CORD-v2 test receipt · read by Qwen3.5-4B (QLoRA) with best-of-4 and the accounting checker
Receipt photo

How the checker chose
The model wrote one normal reading and four sampled ones. The checker keeps the reading that passes the most accounting rules; the model's confidence only breaks ties.
- Greedy67% rulespicked
- Sample 167% rules
- Sample 267% rules
- Sample 367% rules
- Sample 467% rules
Extracted fields
| Item | Qty | Unit price | Amount | Remove |
|---|---|---|---|---|
Accounting checks
goes to human review
- Items add up to the subtotalexpected 238,000 · found 238,000
- Subtotal − discount + tax + service = totalexpected 281,435 · found 281,435
- Cash − change = totalexpected 181,435 · found 281,435
The rules recheck as you edit, using a TypeScript port of the Python checker that agrees with it on all 277 rule results of the test set. Amounts within 1% are treated as equal, to allow for rounding.