| """Plain-Gymnasium ad-creative environment — a single-step CONTEXTUAL BANDIT. | |
| The agent reads a persona feature vector, picks a creative brief from fixed menus, and is graded | |
| against a rule-derived answer key. One action = one ad = one reward; there is no state transition, | |
| so this is a contextual bandit, not an MDP (do not call it an "RL environment"). | |
| """ | |
| __version__ = "0.2.0" | |