jevbooks

Jev in practicewhat to ask, what to keep in code

Stop asking an LLM
for a yes/no.

Send Jev a state and typed questions. It returns decisions and probabilities instead of prose. Your code decides what happens next.

4 Nov 2026 ▪ Which judgments can Jev take?

Enter ↵ · Shift+Enter newline

Every number on this page is Jev’s own; the plan text is from a language model and labelled as such.

01The shape of one request

A stack of cards (the state and three typed questions) goes into a pink square machine (Jev); three outlined bars come out (the three numbers); a stamp (your code) makes the call
state · questionsJev · ~300 msyour code decides
state {task, observation}Q1 noul done?Q2 choice which signal?Q3 score how far?Jev · ~300 ms0.98blocked 0.881.97 / 4if (p > 0.7) stop() ← your code
state {task, observation}Q1 noul done?Q2 choice which signal?Q3 score how far?Jev · ~300 ms0.98blocked 0.881.97 / 4if (p > 0.7) stop() ← your code
One request: a state and three typed questions in, three numbers out, and your code decides.

02One live question

Change the sentence. Watch the number.

Jev mascot: the pink square is Jev, holding up a checked card; the three dials are the three numbers it returns
▪ task_done · noul · live

One live Jev question. The task is fixed; edit the observation and Jev re-answers task_done — a noul, P(yes) — in ~300 ms.

taskBook a table for two at 7pm tonight.

Jev reads literally — sarcasm and hollow "done!" are your code’s job: add a second question or demand details in the state

task_done—

The question we send
{
  "state": {
    "task": "Book a table for two at 7pm tonight.",
    "observation": "Reservation confirmed — 2 guests, 7:00 PM, Table 12."
  },
  "questions": {
    "task_done": {
      "type": "noul",
      "instructions": {
        "what": "Does `observation` confirm that the goal in `task` was achieved?",
        "not_for": "Partial progress, a different time or party size, or a promise to book later."
      },
      "criteria": {
        "true": "The observation confirms the task exactly as stated.",
        "false": "It does not confirm the task as stated."
      }
    }
  }
}

0 = no · 1 = yes · 0.5 is “can’t tell”, not “medium”

Run it yourself: key at console.typesafe.ai/keys · $0.042 per 1M input tokens, output free · this question: … tokens ≈ $… · plain fetch, no SDK needed

No reliability curve published yet — every number on this page is live, not cherry-picked

03Three ways to ask

noul

Does observation confirm the goal in task?

“Reservation confirmed — 2 guests, 7:00 PM, Table 12.”

00.5 can’t tell10.98 = yes
00.5 can’t tell10.98 = yes
One question, one number: 0.98 = yes; 0.5 is “can’t tell”, not “medium”.

choice

“7:00 PM shows unavailable; 7:30 PM selectable”

goal_met0.00subgoal_met0.12blocked0.88other0.00
goal_met0.00subgoal_met0.12blocked0.88other0.00
Jev picks blocked (0.88) and is 0.84 sure of that pick; other is the exit, so a wrong list never forces a wrong answer.

score

0.0600.1410.6020.1730.0341.97 / 4
0.0600.1410.6020.1730.0341.97 / 4
  1. 0 nothing done
  2. 1 started
  3. 2 one subgoal confirmed
  4. 3 all but final
  5. 4 fully confirmed
Levels are situations, not adjectives; the score is the probability-weighted mean.

This playground demonstrates text decisions. Jev also offers experimental image input and Box localization. Read the image and Box guide →