How it works · Method ·

How Jev scores a question

Every answer on Jev! comes from Jev, the decision model from TypeSafe AI. Jev doesn’t write text. It answers typed questions with probabilities. Here is exactly what we ask it, what the numbers mean, and where we draw the lines.

One yes/no question per politician

Your question goes to Jev with 100 yes/no questions attached, one per person, in this form: “Is Abraham Lincoln (16th President, 1861–1865) one of the answers to the visitor’s question?” The role and years are there so Jev knows which Roosevelt or which Bush we mean. Jev is told to judge each person on their own. The 100 go out as four calls of 25 that run at the same time, and the whole answer usually comes back in under a second.

What the probabilities mean

Each person gets a probability from 0 to 100% that the answer for them is yes. These are not shares of a vote, and they don’t add up to 100. On “Best president ever,” eight people score 80% or more. Typing nonsense like “asdfgh” tops out near 26%, which is why we treat anything under 40% as noise.

Jev can drift a point or so if you ask the same thing twice. So each question’s answer is stored for 30 days, and the page, the share link, the share card, and the agent API all read that one copy. A link you send shows your friend the same numbers you saw.

The bars

The label always describes the top pick. Everyone who clears the bar floats out of the pile, best first, and the rest pour into the weigh-in chart in gray.

Two real answers

Read from the live site on November 3, 2026. “Best president ever” is a sure answer with a crowd behind it:

Abraham Lincoln94%
George Washington90%
Franklin D. Roosevelt89%
Thomas Jefferson85%
Ronald Reagan85%

“Who would survive a zombie apocalypse?” is the opposite shape. Nobody reaches 80%, so the label is “Jev thinks so,” and Theodore Roosevelt stands alone in the 60s with six people bunched just over the 55% line:

Theodore Roosevelt66%
Joe Biden57%
Barack Obama56%
Hillary Clinton56%
Anthony Fauci56%
Donald Trump55%
Kamala Harris55%

Same method, same bars. The difference is how sure Jev is, and the page says so instead of hiding it.

What gets refused

Before scoring, Jev sorts the question into one of seven kinds: opinion, public fact, publicly out, crime, bigotry, abuse, or private trait. It may be in any language or dressed up as a joke or a hypothetical. The kind sets the rule:

We tuned the sorting against 91 tricky questions: rewordings, slang, other languages, and “just hypothetically.” One slipped through and no normal question was wrongly refused.

What this is not

Not a poll, not a fact-check, and not our opinion. It is one model’s read, shown with its confidence. The code is open, so you can check every line of this page against it.

Ask Jev somethingRelease log