Most online social networking platforms enable users to define a set of information sources they are interested in. In turn, they are fed with content that stems, at least in part, from this list of manually selected sources: platforms implement a social filtering mechanism that sometimes relies to a large extent on explicit user preferences. A user's informational landscape is also indirectly embedded into a larger fabric of information flows through their neighbors' neighbors. The myriad of ego-centered source choices thus yields a network of decentralized feed links, whose topology is heterogeneous in two main ways. Firstly, a few users benefit largely from the attention of many, while many users attract little attention. Secondly, and perhaps more importantly, clusters of densely interconnected users emerge. Two main types of approaches aim to empirically characterize the heterogeneous and fragmented informational landscape in which users evolve. On the one hand, macro-level studies focusing on the network structure robustly demonstrate the existence of clusters in many online contexts. On the other hand, we find micro-level endeavors to describe the ego-centered informational landscape accessible to users. In general, these studies aim at characterizing the homophilic and/or myopic behavior from the user perspective. Few studies articulate both perspectives where users' access to information is described in terms of local and global network. Users are plausibly nearsighted regarding the latter, and thus to the compound effect of their own direct filtering preferences and that of their neighbors. We aim at characterizing the variety of user configurations regarding both their immediate filtering behavior and the secondary filtering performed on indirect sources. In particular, we propose a double dichotomy based (i) on direct vs. indirect levels and horizons and (ii) on diverse vs. homogeneous sources. For instance, one's apparently diversified source portfolio may actually lead to quite similar sources; conversely, one's visibly small source portfolio might indirectly lead to diversified sources. To this end, we achieve a dual qualification of the macro- and the micro-levels: fragmented holistic clusters, myopic ego-centered landscapes. We use Twitter as a privileged study field of online filtering, for the user-centric source selection process lies at its core, with the central notion of "followers" and "followees". We thus focus on "follower networks", which reflect users' own filtering decisions and are relatively stable in the short term. We further define perimeters of users interested in certain topics by collecting tweets related to specific hashtags and by selecting a few hundreds of the most active users. This yields a series of topical networks which typically feature a certain number of structural clusters, i.e. groups of followers more densely interconnected with one another than with other users. Leveraging these clusters, we compute for each user the proportion of followees belonging to the same cluster ("confinement at distance 1") and similarly for their followees' followees ("confinement at distance 2"). We arrange these values in a two-dimensional map which translates the above double dichotomy. Our contribution aims at further discussing these arrangements both quantitatively and qualitatively.