Score a message against a written rubric

Two anchored Score questions rate clarity and tone. Watch the markers move as the draft improves.

  • General decision tool with a code-side send rule
  • Tool sysone_decide
  • Recorded examples here. Live runs through your account or engine.

Recorded example: Draft 1: vague. The interactive version loads with JavaScript.

Evidence and options

A payment outage is affecting orders. Support is writing to customers who asked whether they should retry.

we're aware of issues. pls wait.

Recorded Jev answer, September 24, 2026

  • Score clarity: 1.77 of 3
  • Score tone: 1.05 of 3

Clarity 1.8 and tone 1.1 on 0 to 3 rubrics.

Revise for tone

Tone is below your 2.0 rule. Aim for the next anchor: "Neutral and polite".

Measured when recorded: 207 ms in the engine, 1 model call, 508 input tokens. A recorded answer, not a live run.

Exact typed request for this example
{
  "state": {
    "situation": "A payment outage is affecting orders. Support is writing to customers who asked whether they should retry.",
    "message": "we're aware of issues. pls wait."
  },
  "questions": {
    "clarity": {
      "type": "score",
      "instructions": "How clearly does the message tell the customer what happened and what to do next? Judge only the message text.",
      "criteria": [
        "The reader cannot tell what happened or what to do",
        "The reader learns what happened but not what to do",
        "The reader learns what happened and what to do, but a detail is vague",
        "The reader learns what happened, what to do and when to expect an update"
      ]
    },
    "tone": {
      "type": "score",
      "instructions": "How respectful and calm is the message toward the customer? Judge only the message text.",
      "criteria": [
        "Blames or dismisses the customer",
        "Curt or defensive",
        "Neutral and polite",
        "Warm and calm, and takes responsibility"
      ]
    }
  }
}

Two independent Score questions, clarity and tone, each with four written anchors.

How this experiment works

  1. Code Skip drafts shorter than 12 characters.
  2. Code In live mode, wait for a pause, cancel older requests and stop after a set number of runs.
  3. Code Apply your send rule to both scores.
  4. Jev Two independent Score questions, clarity and tone, each with four written anchors.
  5. You Write anchors that describe situations, not degrees.
  6. You Choose the send rule from labeled examples.
  7. You Read the draft before sending. A score does not approve it.

Limits

  • A score is a position between anchors, not a percentage or a grade.
  • Confidence describes the spread of an answer, not measured accuracy.
  • Live mode sends a new evaluation after each pause and can use your allowance.
  • Recorded examples are real Jev answers replayed only for an identical request. An edit that changes the request needs a live run.