Most problem parts in Edfinity are auto-graded — multiple choice, short answer, and algorithmic parts have definite correct answers. But some problem parts ask students to show their reasoning, write an explanation, or upload work. Edfinity's free response grader lets you auto-grade these open-ended responses as well, using a rubric that you define.
You create a rubric that describes how a response should be scored — what to look for, how much each element is worth, and what feedback to give. Edfinity then uses AI to grade every student submission against that rubric, returning a score and per-criterion feedback automatically.
To set up rubric auto-grading:
- Create and publish a rubric for a free response problem part using the Rubric Lab.
- Start using the rubric to auto-grade student responses.
What is a rubric?
A rubric is a set of criteria that define how a response is scored. Each criterion describes one aspect of a good answer — for example, "correctly applies the chain rule" or "includes units in the final answer."
There are two types of criteria:
- Scaled — graded on multiple levels. For example, a criterion might have three levels: "not addressed" (0 points), "partially correct" (2 points), and "fully correct" (4 points). The grader selects the level that best matches the student's work.
- Checklist — binary, either present or not. Used for specific elements that are either there or missing, like "includes a diagram." Checklist criteria can also carry negative weight for deductions (e.g., "answer includes extraneous information").
The final score is the sum of points earned across all criteria, divided by the maximum possible weight, normalized to the part's point value.
Each criterion can also include a student hint — a short note shown to the student when they don't earn full marks on that criterion (e.g., "Remember to state the domain of your function").
You can also include grading notes, which are separate from criteria. They're instructions that guide the grader's judgment — for example, "Accept equivalent forms" or "Do not penalize minor arithmetic errors." Students don't see grading notes.
Why sample answers matter
A rubric defines what to grade, but that alone doesn't guarantee it will grade well. A rubric that looks intuitive to an instructor may still confuse an auto-grader when applied to real student work. Sample answers close that gap — they're example student responses at different quality levels (strong, partial, weak, off-topic) with the scores you expect each one to receive.
Sample answers serve as the ground truth for your rubric. By spelling out "this response should earn 7/10 because criterion A is fully met but criterion B is only partial," you make your grading intent concrete and testable. The Rubric Lab can generate samples for you, and you can also add your own — including captures of real student submissions once the assignment is live.
Ensuring consistent auto-grading
Before a rubric goes live, you want confidence that it will grade consistently — not just on the samples you wrote, but on the unpredictable range of answers students will submit. Rubric Lab's scoring step works like a test harness: it grades each sample answer multiple times and checks whether the scores land in the same place every time.
If a criterion is ambiguous or a sample's expected score doesn't match what the rubric actually produces, validation surfaces the discrepancy. You can then triage the results — Rubric Lab analyzes the failures and recommends specific changes to the rubric or the samples. This iterate-and-retest cycle continues until the rubric grades accurately and consistently, giving you confidence before students ever see it.
Supported part types
The following problem part types support creation of an associated rubric for auto-grading:
- Essay
- File upload
- Show your work - when you add a rubric to a Show your work part, the rubric understands the context of the parent part.