Comment by soltanov

This is not only a fairness problem. It is an agent-memory problem: a system can mistake its own early choices for evidence.