Jev Dispatch.

What each round teaches.

Every round sends the same kind of request: a message as the state, plus typed questions. These are the questions the game asks, in the shape the Jev API accepts.

1. choice Route the inbox

   Fernbox support gets every kind of message. Your app asks Jev one Choice question: which team should handle this? Jev picks one of the options you wrote and gives each option a probability.

   {
       "questions": {
           "department": {
               "type": "choice",
               "instructions": "Which team should handle this Fernbox support message?",
               "criteria": {
                   "billing": "Charges, refunds, invoices, or the subscription itself",
                   "technical": "The app or website misbehaving: crashes, errors, broken buttons",
                   "account": "Signing in, passwords, email address, or profile settings",
                   "other": "Anything none of the teams above handles"
               }
           }
       }
   }

   Takeaway: Choice only picks from the options you supply, so give it a way out such as “other”. When the top two options are close, confidence drops. That is your cue to look before acting.

2. noul Catch the refund requests

   A Noul checks whether a statement is true and returns one number: the probability of yes. Messages arrive faster now, so read less and watch the meter.

   {
       "questions": {
           "is_refund_request": {
               "type": "noul",
               "instructions": "The customer asks for a refund or for their money back, directly or indirectly."
           }
       }
   }

   Takeaway: A Noul near 0.5 means Jev is unsure, not that a message is half a refund. Your code picks the thresholds: act above one, ignore below another, and review what falls between.

3. score Triage bug severity

   A Score rates text on levels you describe, from cosmetic to critical. Jev returns a position on that scale, such as 2.6, and a probability for every level.

   {
       "questions": {
           "severity": {
               "type": "score",
               "instructions": "How severe is this bug report for Fernbox customers?",
               "criteria": [
                   "Cosmetic: something looks wrong, but nothing is blocked",
                   "Minor: a feature misbehaves and a workaround exists",
                   "Major: a core feature fails for some customers",
                   "Critical: payments, data loss, or every customer is blocked"
               ]
           }
       }
   }

   Takeaway: A score of 1.46 sits between Minor and Major. Where to cut is a decision your code makes. One level off costs less than a wild guess, so rubrics forgive near misses.

4. all three One call, three answers

   Jev answers several questions about the same message in one request. This round has no Trust Jev button, because you are the code: apply the rules in order to Jev’s three answers.

   {
       "questions": {
           "is_urgent": {
               "type": "noul",
               "instructions": "The message conveys urgency or time-sensitivity."
           },
           "frustration": {
               "type": "score",
               "instructions": "How frustrated the customer appears.",
               "criteria": [
                   "Calm, just stating facts",
                   "Frustrated but civil",
                   "Very angry, strong language"
               ]
           },
           "topic": {
               "type": "choice",
               "instructions": "What kind of message is this?",
               "criteria": {
                   "how_to": "Asking how to do something the help centre already explains",
                   "problem": "Reporting something that is wrong or broken"
               }
           }
       }
   }

   Rules, first match wins:

   1. is_urgent at least 0.7 → Priority
   2. frustration at least 1.5 → Senior agent
   3. topic is how_to → Help article
   4. Anything else → Standard

   Takeaway: That was an if-statement. In your app, code runs these rules on Jev’s answers, and people only see what the rules send them.

Every message, answer, and company in this game is invented for practice. The game makes no Jev calls; the answers are written to resemble real response shapes. JevMade is an independent community site, not affiliated with or endorsed by TypeSafe AI.