Pursuit
Making espresso
Espresso is a process-control problem you can run twice before breakfast. Everything that matters happens inside an opaque basket in about twenty-five seconds, and you never get to watch it.
What it is
Water at pressure through a compacted bed of ground coffee. That is the whole thing. What makes it hard is that the bed is not uniform, the grind is a distribution rather than a number, and the water will always prefer whichever path through the puck offers least resistance. When the shot runs fast and tastes thin, the fault is almost never where it feels like it is.
So you work by proxy. Dose in, liquid out, time on the clock, and a grinder setting that has to be re-found every time the beans change, and sometimes when they have not. Each of those is a measurement of something adjacent to the thing you actually care about, which is how evenly the water met the coffee.
Why it holds me
The feedback is immediate and completely unsentimental. A shot is on the counter in half a minute, it is either better than yesterday or it is not, and no amount of care in the preparation earns any credit at the tasting. I get very few loops that tight anywhere else — most of my professional work reports back in months, through a haze of other variables, if it reports back at all.
It is also a genuine humility exercise. I have made an identical shot two mornings running and had them taste different, and the honest explanation is usually that they were not identical and I could not see the difference. Holding that thought without abandoning measurement altogether is the actual discipline.
What it keeps teaching me
Mostly that the variable you can measure and the variable that matters are rarely the same one, and that the gap between them is where the skill lives. Shot time is easy to read and only loosely related to extraction. Grind size is the thing doing the work and cannot be read at all. It is very tempting to optimise the readable number and call it control.
The other lesson is about changing one thing at a time, which everyone knows and nobody does. When a shot goes wrong the urge is to adjust the grind, the dose, and the tamp at once and hope. It usually works, and you learn nothing, and the next bag starts you from zero again.
The point
What it taught me that transfers
- A number you can read is not the same as the thing you care about. Most measurement in learning design is a proxy, and treating the proxy as the goal is how a programme starts optimising for the wrong outcome.
- A fast feedback loop is worth more than a precise one. Twenty rough shots teach more than two carefully instrumented ones, for the same reason a rough prototype in front of a user beats a specification.
- When you change three things at once and it works, you have bought today at the cost of every future day. The discipline is not caring about the outcome enough to leave it broken while you isolate the cause.
Beginning
Where to start if you are curious
Buy the grinder before the machine. This is the one piece of advice in the whole field that everybody who has done it agrees on, and it is the one every beginner reverses, mine included.
Then use scales and keep a note of what you changed. Not for the data, which will be thin — for the habit of altering one variable and waiting to find out what it did.