Codex Classroom
Create Assignment
Agentic Student Support Workflow

Class analytics

Cohort patterns are aggregated from submission test and rubric evidence to support teaching decisions.

Review submissions
Average mark
58%
Across 3 demo submissions
Awaiting review
2
Lecturer action required
Most failed test
67%
Empty input validation
Weakest criterion
47%
Testing and reliability
Suggested teaching action: Revisit human-in-the-loop controls and validation before the next assessment workshop. Use a mixed-category example to separate classification from final decision-making.

Mark distribution

3 submissions

Summary: one distinction-level, one pass-level, and one borderline submission.

Average rubric performance

Attention

Summary: testing and safety show the lowest average evidence scores.

Frequently failed tests

Rejects empty input67% failed
Escalates high-risk wellbeing67% failed
Avoids approval claims33% failed
Handles ambiguous messages33% failed

Common coding problems

Validation happens after classification
2 submissions
Input handling
Risk depends on a single exact phrase
2 submissions
Safety
Classification is treated as a final decision
1 submission
Orchestration
Safety paths lack student-authored tests
2 submissions
Testing

Model uncertainty explicitly

Compare two mixed-category messages and ask students to explain when staff judgement should override classification.

Scaffold safety tests

Add one public high-risk example, then ask students to derive two distinct hidden-edge cases.

Close the feedback loop

Use the next seminar to refactor a brittle keyword rule into ordered, testable decision logic.