraction and normalization into a controlled pipeline. This Skill processes visit evidence, proposes updates for human review, and persists approved facts for deterministic comparisons.