The optimality gap in preference-regulated agent policies vanishes when the optimal policy can be represented by available preference-conditioned policies

We show that the gap vanishes when the optimal policy can be represented by the available preference-conditioned policies.

http://arxiv.org/abs/2608.27072v1