Add grading criteria

At a glance

A rubric on an activity defines how that question is scored. After you click Create Rubric, add one or more items:

• Manual Score — you award points in Lightning Grader.
• AI Grading — an AI model scores the response against your criteria.
• Code Grading — the sandbox runs the learner’s code and compares output (Code Editor only).

Creating a rubric does not add any items by itself. You must add at least one item.

This is different from Score modifiers in Lightning Grader (bonuses and deductions for the whole lesson). The grader button labeled Add Rubric Item opens a modifier, not a lesson-editor rubric item.

Prerequisites

• Lesson edit mode.
• An activity that supports rubrics (see the table below).

Which activities support which items

Multiple-choice, multiple-select, fill-in-the-blank, and similar blocks use built-in autograding. They do not have a rubric section.

Create the rubric

1. In the lesson editor, select the activity.

2. At the bottom of the block, click Create Rubric. The button disappears once a rubric exists.

3. A Rubric section appears with an empty Rubric Group.

4. Add items:

   • If only one type is available, click Add Manual Score.
   • If several types are available, click Add rubric item and choose Manual Score, AI Grading, or Code Grading.

5. You can add several items in the same group (for example, code tests plus an AI style criterion on a Code Editor).

6. Use each item’s menu for Copy, Paste, Duplicate, or delete. Delete Rubric on the rubric header removes the whole rubric.

Manual Score

Use this when a person should judge the work.

1. Add Manual Score.
2. Set Points (0–1000).
3. In Criterion, name what you are scoring (the placeholder is Manual Score).

On submit, the item is not autograded. Learners see This will be manually graded by your instructor until you score it (and grades are released). In Lightning Grader, click the score (- / N) and enter points.

On graphing questions, a Graph match row can appear next to Manual Score when a model solution exists.

AI Grading

Use this when you want the model to score written work, uploads, chat, or submitted code against a rubric.

1. Add AI Grading.
2. Set Weight (this is the item’s points; it also creates score levels from 0 through that number).
3. Enter a Criterion Name (placeholder: Example: Accuracy).
4. Open Grading Context and describe course norms, partial-credit policy, and how to handle ambiguity. Put shared context on the first AI item in the group; later items can add more.
5. Optionally add Keywords the model should look for (they influence emphasis; they are not all-or-nothing checks). Type a term and press Enter.
6. Under Scoring Criterion, describe what a response looks like at 0 pts, 1 pt, and so on. Click Info in the editor for the full strategy list, or see Best practices for AI grading.

All AI items in the same rubric group are graded in one pass. Learners see the score and student-facing feedback when grades are released. Instructors also see private Instructor Info notes. AI rows in Lightning Grader are autograded (not click-to-edit); use score modifiers or general feedback if you need to adjust the lesson total.

AI Chat activities also have a separate built-in Autograde panel (participation / objectives). That is not the same as a rubric AI item.

Code Grading

Use this on a Code Editor to autograde by running code.

1. Add Code Grading.
2. Set Points and a Title (optional Description is authoring-only).
3. Confirm Environment is Same as Starter Code.
4. Set Run command so it matches the files learners actually submit (defaults are like python3 main.py or javac Main.java && java Main).
5. Enter Input (stdin; one value per line) and Expected Output.
6. Choose Ignore Case, Ignore Whitespace, Comparison Method (Exact match, Regex, or Partial match), and whether to Show diff of actual output and expected output.
7. Optionally Add File under Files for a hidden test harness. At grade time, files merge in this order: student → course library → rubric files (later wins).
8. Click Run Test in the editor to preview before learners submit.

A passing run awards full item points. A failing comparison, a non-zero exit code, any stderr, a timeout (30 seconds), or no submitted files scores 0. See Best practices for code grading for how to write reliable tests.

Code-grading rows in Lightning Grader are autograded. You can still combine them with Manual Score or AI Grading in the same rubric.

Related articles

• Rubrics & scoring modes
• Grade a lesson with Lightning Grader
• Score Modifiers: bonuses and deductions
• Course blocks: code editor library
• Best practices for AI grading
• Best practices for code grading