Jev recipe / implementation steps

Detect spam with a typed Jev decision

Return an explicit review signal instead of hiding a binary classification inside free-form model text.

Build this decision in three steps

01Pass the message and the minimum sender metadata needed for context.

02Use a bounded decision and document what counts as spam for this product.

03Quarantine uncertain messages instead of deleting them automatically.

schema / example

{
  "is_spam": {
    "type": "noul",
    "instructions": "Is this message spam?"
  }
}

Copy this into the Playground, then replace the state and criteria with your own.