02CORTEX–LOOP LAB
Follow a complete human–AI learning loop: a query, a human response, a cognitive teacher, and an independently audited update.
Sample from the candidate pool using skill gaps and human cost. Log every selection probability.
The deployed student is compared here. Static and active-only conditions freeze parameters and reserve new data for offline training. These are not measured RELAY curves.
Low quality, excessive fatigue or sentinel degradation triggers rollback; insufficient evidence triggers caching. Parameter changes can make full-loop performance no better than the static baseline.