Subduction, due to density inversion by top-down cooling, drives plate tectonics and upper-mantle circulation. It organizes plate motions and lithospheric stress, which controls plate boundaries and volcanic chains. Rollback of hinges is the key to plate kinematics. Where a fore-arc basin is present on an overriding plate, hinge and non-shortening plate front track together, and velocity of advance equals velocity of rollback. Convergence velocity commonly also equals rollback velocity (as, California/Farallon for 100 million years) but can be greater. Backarc basins open behind island arcs migrating with hinges. Broadside-sinking slabs push upper mantle back under incoming plates and force rapid spreading in the Pacific (which is being narrowed by rollback), and pull overriding plates forward with retreating hinges shaped to them. Slabs settle on the 660-km discontinuity; if in fact they do penetrate that, they get no deeper than about 1000 km; contrary claims reflect tomographic artifacts. Plates advance over slabs and mantle displaced rearward by them, and ridges spread where advancing plates pull away. Ridges also migrate with respect to underlying mantle, producing geophysical and bathymetric asymmetry, and tap fresh asthenosphere into which slabs are recycled. Sluggish deep-mantle circulation is decoupled from rapid upper-mantle circulation, so plate motions can be referenced to semistable lower mantle. Hotspot and no-net-rotation frames are invalid because, among many flaws, they minimize motions of hinges and ridges. If Antarctica, almost ringed by departing ridges and varying little in Cenozoic paleomagnetic position, is stationary, then, allowing for backarc spreading, hinges roll back, ridges migrate, and the quasi-absolute directions and velocities of plate rotations accord with subduction, including sliding and crowding of oceanic lithosphere toward free edges, as the dominant drive. Northern Eurasia also is almost stationary, Africa rotates very slowly counterclockwise toward Zagros, Pacific plate races toward surface-exit subduction systems, etc. Fast plates have attached slabs and well-developed asthenosphere, and lack cratonic anchors. In accord with paleomagnetic data, but contrary to ad hoc hotspot-frame assumptions, there is little or no true polar wander.