A human example
An incident needs one of hundreds of service and issue categories.
What the caller supplies
The caller supplies a meaningful hierarchy, a question and the incident description.
What happens next
Jev selects a path within a call budget or requests review.
Illustrative example, not a recorded result.
Potential value: medium
Useful when categories are semantic and the input does not already contain a usable ID.
Evidence confidence: low
The current 320-category experiment is synthetic, and a deterministic parser solved every case.
The rating describes support for this claim. It is separate from Jev's returned probability. How we assign ratings.
Evidence, including disagreement
- Navigating beyond a flat choice limit. first-party. Trees avoid truncation; structured data may need no model.
The next test
60 natural descriptions over a fixed hierarchy, with ambiguous siblings and no category fit.
Compare against
- Structured parser
- Keyword rules
- Flat selection where it fits
Measure
- Correct leaf
- Review rate
- Calls per case
- Latency
Decision after the test
Keep the hierarchy only if it beats the parser or rules on natural descriptions within the same call budget.
The report will retain inputs, question versions, every attempt and failure examples. We will update the confidence rating after reviewing the result.
Use a related Engine recipe
Recipes are implementation starting points. Their presence does not mean the protocol above has passed.