Skip to main content

Security and Governance

Lumen’s trust model is based on controlled evaluation, not unattended automation.

Teacher approval​

AI suggestions do not become final marks automatically. Teachers approve, override, or route work into moderation before finalization.

Auditability​

Evaluation records preserve:

  • assessment context
  • rubric version
  • extraction confidence
  • suggested mark
  • explanation
  • review state
  • override decision
  • final approval
  • export state

Moderation​

Moderation catches low-confidence, disputed, or policy-sensitive cases. This gives institutions a defensible review path before results leave the system.

Data handling​

Institutions define how long answer sheets, extracted text, evaluation records, and export artifacts are retained. Retention matches institutional policy and pilot scope.

Export governance​

Final export follows teacher review and moderation policy.