Rules decide, a model explains

Yfraud tells you Stop or Clear before you pay a scammer. The verdict always comes from rules. A language model is allowed to explain the verdict in plain words, and it is never allowed to change it.

Who it is for. Most scams are not clever. They are a link that looks close enough, sent when you are rushed. The people who lose money usually know the warning signs; they just don't get a second look before they press Send. Many of them have been cheated before. "The computer thinks so" is not going to win them back.

Why rules decide. A rule can be read. When Yfraud says Stop, you can trace the verdict to the exact check that fired: the real domain hiding behind a familiar name, a login page on a site that isn't the bank, a deadline written to rush you. A bank can show that line to a regulator. A model's guess can't be shown to anyone.

What the model is for. People don't read rule names. The model turns "lookalike domain" into "the real site is verify.io; paypal.com is only the front of the address, there to fool you." That is the whole job. We test every release to make sure the explanation can never turn a Stop into a Clear.

Why it runs on your device. The rules need no network. What you paste or open is checked on your own machine and never sent anywhere. For people who have already been burned, that matters as much as the verdict.

How it does. In our latest tests on real scam emails, Yfraud catches 80% of phishing with rules alone, up from 42% a week earlier. It wrongly warned on 3 real websites out of more than 134,000. The first thing that round of testing found was our own mistake: Yfraud was warning on one in five ordinary mailing list emails. We fixed that before adding a single new rule.

The cost. A new kind of scam needs a new rule before Yfraud catches it. A model alone might catch it sooner, by luck. We choose explainable over fast, because the first time Yfraud is wrong without a reason, nobody trusts it again.

One colour. Yfraud is ink and paper. Red shows up only in the Stop state, so the one time you see it, it means something.