Inside the clustering threshold for random linear equations