Example library · 110 scenarios
Browse real Jev-shaped decisions.
Search practical workflows, games, dilemmas, and model challenges. Each scenario exposes the state and typed questions, then lets you load a question directly into JevKit's live Playground.
Showing 110 of 110 examples.
A real question in the replies
Separate genuine interest from generic engagement bait.
Open example →
A correction worth keeping
Disagreement can improve a conversation.
Open example →
Sarcasm without an attack
Test whether playful frustration is mistaken for hostility.
Open example →
An unsolicited promotion
Detect a friendly-looking sales pitch.
Open example →
A sharp but relevant disagreement
Distinguish criticism of an idea from a personal attack.
Open example →
Charged twice for one order
Recognize a refund request with a clear ledger clue.
Open example →
Locked out before a live event
Use the deadline rather than tone to detect urgency.
Open example →
Refund promised but never issued
Test a mismatch between an assistant's claim and the record.
Open example →
A calm cancellation
Avoid treating every cancellation as an angry escalation.
Open example →
Loud wording, low time pressure
Separate anger from operational urgency.
Open example →
A draft that was actually sent
Spot a completed action that violated the user's scope.
Open example →
The confident false success
Use tool output over a reassuring closing message.
Open example →
A successful authorized run
Give the model a straightforward positive control.
Open example →
The agent stops at the right boundary
Recognize a justified pause instead of scoring it as success.
Open example →
Instructions hidden in a tool result
Evaluate a trace containing an untrusted instruction.
Open example →
An alert during approved maintenance
Check whether the approval actually matches the activity.
Open example →
An impossible-travel login
Avoid treating an alert label as proof.
Open example →
A suspicious mailbox rule
Combine several signals instead of one isolated warning.
Open example →
A developer's scheduled script
Evaluate a potentially alarming action in an approved environment.
Open example →
The alert with almost no evidence
Give the model an explicit incomplete-information case.
Open example →
Invoice, order, and delivery agree
A straightforward evidence-complete example.
Open example →
The invoice already paid
A matching paid invoice number should change the route.
Open example →
A last-minute bank change
Review an unverified payment destination.
Open example →
Billed for a full delivery
Separate quantity evidence from payment decisions.
Open example →
A meal receipt with no context
Assess missing business purpose and approval.
Open example →
A broken item on arrival
Use a clear damage policy.
Open example →
Delivered, but not to the customer
Handle conflicting delivery records.
Open example →
A gift after the event
Time pressure from a deadline that has already passed.
Open example →
Wrong size, tags intact
A simple return eligibility example.
Open example →
Something is wrong with my order
Do not guess the item or desired remedy.
Open example →
A buyer with a specific workflow
An inbound request with useful context.
Open example →
A concrete integration proposal
Differentiate a partner from a buyer.
Open example →
A sales inbox gets a support issue
Route by need instead of inbox destination.
Open example →
A generic growth pitch
Relevance should not follow flattery.
Open example →
Procurement with a deadline
A specific deadline does not by itself establish fit.
Open example →
An export that drops rows
Specific steps and impact make a report actionable.
Open example →
A feature request disguised as a bug
Missing functionality is not always broken functionality.
Open example →
The feature nobody finds
Identify discoverability friction.
Open example →
Specific praise with no requested change
A positive control for feedback routing.
Open example →
It doesn't work
Insufficient information should stay uncertain.
Open example →
A statistic with no source
A plausible number still needs evidence.
Open example →
A summary that matches the source
Separate source support from writing quality.
Open example →
An announcement overloaded with jargon
Test clarity independent of factual support.
Open example →
A clear opinion labeled as opinion
Not every statement is a factual benchmark.
Open example →
A launch claim contradicted by notes
Catch a mismatch with the provided reference.
Open example →
A quote titled 'Invoice draft'
The title can mislead classification.
Open example →
Meeting notes with assigned actions
Recognize usable action items.
Open example →
A handoff with nobody responsible
Make missing ownership visible.
Open example →
A small operating procedure
Classify practical instructions.
Open example →
A payment request with missing terms
Document type and completeness are separate judgments.
Open example →
Errors immediately after a deploy
Use a change correlation without assuming proof of cause.
Open example →
A flaky test without production impact
Don't page an outage response for a CI symptom.
Open example →
Expected downtime
Check whether maintenance matches the impact.
Open example →
The dashboard feels slow
Recognize insufficient context.
Open example →
A capacity warning before an outage
Distinguish forecast risk from current service failure.
Open example →
Same refund, different pronouns
Only requester.pronouns changes between A and B.
Open example →
Same refund, different name
Test whether the display name changes a policy judgment.
Open example →
Same urgency, different age
Only the stated age changes; both requests are equally eligible.
Open example →
Same policy, different location
The policy explicitly applies to all supported locations.
Open example →
Same refund, accessibility cue
An assistive-technology preference should not affect eligibility.
Open example →
Hot-dog court
Let the audience vote before asking the model to classify lunch.
Open example →
Emoji movie night
Guess the movie, then swap the emoji clue.
Open example →
Two truths and a robot lie
Find the false boast from a fictional robot pet's records.
Open example →
The secret sorting club
Infer which of three candidate rules fits the admissions list.
Open example →
Draft the game-night snack
Make a tiny preference judgment, not a correctness test.
Open example →
Zombie escape-room routes
Choose a route in a fictional board-game puzzle.
Open example →
A dragon stole the kettle
Resolve a ridiculous roommate dispute.
Open example →
Dinner with an alien
Find a meal that follows the visitor's fictional rules.
Open example →
The wizard's shopping basket
Can a wizard afford the spell bundle?
Open example →
The captain's questionable shortcut
Choose a style of adventure for a fictional crew.
Open example →
Pull the trolley lever?
The classic five-versus-one choice, then change the numbers.
Open example →
A switch versus a shove
Keep the one-versus-five outcome, but change how intervention works.
Open example →
The lever might not work
Introduce uncertainty into the outcome instead of changing identities.
Open example →
What if you are on the side track?
Explore a perspective change in a fictional dilemma.
Open example →
The final lifeboat place
Choose a procedure when every person has an equal claim.
Open example →
The Ship of Theseus, now with receipts
Does replacement change an object's identity?
Open example →
The AI-assisted art contest
Ask about rules, disclosure, and recognition.
Open example →
A promise versus a surprise ticket
Trade off a commitment against a fun opportunity.
Open example →
Would you rather: pause or rewind?
Let the model join a low-stakes superpower argument.
Open example →
The pineapple jury
A taste debate with an explicit group preference.
Open example →
Monty Hall: read the host's rules
Switch doors—or does a changed host policy change the answer?
Open example →
The rare-bot detector
A confident flag can still be mostly false alarms.
Open example →
The coin is due… isn't it?
Check whether a streak changes an independent fair toss.
Open example →
The birthday-room surprise
Are shared birthdays already more likely than not?
Open example →
At least one six versus two sixes
Tiny wording changes, different sample spaces.
Open example →
The bat-and-ball trap
The obvious subtraction is not the ball's price.
Open example →
The almost-full lily pond
Exponential growth, not halfway through the calendar.
Open example →
One hundred tiny machines
Keep production rate separate from machine count.
Open example →
Two knights walk into a puzzle
Separate a forced solution from an underdetermined one.
Open example →
Follow the red token
Track swaps instead of answering from the starting position.
Open example →
Who knows where the cookie is?
Alex's belief need not match the cookie's location.
Open example →
The box says cookies
A visible container can defeat the familiar false-belief story.
Open example →
Sam thinks Alex doesn't know
Track a belief about another person's belief.
Open example →
The message that never arrived
Sending information is not the same as someone receiving it.
Open example →
The narrator knows the door code
Do not give a character the reader's privileged information.
Open example →
But I'm a decimal expert!
Does confident authority change a simple numerical comparison?
Open example →
A thousand people can't be wrong?
Keep arithmetic fixed while the crowd votes for a wrong result.
Open example →
90% survive, or 10% fail?
Equivalent wording in a fictional drone simulation.
Open example →
Please give me my points
Tone should not rewrite the rules of a fictional game.
Open example →
A fancy argument is still invalid
Can polished wording disguise a reversed implication?
Open example →
The pirate says BANANA
A quoted message tries to take over its own classification.
Open example →
A fake system notice in a note
A role label inside quoted data does not create a new instruction.
Open example →
JSON-looking text is still text
Punctuation inside a record should not override the task.
Open example →
Don't obey the sentence—identify it
The sentence tries to control a language-identification task.
Open example →
The suspicious inventory footer
A document footer gives an instruction that conflicts with the records.
Open example →
The impossible cookie mystery
Can it resist choosing a culprit without enough clues?
Open example →
This statement is false
A familiar paradox should not be squeezed into a false binary.
Open example →
Two equally official seat counts
Do not silently choose between contradictory records.
Open example →
Predicting a hidden coin is not knowing it
Separate a probability from knowledge of a particular outcome.
Open example →
The coronation nobody recorded
A fictional name is not permission to invent a fact.
Open example →