currently, $\mu_{s,d} = \alpha_s + \betas + \beta{s,d}$, where $\alpha_s$ is passed in as data from the prior. Probably better to remove $\betas$ whole cloth, since that's basically captured by $\beta{s,d} | d=D$ and can cause the forecast to jump around when given few polls (easier for the model to adjust $\betas$ than $\beta{s,D}$
currently, $\mu_{s,d} = \alpha_s + \betas + \beta{s,d}$, where $\alpha_s$ is passed in as data from the prior. Probably better to remove $\betas$ whole cloth, since that's basically captured by $\beta{s,d} | d=D$ and can cause the forecast to jump around when given few polls (easier for the model to adjust $\betas$ than $\beta{s,D}$