Write an analytic rubric for a project and give feedback that learners can act on.
Why it matters
Practical tasks, projects and presentations cannot be marked by an answer key. Without clear criteria, two assessors give different scores, learners do not know what "good" looks like, and feedback becomes vague ("nice work, tidy up the code"). A rubric makes quality visible, and good feedback turns an assessment into the next step of learning.
What a rubric is
A rubric lists the criteria you will judge, the performance levels for each, and a description of what each level looks like. Susan Brookhart's How to Create and Use Rubrics for Formative Assessment and Grading (2013) is a practical reference.
Two main types:
- Analytic rubric: scores each criterion separately. Better for feedback, because learners see which part needs work.
- Holistic rubric: one overall score from a description of the whole. Faster, useful for quick judgments, but less diagnostic.
Example: analytic rubric for a Git team project
Task: in pairs, add a feature to a small app using a branch and a pull request.
| Criterion | 1 Beginning | 2 Developing | 3 Proficient | 4 Exemplary |
|---|---|---|---|---|
| Branching | Work committed directly to main | Feature branch used, but unrelated changes mixed in | One feature branch with only related changes | As 3, and branch name and scope follow the team convention |
| Commits | One large commit, unclear message | Several commits, messages vague ("fix") | Small commits, each message says what and why | As 3, and history tells a clear story a reviewer can follow |
| Pull request | No PR, or no description | PR with a title only | PR explains the change and how to test it | As 3, and responds to review comments with follow-up commits |
| Conflict handling | Conflict left unresolved or work lost | Resolved, but tests fail | Resolved, tests pass | As 3, and explains the resolution in the PR |
Notice what makes it work:
- Criteria match the objectives, not generic qualities like "effort".
- Descriptions are observable, so two assessors would agree.
- Levels describe quality, not just quantity ("several commits" vs "small commits, each with what and why").
- Four levels avoid a lazy middle choice; some teams prefer three.
Writing a rubric step by step
- List the objectives the task assesses.
- Turn each into a criterion (three to six is plenty).
- Describe proficient first: what meets the standard?
- Describe the levels below by what is missing or wrong, and the level above by what goes further.
- Test it: score two or three real samples with a colleague and compare. Rewrite any description you disagreed on.
- Share the rubric with learners before the task, ideally with an example of proficient work.
Feedback that moves learning forward
John Hattie and Helen Timperley (2007) framed effective feedback as answering three questions:
- Where am I going? (the goal, which the rubric shows)
- How am I going? (current performance against it)
- Where to next? (the specific next step)
Feedback can also do harm. Avraham Kluger and Angelo DeNisi's 1996 meta-analysis found that in roughly a third of cases, feedback interventions actually lowered performance, especially when feedback pointed attention at the person ("you're a natural") rather than the task.
Practical rules, drawing on Paul Black and Dylan Wiliam's work on formative assessment ("Inside the Black Box", 1998):
- Be specific and about the task: "Your commit messages say what changed but not why; add the reason in one clause" beats "improve your commits".
- Make it actionable now: give feedback while learners can still use it, before the final submission.
- Limit it: two or three points learners can act on, not twenty.
- Make learners do something with it: for example, ask them to revise one commit message and resubmit. Wiliam puts it this way: feedback should be more work for the recipient than the donor.
- Comments and grades: Ruth Butler (1988) found that students given comments only improved more than those given grades, or grades plus comments, which suggests a grade can draw attention away from the comment. When you can, give comments first and the grade later.
Peer and self-assessment
A clear rubric lets learners assess their own and each other's work. Have pairs swap pull requests and score them with the rubric before submitting. They learn the standard by applying it, and you get fewer weak submissions.
Common mistakes
- Criteria like "creativity" or "professionalism" with no description.
- Levels that differ only by adjectives ("good", "very good", "excellent").
- Rubrics shared only after marking.
- Long feedback lists with no clear next step.
- Praise aimed at the person rather than the work.
Quick recap
- A rubric = criteria + levels + observable descriptions.
- Analytic rubrics give better feedback; holistic rubrics are faster.
- Write proficient first; test with real samples and a colleague.
- Feedback answers where am I going, how am I going, where to next, and asks learners to act on it.
Try it
On the Do tab, write an analytic rubric for a project in your subject.