# Rubric Builder > Builds grading rubrics - analytic, holistic, or single-point - with criteria traced to learning objectives, 3-5 performance levels, observable behavior in every cell, and defensible weighting. Use when someone asks "make a rubric for this essay", "how should I grade this project", "turn these objectives into scoring criteria", or "my rubric feels vague, fix it". Do NOT use for writing the quiz questions themselves - use quiz-generator instead; for rubrics that score job candidates in interviews, use screening-rubric-builder; for narrative comments on a specific student's work, use student-feedback-writer. - Skill: `skillmedev/rubric-builder` (Agent Skill) - Install (CLI): `npx skillmds@latest add skillmedev/rubric-builder` - Raw SKILL.md: https://api.skillmd.com/api/skills/skillmedev/rubric-builder/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: SkillMedev (https://skillmd.com/u/skillmedev) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/skillmedev/rubric-builder --- # Rubric Builder A rubric does two jobs: it tells students exactly what mastery looks like before they begin, and it gives teachers a consistent, defensible standard when scoring. The costly failure this skill prevents is the adjective ladder - "excellent / good / fair / poor" - which students cannot act on before submitting and teachers cannot defend when a grade is challenged. ## Operating procedure ### Step 1: Gather inputs 1. The assignment or task being assessed (required - the actual prompt, not a summary). 2. The learning objectives it assesses (required; if the user has none written, draft them first and label them as guesses for confirmation). 3. Grade or level, and the stakes (minor credit vs major grade - this drives format). 4. Point total, if the gradebook requires one (default: 100). 5. Whether a student-facing version is needed (default: offer one). ### Step 2: Derive criteria from objectives List the objectives, then derive one criterion per objective or per cluster of tightly related objectives. Trace test: if a criterion cannot be traced to a stated objective, cut it. Practical ceiling: 3-6 criteria; beyond 6, scoring time balloons and criteria start overlapping. Mechanics/conventions may earn one criterion at most, and only if the objectives include communication quality. ### Step 3: Choose the format - Analytic (each criterion scored separately): default for any graded assignment worth more than minor credit; the only format that produces diagnostic feedback per criterion. - Holistic (single overall score): quick checks and short, low-stakes tasks only. - Single-point (describes proficient only, with narrative space either side): efficient for iterative draft work where the goal is revision, not a grade. ### Step 4: Set the performance levels 3-5 levels; four is the standard: Exceeds Expectations, Meets Expectations, Approaching Expectations, Does Not Yet Meet Expectations. Three levels work for simple tasks. Five is the ceiling - more levels create scoring drift between graders and are rarely worth the calibration effort. Label levels with descriptive language, not just numbers, so feedback is inherent in the score. ### Step 5: Write descriptors - proficient first, observable only Write the Meets Expectations cell first for each criterion, then adjust up and down. Every cell must describe observable features of the work - what it contains, how it is structured, what is missing - never grader impressions. Banned words in cells: excellent, good, poor, adequate, some, few, needs improvement, appropriate. Cell test: could two graders who have never met apply this cell to the same paper and land on the same level? Adjacent levels must differ by a nameable feature, not an adverb ("cites two or more sources" vs "cites one source", not "cites well" vs "cites adequately"). ### Step 6: Weight the criteria Assign points or percentages reflecting each criterion's importance to the objectives - the central skill should carry the most weight (typically 30-40% of total), mechanics rarely more than 10-15%. Document the weighting rationale in one sentence so it can be shared with students and defended to stakeholders. ### Step 7: Calibrate before use Score one strong, one middling, and one weak sample (real or imagined) against the rubric. Any cell that forced a judgment call gets rewritten before the rubric ships. ## Skeleton: analytic rubric Copy and replace the [FILL] fields. ``` RUBRIC - [FILL: assignment] - [FILL: total points] Assesses objectives: [FILL: list] Weighting rationale: [FILL: one sentence] | Criterion (weight) | Exceeds (pts) | Meets (pts) | Approaching (pts) | Not Yet (pts) | | [FILL: criterion 1] ([FILL]%) | [FILL: observable features beyond proficient] | [FILL: observable features of proficiency - write this cell FIRST] | [FILL: named gap vs Meets] | [FILL: what is absent] | | [FILL: criterion 2] ([FILL]%) | [FILL] | [FILL] | [FILL] | [FILL] | | [FILL: criterion 3] ([FILL]%) | [FILL] | [FILL] | [FILL] | [FILL] | | TOTAL | | | | [FILL] pts | ``` Example of a filled cell pair (essay, criterion "Use of evidence", Meets vs Approaching): - Meets: "Every body paragraph includes at least one quoted or paraphrased source, cited, and followed by a sentence explaining how it supports the claim." - Approaching: "Most body paragraphs include a source, but at least one quotation is dropped in without an explanation connecting it to the claim." ## Deliverable Produce the rubric as a table - rows are criteria, columns are performance levels - with point values per cell, a total-points row, a header naming the assignment and the objectives assessed, and the one-sentence weighting rationale. Offer a student-facing plain-language version. ## Do NOT - Do not write adjective ladders; "excellent vs good" ranks work without describing it, so students cannot self-assess and grades cannot be defended. - Do not include a criterion untraceable to an objective; it grades something never taught. - Do not exceed five levels; fine gradations feel rigorous but produce inter-grader drift. - Do not let mechanics outweigh the target skill; a brilliant argument with comma errors should not score below a vacuous clean one. - Do not use holistic format for major grades; a single number gives the student nothing to fix. ## Quality bar - Every criterion traces to a stated objective; 3-6 criteria total. - Every cell passes the two-grader test and names observable features. - Adjacent levels differ by a nameable feature, not an adverb. - Weights sum correctly and the heaviest weight sits on the central skill. - The rubric was calibrated against at least three sample performances.