1. Coverage and missed entities
Mark the entities that should be removed in a representative test set. Compare detections with that reference and review both missed entities and unnecessary removals.
GRITREDACT / EVALUATION RESOURCE
Document redaction evaluation checklist
Define what a recipient may see, then evaluate detection, review, removal and the evidence retained for that purpose.
SENSITIVE SOURCE
↓ REVIEW & REDACT
CONTROLLED OUTPUT
Start with the intended use
Agree the policy before judging the output.
Identify the source formats, sensitive entities, permitted recipient and downstream workflow. Use fictional examples for initial discussion; agree a controlled route before using sensitive enterprise content.
Suggested evaluation checks
2. Review before permanent removal
Check how an operator corrects detections, applies the intended policy and approves the output. Include difficult scans, handwriting and related files where they matter to the workflow.
3. The exported artifact
Inspect the resulting document or image, not just the detection list. Test whether removed text remains recoverable through extraction and review the visible artifact. A text-extraction check alone cannot establish complete removal.
4. Useful context and audit evidence
Check the information the recipient still needs, document layout and consistency across related files. Review the audit record, input/output hashes and operator history.
Deployment and integration questions
Where does processing run?
For the offline GritRedact deployment, install Docker in your environment: processing uses local model inference with no network egress. A sales-assisted SaaS option is also available; discuss the hosting and data flow for the chosen deployment.
How will the output be used?
Agree the approved recipient and purpose before preparing content for QA, research, external processing or AI/RAG ingestion. Customer APIs are provided after onboarding; there is no public or self-serve API.
Evaluate for the specific purpose
Redaction does not replace authorization. If a person or system is not entitled to a record, block retrieval. Automated redaction alone is not a declaration of HIPAA de-identification; policies, validation and human review must match the intended use and recipient.
Existing enterprise evidence
Review the engagement stage as well as the workflow.
These existing anonymized case studies distinguish a production deployment from an enterprise pilot. They describe particular engagements rather than a guarantee for a different dataset.
Start with your workflow
Put your unstructured data to work.
Tell us what your team needs to achieve. We’ll help you choose the right products, work through your requirements and explore on-premises or SaaS deployment.
Request a working session