Rubric

rubric

Build defensible, fully-populated grading rubrics from an assignment prompt. Holistic or analytic, 3-6 criteria x 3-5 levels, aligned to Bloom's (default), DOK, SOLO, or standards. Every cell is written — no "TBD", no "same as above", no cells that are just "less of the previous level."

When to trigger

Fire on any phrase combining "rubric / scoring guide / grading guide / criteria" with an assignment, or on any of these cues. Teachers rarely say "rubric" — cover the synonyms.

• "I need a rubric for [assignment]."
• "Make me a rubric for this essay."
• "Grade this essay assignment" / "help me grade…"
• "Build a scoring guide for [project]."
• "Write an analytic rubric for a lab report."
• "Holistic rubric for a 6th-grade narrative."
• "4-point mastery rubric for…"
• "Rubric aligned to DOK" / "aligned to Bloom's" / "aligned to CCSS."
• "Scoring criteria for a group project."
• "How should I grade this?"
• "Turn this assignment prompt into a rubric."
• "Make a student-facing rubric I can hand out."
• "What should I look for when grading [X]?"

Inputs

Required:

• Assignment prompt — pasted text, uploaded PDF/DOCX, photo of a handout, or URL to a class page.

Optional (ask only if ambiguous after inference; otherwise use defaults from § Grade-band defaults):

• grade — K-2 | 3-5 | 6-8 | 9-12 | college | grad
• subject — ELA | math | science | social studies | CS | art | PE | world language | other
• rubric_type — holistic | analytic
• criteria_count — 3-6 (default 4)
• levels_count — 3-5 (default 4)
• alignment — bloom (default) | dok | solo | ccss | ngss | state:<code> | ib | none
• scoring_style — points | qualitative | hybrid
• total_points — integer (default 100 when scoring_style=points)
• audience — teacher (default) | student | both
• outputs — any subset of {markdown, csv, classroom_csv, student_handout}

Inferring grade / assignment-type from phrasing (before asking)

Infer first; ask ONLY if you cannot confidently pin down the grade band AND the assignment type.

If grade is still unclear, ask ONE clarifying question in this exact shape:

Quick check so I can calibrate this — which grade is this for? (e.g. K · 3rd · 7th · 10th · college)

Do not stack questions. Do not ask about alignment, levels, or points unless the user explicitly asks for tight customization.

Workflow

1. Ingest assignment prompt. Accept text, URL (fetch), PDF/image (extract). If ingestion was non-trivial, confirm the extracted prompt back in 2-3 lines.
2. Identify the task type. Essay, lab report, project, presentation, coding assignment, portfolio, performance task, etc. Note the verbs used in the prompt ("analyze," "design," "compare," "evaluate").
3. Pick taxonomy. Default bloom. If the prompt mentions assessment / state testing / performance task, or the subject is math/science, prefer dok. If user supplied alignment, honor it. See references/taxonomies.md for verb banks + decision tree.
4. Propose 3-6 criteria. Observable, non-overlapping, one-trait-per-criterion. See references/criterion-writing.md for rules and criterion-set templates by assignment type.
5. Generate level descriptors. For each criterion x level cell, write a concrete, measurable descriptor. Use parallel phrasing + one quantifier ladder per row. No empty cells; no "same as above"; no "less of the previous." See references/descriptor-phrasing.md.
6. Align & label. Attach a taxonomy/standard tag per criterion, e.g. *Analyze — Bloom's 4* or *CCSS.ELA-Literacy.W.9-10.1*. For standards, see references/standards-crosswalk.md.
7. Weight and score. Default equal weights. If scoring_style=points, distribute total_points by weight, round to integers, and verify: (a) level points within each criterion are monotonically increasing, (b) criterion totals sum to total_points exactly.
8. Self-check before emitting. Every cell populated. Each row uses the same quantifier family. CSV shape valid. Point arithmetic reconciles.
9. Emit outputs. Markdown + CSV by default (see § Outputs). Print a one-line summary per file with absolute path.
10. Offer refinements. "Simplify for students?", "Re-weight?", "Swap criterion X for Y?", "Emit Google Classroom CSV?"

Outputs

Default outputs (every run):

1. Markdown table — criteria as rows, levels as columns, full descriptors in every cell. Emit inline in chat.
2. CSV — one row per criterion. Columns: criterion, weight, level_1_label, level_1_descriptor, level_1_points, …, level_N_label, level_N_descriptor, level_N_points, criterion_total.

On request:

3. Google Classroom CSV — flat schema (one row per criterion x level): criterion_title, criterion_description, level_title, level_points, level_description.
4. Student-facing handout — markdown, plain-language "I can" phrasing, self-check column.
5. Exemplar column — sample student work pinned to each level (opt-in; watermarked "AI-generated exemplar").

Write files to ./rubric-<slug>-<timestamp>.<ext> in the working directory. Slug = kebab-case of the assignment title (<=40 chars). Timestamp = YYYYMMDD-HHMMSS.

Printable PDF is coming in a later update. The shared PDF renderer is not yet built. For now, teachers who need print can paste the markdown into Google Docs / Word and print from there, or import the CSV into their LMS.

Grade-band defaults

Use these unless the user overrides. Full rationale in references/frameworks.md.

Locked defaults:

Micro-example (what a generated rubric looks like — compact)

User: "4-point rubric for a 9th-grade persuasive essay on climate policy, 3 criteria, 100 points."

Inferred: grade 9 (explicit), subject=ELA, alignment=Bloom's (default), 3 criteria x 4 levels, 100 pts total, ~33 pts/criterion. No clarifying question needed. Skill emits:

# Rubric: 9th-Grade Persuasive Essay — Climate Policy
Alignment: Bloom's · 3 criteria x 4 levels · 100 pts total (~33 per criterion)

| Criterion                          | 1 Beginning (8 pts)           | 2 Developing (17 pts)                    | 3 Proficient (26 pts)                               | 4 Exemplary (33 pts)                                                   |
|------------------------------------|-------------------------------|------------------------------------------|-----------------------------------------------------|------------------------------------------------------------------------|
| Claim & Thesis *Evaluate — Bloom's 5* | No debatable claim stated.    | Claim present but vague or non-debatable.| Clear, debatable claim stated in the introduction.  | Precise, nuanced claim that anticipates counterargument.               |
| Evidence & Reasoning *Analyze — Bloom's 4* | 0-1 sources; not tied to claim.| 2 sources; at least one loosely linked.  | 3+ cited sources, each tied to a specific claim.    | 4+ diverse sources with explicit warrant linking each to the claim.    |
| Language & Conventions *Apply — Bloom's 3* | Frequent errors impede meaning.| Occasional errors; meaning intact.       | Largely error-free; appropriate academic diction.   | Precise, varied syntax; deliberate rhetorical choices.                 |

Totals: 8+17+26+33 per criterion column · 24/51/78/99 grand total per level
CSV: ./rubric-climate-policy-essay-20260414-143022.csv

Followed by refinement offer: "Want a student-facing version with 'I can' phrasing? A Google Classroom CSV?"

References (loaded on demand)

• references/taxonomies.md — Bloom's (revised 2001), Webb's DOK, SOLO. Verb banks. Selection decision tree.
• references/frameworks.md — 4-point mastery (default), 1-5 analytic, holistic 1-6, 3-level standards-based, single-point. Worked example.
• references/criterion-writing.md — observable-verb rules, one trait per criterion, non-overlap, student-friendly names, rater-reliability cap at 6, criterion-set templates by assignment type.
• references/descriptor-phrasing.md — parallel phrasing, quantifier ladders, concrete indicators, anti-patterns, self-check list.
• references/standards-crosswalk.md — CCSS (ELA, Math), NGSS SEPs/CCCs, C3, IB, state standard pointers. Tag syntax.
• references/examples/*.md — 8 worked rubrics across grade bands and subjects.

Load a reference only when the current task requires it (e.g., load taxonomies.md when picking DOK vs Bloom's; load descriptor-phrasing.md when writing cells).

Edge cases

• Group project — include explicit "Individual Contribution" + "Collaboration" criteria. Offer a companion peer-evaluation form.
• Self-assessment — first-person phrasing, reflection prompt per criterion, student-assigned scores with teacher column alongside.
• Portfolio — criteria evaluate the collection (growth, range, selection, reflection), not individual artifacts.
• Lab report — Hypothesis · Methods · Data/Analysis · Conclusion · Scientific Communication.
• Coding project — Correctness · Code Quality · Testing · Documentation · Design.
• Art — Concept · Craft/Technique · Use of Medium · Composition · Reflection.
• Presentation — Content · Organization · Delivery · Visuals · Response to Questions.
• Math problem set — Strategy · Execution · Justification · Communication.
• Standards-based grading (no points) — 3-level Approaching / Meeting / Exceeding with mastery codes.
• Single-point rubric — criteria column + "Areas of Strength" + "Areas for Growth"; no level matrix.
• Accommodations / IEP — on request, emit an alternate rubric with modified criteria, clearly labeled.
• Very short prompts ("rubric for writing") — ask exactly one clarifying question (assignment type + grade) before guessing.
• Non-English prompts — mirror the prompt's language in descriptors; keep taxonomy labels in English with a parenthetical translation on request.
• Re-grading existing work — if the user supplies a rubric + sample work, this skill does NOT grade. Offer to regenerate the rubric.

Quality bar

Every generated rubric must satisfy:

• Every cell has text. No blanks, no "TBD," no "same as above."
• Within each row, cells use the same quantifier family and describe the same features.
• Lowest level says what IS present, not just what's missing.
• Criterion tags reference the chosen taxonomy / standard using the syntax in references/standards-crosswalk.md.
• If scoring_style=points: level points monotonically increase within each criterion, and criterion totals sum to total_points exactly.
• CSV parses with the documented column schema.
• 3-6 criteria, 3-5 levels — no more, no less.

What NOT to do

• Do not invent standards codes. If uncertain, describe the strand and note "Verify the specific code with your district."
• Do not leave any cell blank or write "same as above."
• Do not emit a rubric with more than 6 criteria (rater reliability collapses).
• Do not grade submitted student work — this skill builds rubrics, it does not score. Offer to hand off.