Stream-bed surfaces are commonly armored. How does this work? Does armor form at low flows and small transport rates and does it disappear during floods? Or is armor persistent over all flows? Field measurements during active transport are almost nonexistent. Lab observations give conflicting answers. In sediment-feed flumes, armor layers form at low transport rates and vanish at large transport rates. In sediment-recirculating flumes, armor layers are persistent over all transport rates. There is persistent confusion over which flume system represents the field and the simulation of armor layers brings the issue to a head. The physical interaction between flow, transport, and bed is identical in both cases. It is the difference in boundary conditions that provokes the difference in armor behavior. In a sediment-feed flume, a feed of constant grain size is typically used for all feed rates. In contrast, the sediment supply in a sediment-recirculating flume coarsens with increasing transport rate (as is typically observed in rivers). These differences are sufficient to qualitatively explain the different armoring behavior of the two flume systems. A quantitative explanation requires a transport model that is referenced to the grains immediately available for transport on the bed surface, thereby allowing computation of transient conditions such as the development of armoring. We have developed a surface-based mixed-size transport model from a set of coupled flow/transport/bed surface measurements. The model is used to develop relations among flow rate, sediment supply, and armor composition. These relations provide a general description of armoring, including the special cases of sediment-feed and sediment-recirculating flumes, as well as natural rivers, which fall somewhere between.