# Question about EKF SLAM (w/ unknown correspondence) update step

In Probabilistic Robotics, page 322, the EKF SLAM update step is shown below.

My question is, why is for every observed feature, $$\bar{\mu}_t,\bar{\Sigma}_t$$ are overwritten in lines 23,24 by the most likely observation in every new loop? It just seems to me, if there are N observed features, then 0th to N-1 th features aren't even accounted for in the final $$\mu_t, \Sigma_t$$ in line 26 and 27, only the Nth feature and its ML correspondence is updating $$\mu_t, \Sigma_t$$. I think I might be wrong about my understanding, but I don't see how the subsequent loops account for the previous loops' ML correspondences.