A rubric is a scoring map—criteria, performance levels and concrete anchors. How to write one two raters can't disagree with.
A rubric defines what each level of performance looks like before anyone grades: rows are criteria, columns are levels, cells are observable descriptions. Its job is to move judgment from the grader's head into shared, citable language.
Criteria × Levels × Anchored descriptors (+ optional point weights per cell).
A presentation rubric's top cell reads "answers questions with evidence, admits unknowns"—not "excellent". Two raters, same grade, no meeting needed.
Anything scored by humans on subjective material: projects, interviews, exams, portfolios—and every calibration session you'll ever run.
Adjective-only cells ("good", "adequate") outsource the decision back to the grader. If it can't be observed, it isn't an anchor.
Holistic (one overall level) is fast for high-volume scoring; analytic (per-criterion) gives feedback and reliability data. Certifications want analytic.
3-5. Beyond five, raters start inventing differences that aren't in the work.
Turn the concept into a live assessment：Rubric
Browse templates →