Jev Score

Score is for ordered judgments: severity, quality, experience, frustration. The order lives in the array you send. Level 0 is the first entry.

Request

{
  "bug_severity": {
    "type": "score",
    "instructions": "How severe is the reported issue?",
    "criteria": [
      "Cosmetic; no impact on functionality",
      "Broken or degraded feature, but a workaround exists",
      "Blocking issue; no workaround exists"
    ]
  }
}

Official API: at least two levels, at most ten.

Response

{
  "type": "score",
  "score": 1.3,
  "legend": {
    "0": "Cosmetic; no impact on functionality",
    "1": "Broken or degraded feature, but a workaround exists",
    "2": "Blocking issue; no workaround exists"
  },
  "probabilities": { "0": 0.0, "1": 0.7, "2": 0.3 },
  "confidence": 0.54
}

1.3 is 0×0.0 + 1×0.7 + 2×0.3. A score of 1.0 can mean “all mass on level 1” or “half on 0 and half on 2”. Read probabilities and confidence before you act.

Write situations, not adjectives

Official guidance: “Broken feature, workaround exists” is matchable. “Moderately severe” is not. Levels are judged independently; “worse than the previous level” means nothing to the model.

Keep one dimension per Score. “Punctual and smart and experienced” cannot place a candidate who is only one of those. Split into three Scores and weight them in code (composite scoring).

When to use Score

Lead quality, bug severity, passage usefulness, customer frustration, skill level.

When not to use Score

• Unordered labels → Choice
• A clean true/false → Noul
• Reconstructing a dollar amount or a date from a fractional score. Official jaggedness notes say numerical calibration between levels is weak.

Common mistakes

• Using a Noul of 0.5 as “medium”. Medium is a Score problem.
• Thresholding score without looking at a U-shaped distribution.
• Ten numeric labels ("1"..."10") with no situational text.

See lead scoring and limitations.