While passing through the deep mantle, seismic waves that graze the core-mantle boundary interface encounter anomalously soft regions that may be attributed to the presence of partial melt (although this interpretation is debated). Such a partial melt would have numerous dynamical and chemical implications. But how would such a melt form? What is its properties? Common sense, and modeling of partial melting, shows that such a melt, if it were to exist, must be more dense than the solid silicate phases in the surrounding mantle, and it ought to segregate from the crystalline solids via porous Darcy flow. This is not a problem in terms of material physics, since many materials may produce more dense melts at high pressures. One dynamically troubling aspect of this result is that such a melt should accumulate continually over time as fertile material passes through the boundary layer, unless segregation is very slow. When the Earth was younger, and hotter, more melt would have been present. Incompatible elements might accumulate in such a partial melt, perhaps adding excess heat production to the melt piles at the base of the mantle. A regulative mechanism that buffers the maximum amount of melt would be necessary if the pore spaces are well-connected and the melt's viscosity is very small, as expected. Heat producing elements concentrated at the CMB will also reduce the amount of heat flow from the core, which has implications for the energetics of the dynamo process. It is also possible that these melt pockets could become rather hot over time, perhaps expanding enough to rise as diapirs into the overlying mantle and eventually freezing, thus carrying a good amount of latent and radiogenic heat away from the CMB and into the mid-mantle. Another interesting possibility arises as well: one wonders what would happen to melts percolating downward and reaching the core-mantle interface. They should be squeezed out and form a thin 100% silicate melt layer at the top of the core. Having a small viscosity would ensure that this material should spread out horizontally along the top of the core, and it could form large inverted magma lakes in regions where the core-mantle boundary has a topographic high. Over time, as this molten material accumulates, it will form a global layer of silicate melt that might then be able to partly rise and freeze in cool regions of the lowermost mantle, and would otherwise form a thin compaction boundary layer. Also, if thermal gradients become high, as would occur beneath a slab impinging on the lower boundary, thermal convection could be set up within the melt layer itself, perhaps leading to partial freezing of the upper portions of the silicate melt layer. Silicate solids that freeze out will then be driven upward by virtue of their smaller density and compact onto the base of the solid mantle. Being furthest from the eutectic composition, these frozen out bits would also have a higher effective melting temperature than the bulk melt composition, and could form a more refractory solid layer of "sediment" above the silicate melt layer. There are many other things to think about: chemical interaction between the core and silicate melts, effect on heat transport from the core to the mantle, temporal evolution of the process from Earth formation to present, etc.. We've only begun scratching the surface. There are some very promising avenues, however. If seismologists can provide a really good constraint on the size and shape of thes ULVZ regions, then we'll know for sure if a partial melting process is responsible. This is because partial melting and segregation produces mesa-shaped regions that cannot be formed by any other known dynamic process, such as chemical variations within the solid, etc.. Each type of anomaly produces very distinct shapes. A flat top is very strong evidence for partial melting because only segregation can act to produce such a topology. Also, the thickness of ULVZ and magnitude of the seismic wavespeed anomaly can help to constrain the relevant parameters.