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.
- Greedy80% rules
- Sample 1100% rulespicked
- Sample 2100% rules
- Sample 380% rules
- Sample 480% rules
Extracted fields
| Item | Qty | Unit price | Amount | Remove |
|---|---|---|---|---|
Accounting checks
would be accepted
- Line 1: qty × unit price = amountexpected 1,944,000 · found 1,944,000
- Line 2: qty × unit price = amountexpected 975,000 · found 975,000
- Line 3: qty × unit price = amountexpected 972,000 · found 972,000
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.