How bank statement balance-checking works — and what it cannot prove

When software hands you data extracted from a document, the honest question is: how do you know it's right? For most extraction tools the answer is "it usually is" — confidence expressed as a percentage, measured on someone else's documents. We built NoRekey around a stronger answer: where a statement prints usable balances, we can compare the extracted movement against them and flag disagreement.

The statement is a self-verifying document

A bank statement declares an opening balance, a closing balance, and the transactions between them. That structure is a gift: it gives the extraction an independent arithmetic check. Sum the extracted amounts, walk the opening balance forward, and you either land on the declared closing balance to the penny or you don't.

opening + sum(amounts) == closing   (within 0.01)

NoRekey checks extracted amounts against available statement balances. A single missing nonzero row or sign error will normally change the total. But matching totals do not prove every row is correct: two omitted transactions of equal and opposite amounts can cancel out. Dates and descriptions can also be wrong without changing the sum. Keep the source and review suspicious rows even when the result is balanced.

Running balances tighten the net further

Most statements also print a running balance beside transactions. That per-row trail lets verification act locally, not just end-to-end: the chain must agree at every printed balance, which helps locate the point at which the amounts stop agreeing. It also enables something better than detection — evidence-based repair. If a running balance disagrees with the amount chain but the magnitudes match exactly, a sign discrepancy may be identifiable from the surrounding evidence; overdrawn balances marked "OD" are the classic case. We only ever correct what the document demonstrates, never what a heuristic guesses.

What you see

Balanced means the checked arithmetic agrees; it is not a certification of tax treatment or every extracted field. Needs review flags disagreement so you can compare the rows with the original. When a statement omits usable opening or closing balances, a complete check may not be possible. The help-centre explanation covers those cases.

Why this is the feature that matters

A balance comparison is a test you can reproduce on your own document. It provides more useful evidence than an unexplained confidence percentage, while still leaving dates, descriptions, account coverage and bookkeeping classifications for review.

Try it on the ugliest statement you have — the dense one, the scanned one — at norekey.com. The balance check is exactly the kind of promise you should test before trusting.