Grounding fidelity
Demo visual only. Production test: does the answer stay within approved course/context evidence?
AI Governance
No AI feature should go live because it looked impressive once.
The platform adds repeatable evaluation gates before a model/provider/prompt combination can be activated for learners or staff.
Pedagogical usefulness
Production test: does the feature guide understanding rather than merely reveal answers?
Safety / integrity
Production test: does the feature respect safeguarding and academic-integrity policy?
Privacy behaviour
Production test: does it avoid unnecessary personal-data exposure and cross-tenant leakage?
Percentages are illustrative UI values, not real model test results.