Confidence Threshold
ConceptsA confidence threshold is a minimum confidence score that you set to decide how to treat results. For extraction, values below the threshold might be flagged for human review, excluded from automated downstream use, or returned with a warning. Thresholds help balance automation with quality and reduce risk of acting on low-confidence extractions.
In DocLD
- Extraction — Each extracted field has a confidence score; you can set a threshold (e.g., 0.8) and route low-confidence extractions to review or ground truth workflows.
- Workflows — A workflow step might branch on confidence (e.g., "if confidence below 0.7, notify for review").
- Tuning — Compare thresholds to ground truth to choose a level that matches your accuracy and review capacity.
Setting the threshold is a product or integration choice; DocLD returns scores so you can apply your own thresholds.
Related Concepts
Confidence threshold uses confidence score from extraction. Ground truth helps calibrate thresholds. Bounding box and citation help reviewers verify low-confidence results.