Translate standards into observable, distinct, and consistently scorable criteria
9 min setupTested with:ChatGPTReviewed: 2026-08-28
1
Add your context
Your text stays in this browser. AILesson Prompts does not send it to a model or server.
2
Your prompt
Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt
Design a rubric strictly from the supplied task and standards.
Task, standards, objectives, evidence, and assistance rules:
[task]
Level, type, performance levels, points, weights, and use:
[scoring]
Access, language, formats, collaboration, and separation constraints:
[fairness]
Map requirements to criteria and flag requested criteria not supported by the standards. Use observable evidence, not traits, effort guesses, or vague adjectives. Keep criteria non-overlapping and avoid double penalties. Describe coherent differences in quality, accuracy, completeness, independence, or transfer at every level; do not rely only on error counts or labels such as excellent. Verify points and weights. Add rules for missing, partial, contradictory, and format-equivalent evidence, plus what is not scored. Provide two borderline calibration cases without inventing student work. Audit alignment, clarity, accessibility, bias proxies, overlap, arithmetic, and inter-rater reliability.
Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.
From input to outcome
A worked example
See how concrete context turns this recipe into a usable result
Actual input
Task and standards
Adult workplace course: give a 4-minute recommendation from a provided dataset. Standards: decision-relevant claim with accurate evidence; one material limitation; organization for the named audience. Notes allowed; no outside data.
Scoring design
Analytic, Ready/Developing/Not yet, 20 points. Evidence accuracy at least half. Certification and feedback.
Fairness constraints
Live or prerecorded. Do not score accent, charisma, slide design, or eye contact. Captions and assistive technology allowed. Keep claim quality separate from organization.
Example output
Weights: claim and evidence 10; material limitation 5; audience organization 5; total 20.
Claim and evidence — Ready 9-10: an actionable recommendation uses dataset values accurately and explains their connection without outside data. Developing 5-8: the recommendation is identifiable, but relevance, selection, accuracy, or connection is incomplete; a material numerical error caps this at 6. Not yet 0-4: no actionable claim, evidence is absent or substantially inaccurate, or outside data is essential.
Limitation — Ready 5: an evidenced limitation is explained in terms of confidence or use. Developing 2-4: relevant limitation, incomplete consequence. Not yet 0-1: absent or unrelated.
Organization — Ready 5: the audience can locate recommendation, support, limitation, and next step within four minutes. Developing 2-4: elements exist but ordering obscures a decision need. Not yet 0-1: essential elements cannot be located.
Live and prerecorded evidence use identical descriptors. Do not score accent, charisma, eye contact, visuals, or production polish. Calibration A: accurate evidence plus a named but unexplained limitation is Ready-range on evidence and Developing on limitation. Calibration B: polished organization with a decision-changing number misstated cannot be Ready on evidence. Calibrate the 8-to-9 boundary on one shared sample before certification scoring.
Why this works
1
Observable descriptors reduce impression-based scoring and make feedback actionable.
2
Calibration cases expose ambiguous boundaries before real grading.
Check the result
Can two scorers locate the same evidence for each descriptor?
Are criteria distinct, aligned, and arithmetically correct?
Does the rubric avoid unsupported style or access penalties?
ChatGPT