Receipt #12

3 / 4 rules pass Model: needs reviewmatches human label
CORD-v2 test receipt · read by Qwen3.5-4B (QLoRA) with best-of-4 and the accounting checker

Receipt photo

Photo of receipt #12 from the CORD-v2 test set

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.

  • Greedy75% rulespicked
  • Sample 175% rules
  • Sample 275% rules
  • Sample 375% rules
  • Sample 475% rules

Extracted fields

ItemQtyUnit priceAmountRemove

Accounting checks

goes to human review
  • Line 1: qty × unit price = amountexpected 120,000 · found 120,000
  • Line 2: qty × unit price = amountexpected 880 · found 0
  • Items add up to the totalexpected 120,000 · found 120,000
  • Card charge = totalexpected 120,000 · found 120,000

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.