Monthly Archives: September 2011

On the discrete nature of coffee bean blends

In which I wax scientific while reading Newton’s biography:

Blending coffee beans does not a homogeneous mixture make. Unlike much cooking and mixing, such as the incorporation of water into flour, the small dosages of coffee used when making espresso actually serve to highlight discrete nature of the mixture. Remember that “discrete” is the opposite of continuous. When you graph it, it moves in stair-steps instead of a smooth curve. The probability of the mixture being dramatically “heavy” on a certain variety of bean increases as the sample size decreases.

Say that your “espresso blend” is a combination of 4 different bean types, 25% of each: a deep dark Sumatra, a couple of medium-bodied central American batches, and one bright and floral African. Assume they are mixed as evenly as possible, but that nothing mechanical enforces this ratio. Say that a shot contains the grounds produced from 32 beans. Ideally, there will be 8 beans of each type in that shot (8+8+8+8=32).

What is the chance that the mixture will be off a bit? Ex:(8+6+10+8) Pretty high, but no big deal. What is the chance that it will be really screwed up? Ex:(15+0+1+16) Pretty unlikely but it’s got to happen every once in a while. Now say your blend is more complicated: (8+8+8+4+4). Now the chance that one of your types will be missing completely skyrockets. You can’t make subtle changes and have them “take” in the real world. The discrete nature of the mixture resists that subtlety. Sure, it will be there across the mean of 1000 doses, but the distribution will be all over the place. This kind of variation can be frequently blamed for inconsistently tasting shots, I am certain.

This is a quality control challenge that, as far as I can tell, nobody is dealing with directly. It seems to me that there are some relatively simple mechanical solutions that could be devised to keep the mixture stable across dosages. An array of single-bean dispensing hoppers feeding into a buffer would be the most obvious solution. This would be versus a traditional single conical hopper feeding directly into the grinder. Is anyone building these? Why not?

Photo credit