Despite people pushing the idea; I doubt that split roast levels are commonly useful. This is based both on experiment and reason. You'll have to take my word on the experiment part; but the reason part is worth considering.

To even consider a split roast, there has to be a flavor or flavors that are best at a light roast level, and another set that are best at a darker roast. But this isn't enough. Suppose the preferred roast light roast levels drop in a convex curve (bowed outward, like the graph of a square root), and the preferred dark roast flavors rise in a convex curve. Then an intermediate roast will have more of both than a mix of a light and dark roast. The only time a dual roast pays is if the flavors are rising or falling in a concave curve (like the graph of a quadratic).

In general, logic dictates that flavors appear and disappear in roasts in a bell shaped curve, starting fast and concave, then slowing down to a gentle convex rise, then over the top to a convex fall, then slowing to a concave fall. In this case, it makes sense to mix very different color roasts, so the flavors are far apart and divided by concave portions of their curve, but not for roast levels that are close, with the curves overlapping in their convex section.

This is easier to see than to describe:



When it comes to Yemen, melanges make even less sense, since they roast so unevenly, that even a single roast is a melange!

swines wrote:Why? Why fast? Why concave? This isn't "logic" - it's pure speculation, unless you have some repeatable method of quantifying the taste versus roast levels. Or, unless you can prove the beans took calculus and believe in quadratic equations as their preferred method of flavor response.

These are not quadratic curves, but end to end sigmoid growth curves. Flavors are produced and destroyed by chemical reactions which start slow, pick up steam (if they are autocatalytic, which Maillard reactions are), then slow down again when they run out of raw materials (try producing "ash flavor" by setting something on fire to see what I mean). One or more such reactions produces the flavor giving the rising side, then one or more destroy it, giving the falling side. My illustration is biased since the plateau can be long or short, and the fall need not be symmetrical with the rise.