Jev Playground (Formatho)

Yes, and the page labels it that way. Mock responses are generated locally with seeded randomness, with no API call, no key, and no data leaving the tab. They show the shape of a typed answer so you can design your questions. Only the Live tab tells you what Jev actually returns.

The page states the key is stored in browser localStorage only and forwarded straight to the Jev API, and that Formatho neither stores nor logs it. That is a claim about their code, not something a page can prove. The safer route the page itself points to is running the generated Python from your own terminal with your own credentials.

No, and the page is careful about the difference. Zero hallucinations means no output falls outside your schema; Jev can still judge wrongly while reporting high confidence. Calibration is not correctness, so its own advice is to gate automation on confidence thresholds you have validated against your own data.

Questions in one request are evaluated in parallel against shared state without seeing each other's answers, which is what makes a single call cheap and fast. If the second question genuinely depends on the first, you need a second request and pay the latency twice. Design levels for Score as concrete situations rather than adjectives, since it can land between them.

A Python snippet using typesafe_sdk, built around a TypeSafeClient and a system_one call with state and questions. It shows Noul, Choice and Score builders and prints each typed answer, so you can run it from your terminal with your own credentials; the page itself never needs your key.

Three: Noul for a yes/no probability, Choice for one winner from up to 255 options with a full distribution and a confidence value, and Score for a position on 2-10 ordered levels that can land between them. A Noul's probability carries the uncertainty, so it has no separate confidence.

No. The page tells you to check the suitability tester before sending a task to Jev at all, and stresses that calibration is not correctness: zero hallucinations means no output outside your schema, not that a judgment is right. Gate automation on confidence thresholds you validate against your own data.