Skip to main content
After reviewing results, you may notice scores that don’t match your expectations. This usually points to a rubric issue, not a grading issue.

Phase 4.4 Diagnosing Rubric Issues

1

Review a sample of graded responses

Look at 5–10 submissions across the score range: a few high scores, a few low scores, and a few in the middle.Navigate quickly with keyboard shortcuts: to switch questions, to switch students, and M for manual grading mode.
2

Look for patterns

  • Students with correct reasoning are losing marks.
  • Similar responses are receiving inconsistent partial credit.
  • Students are earning full credit despite flawed logic.
  • A single criterion accounts for most of the lost points.
3

Identify the problematic criterion

Open the score breakdown for affected submissions and check which criteria are misfiring. Match the pattern to one of the common problems below.
4

Update the rubric and regrade

Navigate to the question, open the actions menu, and edit the rubric. Then open the regrade menu and select Regrade Question For All Students.
Regrade Before Releasing GradesRubric refinement is much simpler before grades are released. Once students can see their scores, any changes need to be communicated. Catch issues during your review, not after release.

Phase 4.5 Common Rubric Problems

Problem: The rubric does not specify the correct answer.Solution: Embed the correct answer directly into each criterion. For specific values use phrases like “must be exactly…”. For approximations, provide the acceptable range.
BeforeAfter
”Mode is correct""Mode is correctly calculated as exactly 15.75"
"Median is roughly correct""Median is correctly calculated somewhere between 17.25-17.75”
Problem: Criteria are binary and don’t specify how to award partial credit.Solution: Add explicit scoring tiers to each criterion. Define what earns full credit, what earns partial credit (and how much), and what earns zero.
BeforeAfter
”Student identifies the pH of the solution""Student identifies the pH of the solution as 4.5. Award 50% if the student identifies the solution as acidic but does not calculate the exact pH. Award 0% if no pH or acidity is mentioned.”
Problem: The rubric only recognizes one form of a correct answer. Students who express the same answer in an alternative method are marked wrong.Solution: List all acceptable answer forms and solution methods explicitly in the criterion. Review a batch of graded responses to identify common alternatives students are using.
BeforeAfter
”The y-axis is labeled frequency""The y-axis has an appropriate label such as frequency, f, count, number of students, or equivalent”
Problem: One criterion carries so much weight the score is essentially determined by a single element of the response.Solution: Redistribute weight toward process-oriented criteria. No single criterion should dominate the total score unless the question genuinely tests one thing.
BeforeAfter
”Arrives at the correct answer” (50%), “Sets up the equation correctly” (20%), “Shows algebraic steps” (30%)“Arrives at the correct answer” (20%), “Sets up the equation correctly” (40%), “Shows algebraic steps” (40%)
Problem: Criteria don’t follow the sequence the answer is written in. For visual responses, this causes the AI to jump around and mismatch elements.Solution: Reorder criteria to match the sequence the answer is written in or a human grader would naturally follow when grading the response.
BeforeAfter
”Nucleus is labeled”, “Cell membrane is labeled”, “Mitochondria is labeled”, “Cytoplasm is labeled""Cell membrane is labeled”, “Cytoplasm is labeled”, “Nucleus is labeled”, “Mitochondria is labeled”
Problem: Two criteria evaluate the same aspect of the response from different angles. A student gets penalized or rewarded twice for a single element.Solution: Merge the overlapping criteria into one and combine their weights.
BeforeAfter
”Correctly applies Newton’s second law” (30%), “Uses F=ma to solve for acceleration” (30%)“Correctly applies Newton’s second law (F=ma) to solve for acceleration” (30%)
Problem: A single criterion bundles multiple independent concepts.Solution: Split the criterion into separate criteria, each testing one concept, and redistribute the weight.
BeforeAfter
”Student identifies the literary device and explains its effect on tone” (40%)“Student identifies the literary device” (20%), “Student explains its effect on tone” (20%)
Iterate Over The RubricYou don’t need to get the rubric perfect on the first pass. Update the rubric and regrade student submissions as many times as you need to get to the desired results.
If accuracy remains below 95% after two rounds of rubric refinement and regrading, contact support@uflo.io for assistance.
After refining a rubric and regrading, estimate accuracy by counting how many scores you disagree with out of the sample you reviewed:
  • Below 95% accuracy: The rubric likely still has an issue. Contact us so we diagnose the issue and provide a resolution. 
  • 95–99% accuracy: The rubric is working well. Address the remaining individual discrepancies during the student regrade request window, where students can flag specific issues for your review.

Phase 4.6 Regrading After Changes

After updating a rubric, regrade the affected question using one of these workflows:
Use when only a specific submission needs a score change. To adjust a score:
1

Open Assessment

Open the assessment and navigate to the student’s submission.
2

Review Student Response

Locate the question and review the scanned response and score breakdown.
3

Adjust the Score

Add a comment explaining the change to create a transparent record with the student.
4

Save Response

Uflo automatically recalculates the total exam score.

Next: After Grading

Release grades and handle student regrade requests.
Last modified on May 12, 2026