Reporting feedback
If anonymizer misses an entity that should have been redacted, or redacts something that shouldn’t be, please report it. Feedback is opt-in — nothing is sent automatically.
In the web UI
- Click Report next to the entity (false positive) or highlight the missed text and click Report as miss in the preview
- A dialog appears with a copyable JSON payload
- Send the JSON to your team contact (Slack, email, ticket system) — the destination is your call
What’s in the payload
Each report contains:
- Category — what category was expected/detected
- Context hash — SHA-256 of the surrounding sentence (so duplicate reports collapse)
- Sanitized context — 5 words on each side with the entity itself replaced by a placeholder
- Detector source — which detector handled (or missed) the span
- Anonymizer version — for reproducibility
Never included: the raw entity text, the file content, the filename, your user info.
Why no telemetry?
Automatic feedback would mean the tool phones home — which contradicts the “zero network in core” promise. Feedback is a deliberate, manual action.
Where the feedback log lives
If you click Save locally, the report is appended to ~/.anonymizer/logs/feedback-YYYY-MM-DD.jsonl. This file never leaves your machine unless you explicitly export it (anonymize feedback export) or share the dialog payload.
To clear: anonymize feedback clear.
What happens after you send a report
The development team uses these reports to drive detector improvements via a structured feedback loop — analyze → reproduce as a failing test → improve detector → ship a patch.