Grading and Attention
How SOL scores open responses, when items wait for you, and how to review or override scores.
SOL grades with rubric matches from the model and scores from application code. You remain the grader of record. Unsure answers wait in Attention instead of quietly inventing a score.
How short-answer grading works
- The model returns which rubric criteria the student met (and rationale)
- Application code computes the numeric score from those matches
- If the model is unsure, or the AI path is unavailable, the attempt surfaces for manual review
- You can accept, adjust, or regrade from the attempt page or Attention queue
This keeps grading auditable: every score should be explainable from criterion-level evidence.
Attention queue
Attention collects attempts that need a human eye: pending review, manual grading, or fallback paths.
Typical actions:
- Open the attempt and read the student answer + rubric matches
- Adjust the score when you disagree
- Regrade when materials or rubrics changed
- Resolve items so they leave the queue
Discussions
AI-evaluated discussions follow a similar pattern: the model evaluates against your prompts/rubric; you can audit sessions and intervene when needed.
What students see
Students see scores and feedback you have released through the normal results / grades flow. They do not see internal Attention state.
Honesty about AI
- Say “SOL suggests” and “you review,” not “SOL automatically knows”
- Prefer criterion language over opaque total scores alone
- Use Analytics after graded work to see which concepts to reteach
Related guides
- Authoring from materials
- Course analytics
- Faculty onboarding
- Security (AI and education records)
