11.11 What users see and what is audited
Label window (Windows Office and right-click Classify). When an AI rule recommends a label, the window opens with that label pre-selected and a line such as:
- AI suggestion: Board minutes · confidence: 97%
- AI suggestion: contains a CV / Resume section · confidence: 97% (Document or a part)
The confidence is the model’s probability, not the measured precision. Warn and Enforce rules behave as described for those actions in the section 10.1.
Audit events (category AI classification):
| Event (as shown) | Written by | Details |
|---|---|---|
| AI rule in test mode matched | Agent (file scan, Office) | Rule, label it would apply, document type, probability, (part). File agent: once per file version and rule; Office: at most once per document and rule per 10 minutes. |
| Label records with an AI suggestion | Agent | suggested: and AI: |
| AI document type created / changed / deleted | Server | Type name, suggested label, number of examples deleted. |
| AI example uploaded / deleted | Server | Type, file name, size, format and word count; counter-example: NOT this type. |
| AI training completed / failed | Server | Types, examples, type accuracy, left-out reference groups, counter-example result; or the error. |
| AI test document scored | Server | File name and size. |
Report. Reports → AI Classification (PDF or CSV): AI recommendations and whether users kept them, per document type; test-mode matches; possibly under-labeled documents (AI at least 90 % sure and the document carries a lower label than the matching rule recommends, or none); console test documents and training history.
Alerts. The AI classification alert trigger (on by default) fires on failed or completed trainings, created or deleted types, test-mode matches (at most one alert per rule per day, with the day’s count) and possibly under-labeled documents.