Choice
categoryWhat kind of ticket is `ticket`?
bug_report
84% confidence
Option probabilities
bug_report89%billing8%other2%feature_request1%A WORKING LAB FOR DEVELOPERS & AGENT BUILDERS
Give your code
better judgment.
TypeSafe’s Jev turns messy input into typed judgments.
Your code keeps the thresholds, routing, and final say.
No signup. No visitor API key. Start with a saved example.
01 / THE WORKING STUDIO
See the if. Not the chat.
Change the input. Inspect the judgment.
Move the boundary your code controls.
1
You provide the state
Threshold lessonRoutes support tickets by category and urgency.
Start here: raise the confidence floor in step 3 to 0.90. The same saved judgment switches to a human — no API call.
Start with a variant
Try “Calm billing” to compare it with this angry outage. Then run Jev.
Live runs send this input and context to TypeSafe. Use sample data, not secrets.
Saved example loaded. Edit the input or choose another variant to make a live request.
2
Jev returns judgments
SAVED DEMOTyped values, not generated prose. Your code uses these.
View the input behind these demo answers
My production API has been down for 2 hours and I'm losing money! This is completely unacceptable. I need someone to fix this NOW or I want a full refund.Saved illustrative answers. No API request or live timing.
Noul
urgencyDoes `ticket` convey urgency or time-sensitivity?
0.94probability of yes
0 · no1 · yes
Score
frustrationHow frustrated is the author of `ticket`?
1.81 / 2weighted score
Between rubric levels 1 and 2
See the rubric
- Calm, just stating facts
- Frustrated but civil
- Very angry, strong language, or threatening to leave
Noul
refund_requestedDoes `ticket` ask for money back?
0.76probability of yes
0 · no1 · yes
Saved judgments, real policy. Edit the input to request fresh answers.
Choice Which option?Noul Probability of yesScore Position on a rubric
3
Your code decides
Same judgments. Your threshold. Your outcome.
POLICY OUTCOME · DEMO INPUTengineering-priority
0.50Higher = more cautious0.95
category.confidence0.84 ≥ 0.70Answer confidence vs. your floor. Other rules still apply.
Try 0.90. The same 0.84 confidence now sends this ticket to a human. No new API call.
if (answers.category.confidence < 0.70) {
return { route: "human" };
}
if (answers.category.choice === "other") {
return { route: "human" };
}
if (answers.category.choice === "bug_report" && answers.urgency.noul > 0.8) {
return { route: "engineering-priority" };
}
if (answers.category.choice === "billing" && answers.refund_requested.noul > 0.7) {
return { route: "billing-refund" };
}
return { route: `queue:${answers.category.choice}` };Jev returned bug_report with 84% confidence and 0.94 urgency. Your policy selects engineering-priority.
Illustrative policy, evaluated locally. No tickets routed, commands executed, content published, or payments made.
Jev judges. Code decides. You can inspect both.Learn to design your own
02 / FROM THE DEMO TO YOUR CODEBASE
You’ve seen the branch.
Now design the system.
The slider is the easy part. Choosing the right questions, setting a defensible floor, and knowing when to ask a human — that’s the work.
The ebook picks up exactly where this Studio stops.
IN THE STUDIO
Try a prepared decision.
IN THE EBOOK
Learn why the policy works.
IN YOUR CODEBASE
Test a gate on your own data.
A practical field guide. Not another prompt collection.
THE COMPANION EBOOK · BY SKP / TENX LABS
Hands-On Jev
Build the judgment layer.
Keep control of everything after it.
For developers, indie hackers, and agent builders who want to classify, route, score, and gate — without turning every branch into a chat completion.
- Design better questions. API shapes, typed primitives, and concrete rubrics.
- Own the policy. Fan-out, confidence gates, composite scores, and intent routing.
- Know what can fail. Jaggedness, overrides, shadow testing, and version pinning.
- Practice, then adapt. Eight exercises, worked case studies, and a record-your-own live lab.
CHOOSE YOUR EDITION
Two editions. One book.
FIELD GUIDE + PRO KIT · RECOMMENDED
$49USD
The book + the kit for $20 more.
The book teaches the method. The kit puts it in your codebase — and in your coding agent.
38 files · 4 working gates · 67 offline tests · TypeScript + Python
- Everything in the Field Guide, plus:
- Four working gates — support triage, publish gate, tool review (TypeScript) and the offer picker (Python), with Python versions of every policy and runner
- An agent skill for gate #5 — a
SKILL.mdthat walks your coding agent through the book’s 7-step gate design (cost of error, state, primitive, criteria, policy, validation, tests) and has it extend the kit’s patterns instead of inventing its own - 67 offline tests, no API key —
npm run verifyruns both suites, including 29 named boundary cases - Fixtures to live calls, and back —
--livecalls Jev;--recordsaves real responses as new test fixtures - Ship it — use the kit’s code in your own commercial projects, unlimited projects
- Eval harness + 5 setup guides — score the model against labeled examples and adapt each gate to your data
- Every book listing as copy-paste Markdown
30-day money-back guarantee · Instant download · Free v1.x updates
THE FIELD GUIDE · JUST THE BOOK
$29USD
- PDF and EPUB editions
- Two-page desk reference (also free)
- 11 chapters, 8 exercises + answer key
- Decision-design worksheet, 6 diagrams, docs map
30-day money-back guarantee · Instant download · Free v1.x updates
| Feature | Field Guide | + Pro Kit |
|---|---|---|
| Learn the method (book, exercises, diagrams) | ✓ | ✓ |
| Copy-paste listings | From the PDF | ✓ Markdown file |
| Working, tested gate code (TS + Python) | — | ✓ |
| Agent skill for designing new gates | — | ✓ |
| Runs offline without an API key | Book examples | ✓ Full test suite |
| Commercial use of the code | Book snippets | ✓ Whole kit |
| Files in your download | 14 | 53 |
- Delivery. Instant download via Gumroad.
- Prices. US dollars; any applicable tax is shown at checkout.
- API access. Not included — the book and kit run offline; live integrations use your own TypeSafe access.
- Updates. Free corrections and minor additions (v1.x); a major revision ships as a new edition.
- Terms. License, refund, and support terms are listed on the Gumroad page.
Published by TenX Labs. Written by SKP — founder, AI engineer, and product manager, previously at Microsoft. An independent guide to TypeSafe Jev.
03 / LOOK INSIDE, BEFORE YOU BUY
Less mystery. More working knowledge.
FROM CHAPTER 5.2Confidence gating
“A judgment describes the input; it is evidence, not authority. Confidence helps your code decide whether that evidence is usable or needs review. Permissions, hard rules, and the cost of being wrong still decide what happens.”
“Confidence informs policy. It does not grant permission.”
“Low confidence is not failure; it is a first-class outcome: review, escalate, or ask a different question.”
Hands-On Jev · SKP
A short guide. The whole decision path.
From your first typed question to a policy you can test and refine.
01—05
The mental model, primitives, first request, and policy boundaries
06—08
Worked patterns and offer selection
09—11
Composition, testing, and the live lab
See all 11 chapters
- Your first smart if statement
- Three primitives, three different questions
- Design the state before the question
- Make one real request
- Put uncertainty and overrides in code
- Worked pattern: an editorial publish gate
- Worked pattern: a tool-review boundary
- Select an offer without inventing dependencies
- Advanced: compose judgments without losing meaning
- Test the policy; evaluate the model separately
- Run a small live lab, then decide what to ship
KEEP THE MENTAL MODEL
Not buying yet? Take the desk reference.
Three primitives. Four patterns. A before-production checklist.
The same two-page reference that ships with the ebook — free, no email required.
A FEW STRAIGHT ANSWERS
Know what
you’re getting.
Is this demo actually calling Jev?
Default examples use saved, illustrative answers and make no API call. Edit the text or select a non-default variant, then Run, to request live judgments through the Studio’s server. Live results are labeled. If a request fails, we show the error — never substitute saved answers for your edits. Moving the threshold only reruns local policy.
Do I need an API key or an account?
Not to use the Studio. Live runs use the host’s server-side TypeSafe key when configured. To integrate Jev into your own application, you’ll need your own TypeSafe access. The ebook does not include API credits or guarantee access to TypeSafe’s early-access service.
What does the Pro Kit add?
The $29 Field Guide teaches every pattern through the book’s listings, exercises, and worked case studies. The $49 Field Guide + Pro Kit adds the runnable developer kit — 38 files: four working gates with Python and TypeScript versions, 67 offline tests, an agent skill for designing new gates, an eval harness, setup guides, and every listing as copy-paste Markdown — and you can ship the kit’s code in your own commercial projects. The kit is only in the $49 edition.
What’s the agent skill?
A SKILL.md file your coding agent follows to design a new gate with the book’s method: decide the cost of being wrong, design the state, pick the primitive, write criteria, put the policy in code, validate responses, and test on fixtures. It tells the agent to extend the kit’s own questions, policies, and tests rather than invent new ones. If your agent doesn’t load skill files, paste it into its instructions.
Is this a production-ready safety system?
No. These are teaching examples with illustrative thresholds. Jev’s judgments are probabilistic. Validate against your own labeled data, keep deterministic checks and side effects in code, and add human review where the stakes demand it. The Studio never executes the actions it displays.
Who made this?
TenX Labs publishes Hands-On Jev Studio and Hands-On Jev. SKP is the author. This is an independent educational product about TypeSafe’s Jev, not an official TypeSafe product. Technical references are linked in the ebook.
KEEP THE IF. IMPROVE THE JUDGMENT.