Written feedback on every question, on every exam. The class roster stays out of the AI request.
Rubrica scores scanned, handwritten exams against the instructor's rubric, writes feedback on each question for each student, and shows the instructor where the class struggled. Stored names, student IDs, and cover pages are left out of every grading request. The instructor reviews the results and decides what goes to students.
A student or parent? Read the plain-language guide.
Illustrative example. Results lists students by random code until the instructor chooses to reveal names. Amber marks an exam near a letter-grade line. Red marks an answer to check.
The line between who a student is and what they wrote.
On the instructor's Mac
Sent for grading
Only when the instructor acts
Applies when cover pages are read by the local model, the default setup. A name written on an answer page, or typed into a rubric, goes with it. Rubrica does not detect or remove these, so instructors should check first. Ask students to write their name only on the cover page.
Every path outA score says how a student did. It does not say why.
Writing an explanation for every question on every exam is slow, so large courses often return a number and little else. Research on feedback points to what that costs. In one meta-analysis of item-level feedback, explanations had the largest effect, showing the correct answer had less, and saying only right or wrong had almost none.
Those studies used computer-based tasks, not handwritten university exams. Effects vary widely, and some feedback lowers performance. Rubrica has not yet measured whether its feedback improves learning. Sources are on the Evidence page.
Score only
With Rubrica
Every student gets the reasoning behind their score.
Each report shows the total and letter grade, a card for every question with its sub-parts, written feedback on each, and an overall summary. The instructor can edit any feedback before it goes out. Class-wide sends include a link to request a regrade by email.
Shifts supply left and correctly identifies the higher price and lower quantity.
- 1Total score, letter grade, and a score bar marked at each letter-grade line.
- 2A card for every question, colored by outcome: green for full credit, amber for partial credit, red for no credit.
- 3Sub-part rows, each with written feedback on what was right or wrong and why.
- 4An overall summary. The instructor can edit any feedback before it goes out.
- 5A link to request a regrade by email, included in class-wide sends.
Names and content are invented. Free-response feedback is shown. Multiple-choice feedback is deliberately short.
See where the class struggled, question by question.
The Analytics screen shows the score distribution and how each question performed. The diagnostic report adds a question table with average score, zero rate and full-credit rate, the lowest-scoring multiple-choice questions, and a comparison across exam versions. It then adds an AI-written summary of common error patterns and concepts to re-teach.
Score distribution
| Question | Max | Avg % | Zero rate | Full credit |
|---|---|---|---|---|
| Q1 | 10 | 84% | 2% | 63% |
| Q2 | 10 | 66% | 4% | 21% |
| Q3 | 5 | 43% | 27% | 9% |
| Q4 | 2 | 71% | 29% | 71% |
Top error patterns
- ~14 students named the externality but gave no efficient quantity (Q3b)
- ~9 calculated elasticity correctly but did not link it to total revenue (Q2b)
- ~5 reversed the direction of the supply shift (Q1)
Concepts to re-teach
- Finding the efficient quantity with an externality
- Elasticity and total revenue
The written summary draws on a sample of up to ten feedback comments per question, so its counts are estimates. Generating it sends those comments and class statistics to the model provider. It includes every graded exam, reviewed or not, so run it after review.
Four steps, with the instructor deciding what goes to students.
Scan and upload
Rubrica can read each cover page with a model running on the Mac. The instructor confirms the name, and the app links it to a random code.
Grade against the rubric
Each question is scored against the instructor's rubric, using its partial-credit levels where the rubric has them. Every score comes with written feedback.
Review what is flagged
Exams near a cutoff are graded twice. Hard-to-read work and possible errors are flagged with what to verify. The instructor can change any score or comment.
Return reports
Students receive a PDF report from the instructor's own Gmail. The instructor gets a class report showing how each question went, plus an AI-written summary of common errors and concepts to re-teach.
Early studies, reported with their limits.
Three small studies from March to July 2026, run by Rubrica's author. All three ran on development builds, before the first pilot build, and have not been re-run on the current version.
Average difference per exam from human scores on one midterm's structured section.
Compared with human scores, 25 examsLower than the human score on average on a 25-point essay, with the largest gaps on the strongest essays.
Compared with human scores, 25 examsOf structured-section scores matched the human score exactly (810 of 949).
Compared with human scores, 25 examsOf exams had the same letter grade from two automated graders (21 of 30), before safeguards.
Agreement between two automated gradersEach study is small. The Evidence page states what each one does and does not measure, and what has not been measured at all.
Read about the studiesDocumented for IT, compliance, and faculty review.
Stored on the Mac
Rosters, grades, and the link to each student are stored on the Mac. Rubrica runs no cloud service that holds them. An optional backup to the instructor's own iCloud Drive or OneDrive is off unless they turn it on.
Restricted access
Approved Google accounts only, plus a PIN at each launch. These control the app. They do not encrypt the files on disk.
Nothing sent on its own
Rubrica never emails a grade on its own. A class-wide send goes only to exams the instructor has marked reviewed.
Questions, press, or a pilot for your course?
Instructors, departments, and reporters can reach us by email. For a pilot, include your course, class size, and exam format.
Email hello@tryrubrica.com