I love coffee. To approximate Mark Twain, quitting is easy, I’ve done it hundreds of times. I feel better, overall, when I don’t drink coffee regularly. But I love coffee. There is so much variation—the most consistent brew is still different from drink to drink. Coffee is marvelously complex; it tastes different at different temperatures, or at different times of day.

With all this variety, you might expect that some cups are better than others. To an engineer (ahem) this implies that there is an optimum. I’m fairly sure the best cup of coffee is stored next to the Golden Fleece. But maybe, just maybe, there is a very good cup of coffee I can make. In an upcoming series of postings I will write about my ongoing effort to brew, and define, the perfect cup. The search continues.

Brewing Methods

There are three classes of coffee brewing methodology

1. Hot diffusion extraction, including drip, French press, percolator, and vacuum
2. Cold diffusion extraction, also called simply cold extraction
3. Forced extraction, in particular espresso

I love espresso, but I’m not interested in spending the money required to brew my own. The word expresso is a linguistic abomination, as loathsome and vile as “12 items or less”.

Hot and cold extractions are both accessible to me, and each offers a wealth of experimental opportunity. I’ll begin my discussion with hot extraction.

Hot Diffusion Extraction

A good cup of drip coffee is hard to beat. I love Starbucks’ drip because it is very rich, very aromatic, and very complex. I know that many loathe Starbucks, and in the name of political correctness, I’ll be the first to state that these horrible people are wrong. Terribly wrong. Even so, it is safest to disagree from a distance, rather than risk offending a disenchanted ex-Starbucks drinker and his certainty that “multinational corporations” are secretly making his life suck while he types on his or her pretty little Mac.

I assert that there are five principal variables in the diffusion extraction process:

• Coffee-to-water ratio (r)
• The source and roast of the beans (b)
• Distribution of coffee bean particle sizes (F(d)). Fineness at first glance, grind uniformity upon consideration.
• Period of time in which the coffee beans are in contact with the water (t)
• Temperature of the water (T)

One might suggest that some specific set of these variables would produce a truly excellent, repeatable, cup of coffee. If Q is the quality function, then we seek the r, b, F(d), t, and T such that we (nearly) maximize Q. So, then, this is science. In practice there are other variables, not least of which are the temperature at the time of drinking, the source of the water, and the vessel from which the coffee is consumed. I’m ignoring those. Let us explore the support of each of the chosen parameters.

Coffee-to-Water Ratio r

I have never been able to measure coffee repeatably using the oft-advised “tablespoons”. I find the grain size of whole beans to be too close to the size of the spoon—leveling is impossible. Picture (below left) is a level tablespoon, and (below right) is a very different, but also reasonable interpretation of tablespoon.

Starbucks provides an interpretation of a tablespoon—namely 5 g. For reference, they are recommending a coffee/water (C/W) ratio of 0.055 g/ml.

I have resorted to weighing for the beans, to provide consistency. Volume is acceptable for water, since that is not subject to nearly the same variability.

A reasonable starting range for coffee-to-water ratio might be in the neighborhood of 0.047 g/ml to 0.063 g/ml for drip (according to Black Bear). Allowing some margin for exploration, I assert that we should explore coffee to water ratios between 0.035 and 0.075 g/ml—which puts Starbucks 0.055 g/ml right in the middle.

r in (0.035 g/ml, 0.075 g/ml)

Source and Roast of Beans b

The choice of beans is important, I think. Perhaps the roast is even more important. Starbucks appears to prefer to burn their beans, and in fact a cursory glance at Trader Joe’s coffee shelf suggests that most beans sold are darkly roasted. Watch the shelves, you probably won’t see more than one choice that in “light roast” and maybe only one in “medium roast”, the rest will be French roast, city roast, or dark roast. As far as I can tell the last three are all equivalent.

Nevermind, I will eventually test several varieties. Because I don’t drink that much coffee, the bean for early experiments must be consistently available. Because I’m cheap, it must also be affordable. Not all the way to Folgers, but there is no chance I’ll regularly buy Starbucks beans. My main choice, for these experiments, is Costco’s house brand.

Particle Size Distribution F(d)

Coffee grind must be one of the most ill-described nonsensical terms in the wide field of cooking. Particle size ranges from coarse grind (perc) to a nearly molecular dust (Turkish). And “drip” means something, but nobody quantifies it. In my estimation, about half the grinder reviews I can find on the net base their rating on the noise, appearance, or the speed with which the shippers delivered their grinder(!?). I have yet to find a single review by any organization or individual which examines the grind in any repeatable way. I’d do it myself, but I lack two things: a graduated sieve set and a plausibly decent grinder. Donations welcome.

This is, by far, the weakest controlled variable in my process. I can only hope that the alarmingly wide range of particle sizes I get makes the rest of the processes insensitive to the grind. Note that I have two grinders, a whirly-blade grinder and a heavily modified lump-o’-junk burr grinder based on a portable which I reviewed on Amazon, though the product appears not to be sold anymore.

The whirly blade grinder produces grind distribution something like the following.

Extraction Period t

This is very easy to control…sort of. In a typical drip system the extraction period is equal to the time it takes to run the water through. This means that the more coffee you are brewing, the longer the extraction time. I didn’t want that to be a limit, so for the purposes of experiment I use a French press. With a French press the extraction time is easily controlled with a countdown timer—depress the plunger when the timer dings. I presume my cholesterol is suffering accordingly.

The admissible range of times could reasonably go from zero to approximately 5 minutes. By zero, I mean about five seconds; pour the water in the press pot, then depress the plunger.

t = (0, 5 minutes)

Temperature T

I have read recommendations for temperature ranging from about 195 F to 205 F. I live at approximately 1 mile altitude, and therefore 205 F is the hottest I can get outside the pressure cooker. Since I’ve never seen a justification for the indicated temperatures, I provide a little margin.

T = (185 F, ~205 F)

The Quality Function

My objective is to optimize the quality, or enjoyment, of the cup of coffee. This is a highly subjective measure, and proves difficult even in that context. My tasting methodology is to spend a while smelling the hot (but not undrinkably hot) cup of coffee. I try to smell different elements. Then I take a large sip, and swallow it. I know professional tasters spit, but I’m interested only in optimizations that reflect the way I drink coffee. Practically, I find that swallowing allows me to better characterize the bitterness and the mouth feel of the beverage.

In addition to being subjective, a taster is variable. I could, in theory, drink the same cup of coffee on three different days and get three different answers. This will prove to be the source of the greatest uncertainty in my estimation problem.

I briefly toyed with the idea of expressing a vector quality function, comprised of elements like “mouth feel”, “aroma”, “flavor”, and “strength”. However, I decided that using those attributes would bias my assessment, since I might be carrying an unconscious bias that weak coffee is bad coffee. My rating scale is 1 to 5, where 1 is so bad I might hesitate before having another cup (I’d feel gouged at 30 cents), 3 is a passable daily cup, but I’d feel gouged at more than one dollar, and 5 is very good, where I’d feel I got a deal at $2.50.

Experimental Space

Suppose all our parameters vary over their ranges fairly coarsely:

• Ratio r in (0.035, 0.045, 0.055, 0.065, 0.075) g/ml
• Beans b in (Costco)
• Particle size distribution F in whatever my lousy equipment provides
• Extraction time t in (10, 23, 55, 128, 300) seconds—log spaced
• Temperature T in (185, 190, 195, 200, 205) Fahrenheit

To do even a single experiment for each combination would require 5 × 5 × 5 = 125 experiments. Assuming a pot a day, this is almost half a year from a useful result. This radical shortcoming leads to the next topic: Design of Experiments.