Jev gives your code decisions, not more text. JevStation is where you run them.
Send a ticket, review, tool call or model answer, ask Choice, Score or Noul questions, and get a typed answer with probabilities and a confidence value — ready to threshold, nothing to parse. JevStation runs Jev, TypeSafe's System One model, and every decision model that speaks the same contract on one key: Cloudflare Clef, Perplexity Decider and OpenAI Luna Decisions.
- Question types: Choice, Score, Noul
- 3Question types: Choice, Score, Noul
- Typed questions per evaluation
- 12Typed questions per evaluation
- Per evaluation: 1 credit standard, 3 above 8,000 characters or 5 questions
- 1–3 creditsPer evaluation: 1 credit standard, 3 above 8,000 characters or 5 questions
Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing with 'invalid redirect'. I'm losing sales every hour this is down. Please help ASAP.
Answers
- Team
- integrations99%
- Urgency
- 3.7/4 · Critical77%
- Churn risk
- Yes80%
Real Jev output · 0.89 s · captured 2026-10-09
Private by default. Your threads are never used for training.
What's inside
Everything around Jev, handled
Jev returns the judgement; JevStation handles everything that decides whether an evaluation survives contact with real work. A standard run costs 1 credit — state up to 8,000 characters and up to 5 questions — and a larger one costs 3. A failed run costs 0.
Typed answers, not prose
Choice, Score and Noul answers arrive as values your code can branch on, each with the full probability distribution behind it.
Read the contractSaved threads
Every evaluation is stored against your account with an auto-derived title, so the state you built is still there weeks later.
Open the playgroundOne round trip
Jev does not stream tokens, so there is nothing to reassemble. A run either returns a complete answer set or fails cleanly.
See a requestCredit metering
A standard evaluation costs 1 credit — state up to 8,000 characters and up to 5 questions. A larger run costs 3 credits. Credits are deducted when the answers land, and a provider error refunds them automatically.
Check your balanceReusable state
Pin standing instructions or reference material to a thread, so the context you already gathered becomes the state it evaluates.
Read the docsOpen submissions
Publish a question set, tool or write-up you built on Jev and it lands in the shared catalog for everyone else.
Submit a playbookHow it works
State in, structured answers out
Three moves, in the order you actually meet them: paste something to judge, write up to 12 typed questions, then read the answers. Each question comes back as a typed value with its probability distribution, so the branch your code takes is decided by the response rather than by parsing it.
- 01
Paste the state
Open the playground and paste whatever Jev should judge: a support ticket, a review, a spec or a JSON record.
- 02
Write typed questions
Add Choice, Score or Noul questions, up to twelve per evaluation, each with the criteria that define its options.
- 03
Read the answers
Jev returns the typed value, its probability distribution and a confidence score. Keep the thread, or publish the question set as a playbook.
Positioning
JevStation and a text-generating LLM
A language model writes prose that you then have to parse. Jev returns the judgement itself, in the shape your code already expects.
What it is
JevStation
A JevStation workspace over a System One model
Text-generating LLM
A model that predicts the next token
Output shape
JevStation
A typed value plus the probability distribution behind it
Text-generating LLM
Prose you coerce into JSON and parse back
Confidence
JevStation
A calibrated confidence value on every Choice and Score answer
Text-generating LLM
None, unless you ask for it in the prompt
Cost control
JevStation
1 credit per standard evaluation, 3 for a large one, refunded automatically on failure
Text-generating LLM
You count tokens, retries and malformed output
Use cases
What people ask Jev
Jev is deliberately narrow: it judges, it does not write. A run takes up to 12 typed questions and returns a typed value, the full probability distribution and a confidence score for each one. These are the shapes that come up most often in the shared catalog.
Support triage
Paste a batch of tickets as the state, then ask a Noul question for urgency and a Score question for complexity.
Review moderation
Ask Choice whether a review is positive, neutral or negative, then route on the answer's probability instead of a keyword list.
Rubric scoring
Turn a long document into the state and score each section against the same rubric, so the results stay comparable.
Team playbooks
Publish the question set that finally worked, so the next person starts from your criteria instead of a blank prompt.
Learn Jev
Start with how System One models work.
Short, sourced explainers on what Jev is, what an evaluation actually costs, and how teams use it inside agent loops.
What Is a System One Model? Jev, Explained
A System One model like Jev returns typed decisions with calibrated probabilities instead of text. How it works, how RLCD differs from RLHF, and its limits.
Read the guideJev Use Cases: 12 System One Model Decisions
What can Jev do? Twelve decisions it makes in an agent or pipeline, from ticket triage to tool-call guards, each with its question set and a free tool page.
Read the guideJev as a Judge: What One Yes/No Question Catches
Can a text-free model judge LLM output? One Jev question ranks backdoored code at AUROC 0.976 for ~$0.04 per 1,000 checks. Where it holds, where it breaks.
Read the guideWhat One Jev Evaluation Costs, Measured
We made 11 real Jev calls across every state size and question count, fitted the token model and measured the cost: $0.000012 to $0.000377 per evaluation.
Read the guideRun your first Jev evaluation.
Create an account and you get 200 free credits — enough for 200 standard evaluations. No card, no provider API key and no setup: open the playground and run your first evaluation in the browser.