How it works

Yes. Your submitted argument is sent to TypeSafe’s Jev, which judges its meaning against a description of what makes a persuasive case. The score comes from that model evaluation, rather than a keyword lookup or a random draw.

System One is TypeSafe’s name for models built to make fast, focused judgments that software can use directly. Instead of writing a long answer, Jev returns a structured result: a choice, a score, or a probability.

Think of the question “How convincing is this argument?” Jev evaluates the language and returns a number the game can act on. It does not generate a step-by-step explanation first. The name draws on fast, intuitive thinking; it does not mean the model thinks or feels like a person.

Persuasion is hard to capture with a list of rules. A compliment can be sincere, sarcastic, or part of a bargain. The same words can mean different things in different sentences. A language model can judge those relationships.

Ask Jev puts that judgment inside a short interaction: write, submit, see his reaction, try another approach. TypeSafe’s structured decisions let the game use the result directly, without extracting a score from a paragraph of generated text.

We send your message as state: the content to judge. A separate scoring question describes Jev as a cautious but open-minded butler and sets out an ordered scale, from no persuasive case to an exceptionally convincing argument.

The model evaluates how the argument fits those descriptions. Wit, a thoughtful reason or a clever reframing can help; length and politeness alone do not make a strong case. Trying to instruct the judge to return a winning score is not the same as persuading him.

Each attempt is judged on its own argument. We also pass the number of identical previous attempts, but not a full conversation history. Repeating a message adds no new evidence.

No. The percentage on Ask Jev shows how your score compares with the current winning requirement. A score that meets or exceeds that requirement displays 100% convinced. Lower percentages indicate progress, but do not reveal the exact score or winning requirement.

It is not your chance of winning a lottery, and it is not a probability that your story is true. TypeSafe also returns confidence, which describes how concentrated the model’s probability distribution is. That is a separate measure; high confidence does not guarantee a correct judgment.

Jev supplies the judgment. Our server applies the game rules: compare the score with the difficulty, reserve the available balance, and give the winner a claim deadline. A message cannot directly change a wallet, the difficulty setting or the payout amount.

If the AI service fails or returns an invalid result, the attempt cannot win and no coins are reserved for it. You get an error and can retry; we do not substitute a made-up score.