The acquisition policies use Bayesian updates with \( \mu_i = \frac{\alpha_i}{\alpha_i + \beta_i} \), sampling \( \theta_i \sim \text{Beta}(\alpha_i, \beta_i) \) for Thompson Sampling, scoring \( \mu_i + c \sqrt{\frac{\ln N}{n_i}} \) for Upper Confidence Bound (UCB), and selecting \( \arg\max \mu_i \) with probability \( 1 - \epsilon \) for epsilon-greedy.