Identity verification accuracy, a comparison

Identity verification accuracy is not one number, it is a trade-off between false acceptances and false rejections, measured on a specific population against a specific threat, which is why a method's structure predicts real-world accuracy better than any advertised percentage.

Accuracy is the most quoted and least examined claim in identity verification. A headline figure feels decisive, but it hides three questions that decide whether the number means anything: accurate at what, measured how, and against whom. This comparison walks through each, then ranks the methods on the accuracy that survives an actual attack. For a companion view that scores the same methods on binding, spoof resistance, and auditability, see accurate identity verification methods, compared.

Accuracy Is Two Numbers, Not One

Every verification decision can fail in two directions, and they trade against each other:

• False acceptance rate (FAR). How often an impostor is approved. This is the number fraud cares about.
• False rejection rate (FRR). How often a legitimate person is denied. This is the number your candidates and your funnel care about.

Turn one down and the other goes up. A vendor can advertise a stunning FAR by loosening thresholds until real users get rejected, or a smooth pass-rate by quietly letting impostors through. A single "accuracy" percentage usually blends these into a figure that tells you neither, so always ask which error the number is counting.

Measured How, and Against Whom

The second problem is the test set. Accuracy measured on cooperative users in good lighting says nothing about accuracy under attack. The threat model is everything: a method that scores 99.9% against honest mistakes can collapse against deepfakes, injected video, template forgeries, and paid proxies, the adversaries that actually target hiring. As the rise of deepfake hiring fraud shows, the tooling aimed at digital checks is improving faster than the checks themselves, so an accuracy figure from last year describes a threat that no longer exists.

How the Methods Compare on Real Accuracy

Judged on adversarial accuracy, the false acceptance rate that holds when someone is actively trying to beat the check, the methods separate cleanly:

• Knowledge-based authentication. Poor. The "secret" answers are breached and for sale, so a prepared impostor often scores higher than the real person.
• Document plus selfie and liveness. Good against casual fraud, eroding against deepfakes and injection, because the whole decision runs on images the attacker supplies.
• Live-agent video review. Slightly better on clumsy fakes, but a remote human sees only the feed that real-time face-swaps are built to fool.
• Biometrics. Excellent at re-recognizing an enrolled user, but its accuracy is inherited from whatever verified the very first enrollment. Fool enrollment once and it re-recognizes the fraud perfectly.
• In-person notarization. Highest adversarial accuracy, because a deepfake or a VPN cannot put a body in a chair, and a proxy cannot present a stolen ID without being the person examined.

The Honest Ranking

On cooperative users, most of these methods post similar numbers. On adversarial users, the population that matters when the cost of a wrong hire is high, the ranking inverts toward whichever method requires a physically present human. That is not a marketing claim; it is a structural fact. A check that only ever inspects data or pixels can, in principle, be satisfied by manufactured data or pixels. A check that requires a real person in a real room cannot.

This does not mean every decision needs a notary. For low-stakes, high-volume verification, a fast digital method at a sensible FRR is the right trade. But when accuracy has to hold against a motivated adversary, the method with the lowest real-world false acceptance rate is the one that puts a trained official in front of the person. For the full menu of options and how they behave on presence, see presence verification methods compared.

Where PinpointVerify Lands

PinpointVerify runs the method with the lowest adversarial false acceptance rate, and in practice a false acceptance is nearly impossible. The check requires the candidate to appear in person, at a specific location, in front of a state-licensed notary who examines their government-issued photo ID face-to-face. That single requirement closes the two doors a false acceptance would have to pass through.

• Location is self-enforcing. A candidate who is not actually where they claim to be cannot show up for an in-person appointment near that location, so a spoofed location eliminates itself.
• Identity is self-enforcing. A candidate applying under a name that is not theirs cannot produce a matching government-issued ID for the notary to inspect, so a fabricated identity has nothing to present.

What remains are logistical errors, a missed appointment or a scheduling mix-up, not the identity gaps that inflate the false acceptance rates of digital-only checks. The output is a notarized record plus a location-confirmed trace. See how in-person verification works.