Coffee at Starbucks. It's advertised in three sizes. What wasn't obvious to me is that the sizes actually contain different concentrations of coffee. A tall (12 oz.) coffee has a shot of coffee, as does a grande (16 oz.), but the venti (20 oz.) coffee has two coffee-shots. So, depending on which size you get, you get volume-to-coffee ratios of 12 oz./shot, 16 oz./shot and 10 oz/shot.
Now of course there's nothing inherent in the taste of lattemedium that should make it best served in medium-sized portions. The form of the lattesmall does not come equipped with an intrinsic volume-scale. In fact, you should be able to buy each of coffeesmall, coffeemedium and coffeelarge in sizes small, medium and large as desired. Think of it: you really like the especially mild and milky flavor of lattemedium, sufficiently in fact to to want it in the large size. A largelattemedium, so to speak. For when mediumlattemedium just isn't enough. Or maybe the robust heartiness of largemochalarge is appealing, but who wants that much coffee? Is not the mediummochalarge a worthy alternative? Coffee is not a univariate entity - no, it is inherently a matrix-type proposition. It seems to me Starbucks underserves us by offering us only the diagonal coffee-elements, and compounds the error by identifying the elements, as if a lattemedium were merely a jumbo version of the lattesmall.
I have great trouble convincing people of the essential awesomeness of the proposed scheme; somehow people - even those fluent in the half-syrup-half-caramel-one-percent-macchiato-with-cinnamon lingo - seem to find this a bit complicated. Me, I think a world with grid-valued coffee types, in which people had to think with 3x3 matrices every time they ordered a cuppa, would be a happier place in all of the best ways.