Practical guide / sample v1.0
Your CSV totals can match while record IDs are wrong
Our scan-to-CSV test produced a useful warning. Local Tesseract OCR recovered every quantity, unit price and total correctly, yet five of six record IDs were wrong. For example, the source ID R001 became ROO1, with letters replacing zeros.
The six line totals still added to $323.87. A total-only check would have passed. A later join to an inventory or purchasing system could still fail because those IDs no longer match.
We scored the sample in two ways: exact cells in their original row/column positions, and separate record-ID integrity checks. The OCR route achieved 31 of 36 exact cells, but only one of six complete rows and one of six matching IDs. There were no manual repairs.
Before relying on extracted CSV data, check:
- Shape: expected columns, data-row count and extra headers
- Identity: missing, unexpected and duplicate IDs
- Values: exact dates, decimal points, minus signs and leading zeros
- Arithmetic: line calculations and overall totals
- Exceptions: rejected or unreadable rows, rather than a silent best guess
Do not automatically replace every O with zero. Some valid identifiers contain both letters and digits. Use an approved field rule or an authoritative reference, and retain the original value in an audit trail.
This is a clean 200-DPI synthetic scan with known cell positions. It demonstrates a failure mode, not how often a tool will fail on your documents. PDF.co and Adobe Acrobat remain untested.
Every ID and amount, side by side
| Expected ID | OCR ID | ID check | OCR amount |
|---|---|---|---|
R001 | ROO1 | O / zero mismatch | $25.00 |
R002 | ROO2 | O / zero mismatch | $26.97 |
R003 | ROO3 | O / zero mismatch | $105.00 |
R004 | R004 | Exact | $3.00 |
R005 | ROO5 | O / zero mismatch | $124.00 |
R006 | ROO6 | O / zero mismatch | $39.90 |
All six displayed amounts match the expected CSV. The problem is record identity. This is why both field-level and ID-integrity checks belong in a conversion workflow.
Scope reminder: these are three clean synthetic fixtures. The OCR test used known cell positions. No paid converter was tested, and no result here is a general accuracy or vendor-ranking claim.
Expected CSV · Observed OCR CSV · Detailed cell errors · Full methodology