Document fraud detection you can actually buy

Start here: buy documents, not a subscription

A one-off payment for a bundle of documents that never expires, no card on file, nothing to cancel, no renewal date to diary. Most people screening documents do it in bursts, and this is the shape that fits a burst.

Your 50 free documents a month are spent first, every month, before a single credit is touched. Credits are the alternative to subscribing, so they are bought from a Free account, if a plan already covers your volume, you do not need them.

Or subscribe, when the volume is steady

Every plan runs the identical engine: you pay for volume only.

• The API is on every plan, including Free. There is no tier here where you may check a document by hand but not from your own code. A free key calls the same POST /api/v1/inspect the paid ones do, against the same engine and the same signal families.
• No monthly minimum. Nothing has to be committed before the first document is checked: the web checker is free and needs no account, the Free plan needs no card, and a credit pack is a one-off purchase with no floor under it. The cheapest way to check one document is to check one document.

Brazilian cards are charged in reais at checkout. These are the amounts you will be charged, not a conversion.

Which plan fits

Past 50,000 a month, email us, we would rather talk about your volume than guess at a price for it.

What a check costs, against what a case costs

At the credit-pack rate a checked document runs four to five cents. Here is the other side of that, using only numbers with a published method. The sources are named because a number you cannot check is worth nothing on a page like this.

• If you review documents inside a business. In the ACFE's 2026 Report to the Nations (2,402 investigated cases), the median occupational fraud loss was USD 104,000, and organisations under 100 employees carried the highest median of any size band at USD 126,000. In the same study the most common way fraud was concealed was creating or altering documents, and document examination was the first thing to catch it in 5% of cases.
• If you already have identity verification. DHS's Remote Identity Validation Rally tested seven commercial document-validation systems: false-accept rates ran from under 0.9% to under 76.7%. Two of the seven accepted roughly seven in ten forged documents. A neutral federal evaluator, not a vendor, and a reason "we're already covered" deserves a test rather than an assumption. Those systems read photographed ID cards, not PDF structure; the point is that the category is not solved, not that our numbers are theirs.
• If you operate in Brazil. The Fórum Brasileiro de Segurança Pública recorded 2,261,055 estelionato reports in 2025, and 2.8% of them produced a criminal proceeding. Where recovery through the courts is that unlikely, the control that remains is the one you apply before the transaction.

What none of this says is that a check prevents a loss. Tamperlens reports what a file's structure records, with the innocent explanation beside every finding; the decision, and the outcome, stay yours. The arithmetic worth doing is simply that the check is cheap and the case is not.

Private issuer baselines: the answer to a document that was never edited

Everything else on this site reads the record of editing: revisions appended after creation, a value retyped in a second font subset, an editor's name in the metadata. That works, and we publish how well. It has one blind spot, and we would rather sell you the counter than pretend the spot is not there.

A document generated cleanly from scratch carries no editing traces, because nothing was edited. Printed from a template, rebuilt as a flat image, emitted by a generator: the bytes are honest about their own history and silent about whose document it claims to be. That is the growth class in document fraud, and no byte-level engine catches it. Ours included.

What does catch it is a reference. Statement pipelines are remarkably consistent: PDF version, xref style, page geometry, and often the producer field they leave empty. A baseline is a structural profile of an issuer's genuine output, trained from documents you already hold and trust. A file claiming to come from that issuer is then measured against how that issuer actually writes files, and a clean generator's output does not resemble a bank's pipeline by accident.

• Only fingerprints are stored. Training documents are read in memory and discarded like every other upload. See how that works.
• They are yours. A private baseline is visible to your key alone. The built-in public baselines work on every plan, including Free.
• Still signals, not verdicts. A mismatch says this file does not look like that issuer's output, with the differences named. A legitimate pipeline change looks the same, and the report says so.

Private baselines are on Team and up, and the API reference documents training and use. Whether they hold up against generated documents is being measured next, and the number will be published either way.