Home/Jev Score
SJev Primitive

Jev Score

Jev Score places a state on a defined, ordered scale — such as low, medium, high, or critical. It is the rating tool of the Jev toolkit, for when you need to assess quality, risk, or priority against a rubric.

What Score does

You define a scale — for example, low, medium, high, critical. You give Jev the input (a document, a customer profile, a support interaction). Jev evaluates the input against your rubric and returns one level from the scale.

The return value is an ordered level. Your code can compare levels, route based on thresholds, or display them to users. The scale is defined upfront, so the output is always one of your defined levels — never an arbitrary number.

When to use Score

  • Quality assessment. Rate reply quality, content relevance, or code review severity.
  • Risk scoring. Assess fraud risk, compliance risk, or security threat level.
  • Priority assignment. Rate urgency as low, medium, high, or critical.
  • Lead qualification. Score lead intent or fit on a defined scale.
  • Sentiment rating. Classify sentiment as positive, neutral, or negative.

When not to use Score

  • Numeric prediction. Score returns an ordered level, not a number. Use code for exact numeric work.
  • Unordered categories. If the options have no ordering (billing, technical, sales), use Choice.
  • Binary questions. If the answer is yes/no, use Noul.
  • Deterministic rules. If you can compute the rating with code, do that instead.

Defining levels and criteria

Each level in a Score should have a distinct, testable definition. Write criteria that explain what makes an input "low" versus "medium" versus "high." If two levels overlap, the decision will be unreliable.

Usually three to five levels is enough. More levels are fine only if each one has a clear, distinct definition. Fewer levels are easier to test and more reliable.

Avoid hard-coded thresholds. If a case is unclear, route it to a qualitative fallback — a human, an LLM, or a request for more information — instead of a fixed cutoff.

Example: lead intent

You receive a new lead from a form submission. You want to rate their intent as low, medium, or high. Jev Score reads the submission and returns one level.

Decision Spec (excerpt)Score
{
  "type": "score",
  "question": "How strong is this lead's intent?",
  "input": { form_data: lead.submission },
  "levels": ["low", "medium", "high"],
  "criteria": {
    "low": "Vague inquiry, no timeline",
    "medium": "Defined need, exploring options",
    "high": "Clear budget, timeline, decision maker"
  },
  "fallback": "low → nurture sequence"
}

Score in a Decision Spec

A Score Decision Spec includes the question, the input fields, the ordered list of levels, the criteria for each level, and the fallback. Jev Fit + Builder generates this as editable JSON, so you can refine the levels and criteria before generating code.

The code shape

TypeScript
const result = await jev.score({
  question: "How strong is this lead's intent?",
  input: { form_data: lead.submission },
  levels: ["low", "medium", "high"],
});

if (result.value === "high") {
  routeToSales(lead);
} else if (result.value === "medium") {
  scheduleFollowUp(lead);
} else {
  addToNurture(lead);
}

Choice, Noul, or Score?

Choice

Pick one from a set of options.

Noul

Answer a yes/no question.

Score

This one. Rate on an ordered scale.

Try it on your workflow

Describe a scoring or rating task and the analyzer will identify Score steps.

Analyze a scoring workflowWhen to use Jev
FAQ

Frequently asked questions

It's a Jev primitive that places a state on a defined, ordered scale — such as low, medium, high, or critical.

Still have questions?

Contact support@jevmodel.co →