Since 2004, the Sumatra subduction zone has been one of the most active on the planet, generating 30 moderate to great earthquakes (including four Mw>8) that produced complex patterns of postseismic deformation. All these events were recorded by the Sumatran GPS Array (SuGAr). We have been working on self-consistently separating and modeling the different signals present in the GPS time series. We have modeled the moderate earthquakes for the first time, and remodeled the larger events with a careful eye to resolution, realistic slab geometries, and the most precise estimates of coseismic offset. Wherever possible, we have incorporated coral paleogeodetic data and InSAR data, successfully unwrapping the InSAR scenes even over the small outer-arc islands. Here, we present our geodetic results in the context of (a) illustrating our best attempt to piece together the patchwork quilt of all slip on the megathrust since 2004 and (b) our thoughts on how these results fit into long-term behavior along the subduction zone as inferred from paleogeodetic observations. We are particularly interested in understanding behaviors that appear to be relatively persistent over time, and those that show greater variability. For example, coral data (e.g., Meltzner et al., 2015) show that great earthquakes along one section of the megathrust are persistently bounded along-strike by structural features within the downgoing plate, including subducted fracture zones. Our data show varied types of slip near these boundaries and near areas that slipped little during great earthquakes. To illustrate, along the section of the megathrust beneath the Banyak Islands (which was a low-slip patch in the middle of the 2005 Mw 8.6 Nias earthquake rupture), we have modeled a moderate (Mw 7.8) earthquake rupture, a 15-year slow-slip-event (Tsang et al., 2015a), and large amounts of afterslip. These varied behaviors likely result from structural and/or geometrical complexities of the megathrust at this location. In contrast with the idea of persistent behavior of great earthquakes, the Mentawai patch (which is forecast from paleogeodetic data to produce a giant earthquake in the near future) has so far been breaking only in small sections. While a great earthquake may still occur in the coming decades, the slip released during the 2007, 2008, and 2010 events has been piecemeal.