Recent seismological observations suggest the existence of a ≈150-km-thick density-stratified layer with a P-wave velocity gradient that differs slightly from PREM. Such a structure can only be caused by a compositional gradient, effects of a slurry or temperature being too small and probably the wrong sign. We propose a stably stratified, variable concentration layer on the liquidus. Heat is transported by conduction down the liquidus while the light and heavy components migrate through the layer by a process akin to zone refining, similar to the one originally proposed by Braginsky. The layer remains static in a frame of reference moving upwards with the expanding inner core boundary. We determine the gradient using estimates of co, the concentration in the main body of the outer core, and cb, the concentration of the liquid at the inner core boundary. We determine the depression of the melting point and concentrations using ideal solution theory and seismologically determined density jumps at the inner core boundary. We suppose that co determines Δρmod, the jump from normal mode eigenfrequencies that have long resolution lengths straddling the entire layer, and that cb determines Δρbod, the jump determined from body waves, which have fine resolution. A simple calculation then yields the seismic, temperature, and concentration profiles within the layer. Comparison with the distance to the C-cusp of PKP and normal mode eigenfrequencies constrain the model. We explore a wide range of possible input parameters; many fail to predict sensible seismic properties and heat fluxes. A model with Δρmod= 0.8 gm cc−1, Δρbod= 0.6 gm cc−1, and layer thickness 200 km is consistent with the seismic observations and can power the geodynamo with a reasonable inner core heat flux of ≈2 TW and nominal inner core age of ≈1 Ga. It is quite remarkable and encouraging that a model based on direct seismic observations and simple chemistry can predict heat fluxes that are comparable with those derived from recent core thermal history calculations. The model also provides plausible explanations of the observed seismic layer and accounts for the discrepancy between estimates of the inner core density jumps derived from body waves and normal modes.