Silently skipped 3 pages in a scanned PDF
What happened
Pages with low contrast were not processed and content was omitted from the summary.
The catch was triggered by a content freshness check during retrieval.
Evidence summary
- Catch ID
- CTCH-4P2M
- System
- Invoice Document Pipeline
- Detected
- May 18, 4:32 PM
- Severity
- High
- Status
- Resolved
- Repair
- —
Retrieval context
- User query
- “What is our policy on expense report approvals?”
- Response timestamp
- May 19, 9:14 AM
- Model
- gpt-4.1
- Top source
- Policy on Returns (v3.2).pdf
- Source status
- Deleted from approved set 97 days ago