@sardine says Jev caught 27 of 29 fraudulent transactions in a September fraud ring, the most of any approach they tested, and that Jev is now a node in Sardine Flow.

Benchmark post. Sardine says they ran Jev, TypeSafe AI's System One model, on a fraud ring that hit a customer in September. Jev caught 27 of 29 fraudulent transactions, the most of any approach tested. Their Data Analyst Agent's rules, built from the ring's behavior such as four cards per account, would have caught 26 of 29. Jev needed one broad question about each transaction's fraud-ring risk. Every method used the same account links and velocity counts, which their code computed first. They say Jev turned that evidence into a decision faster, cheaper, and more consistently, answering in about 100 milliseconds for a fraction of a cent. On consistency it beat Claude Sonnet 5: for one account's 12 transactions in about 100 minutes, Jev scored all 12 high risk, while Sonnet mixed medium and high. They also added Jev as a node in Sardine Flow. The post links a Sardine blog and a November 4 fireside registration, and attaches one photo.