If fe != 0, is it correctly included in predict() and group_effects() for non-Rubin models?

[wwiecek]

In group_effects.R I have

      # If dealing with a meta-regression model, we automatically add effect of covariates
      # unless user requests random_only
      if(bg$model == "rubin" && !random_only && !is.null(bg$covariates)) {
        m_fe <- fixed_effects(bg) %*% t(bg$inputs$X)
        m <- m + m_fe
      }

This looks wrong to me!