Report 1: Baseline
The tool as it stands, measured on two sets of invoices, with the issues found. Later reports score the same invoices with the same commands, so they can be compared with this one.
| Measure (rules extractor) | Corpus A: independent invoices | Corpus B: built-in generator |
|---|---|---|
| Field accuracy | 84.5% | 99.9% |
| Planted errors found | 25 of 32 | 20 of 20 |
| False alarms | 30 | 0 |
| Rule engine on perfect input (oracle) | 32 of 32, 0 false alarms | 20 of 20, 0 false alarms |
Raw results