“We should write a paper on that’
— Christopher Hendon, 2016
CAMERON ET AL, 2020
In the beginning …
Cue interior setting sometime in January 2016. I’m in a food court eating lunch, and Chris sends me a photo of a glass of rosé.
‘I’m drinking rosé’ he says.
‘Cool man!’ I reply. ‘I lowered the pressure on my machine today. Tastes amazing.’
‘Wow,’ Chris says. ‘We should write a paper on that.’
And that’s how this all started.
THIS LOW PRESSURE LARK (fig. 1)
Just drop the needle.
As far as an origin story goes, that’s pretty much it. Four years later, thousands of shots pulled, research and writing from six scientists across five countries, and some funky math matched to an experimental model, Cell Press’s new journal, Matter, published our paper in January 2020.
“Systematically Improving Espresso: Insights from Mathematical Modeling and Experiment” is a beast of a paper. But at the start, we were simply writing a paper focusing on the benefits of lowering your pump pressure. And within the first data set, we discovered how much better these shots were—we achieved higher extraction yields at 6 bars, and we did this consistently across multiple samples. At 9 bars, there was significant—and major—shot to shot variation across all samples.
So just lower your pump pressure, already, ok?.
THE VOLCANO OF REPRODUCIBILITY (fig. 2)
From variable to predictable.
Our next finding was so counter-intuitive I repeated the entire damn experiment to confirm the results, as Chris was sure I’d messed something up. Mapping out extraction yields we discovered a “volcano” effect—from a very fine setting, as I ground coarser (but kept the other variables constant) extraction yields went up. They would then plateau at a point, and gradually decline. On paper, the results looked just like an upside-down ‘V’, or in other words, a volcano.
We had long believed that the finer you ground coffee, the more surface area you’d expose, resulting in a higher EY. Or simplified: more finer = more surface area = higher EY. Only we were grinding coarser and getting a higher EY. We were a little weirded out by this.
What was really interesting though, was the EY on the right side of the peak (or the coarser grind settings relative to the peak) had a lower variability shot to shot than those at the finer settings to the left of the peak. Not only was grinding coarser raising our EY, but it was also consistently—and repeatably—higher than the finer settings. At low pressure and coarse grind settings, we were making reproducible coffee at a statistically significant level.
The thing about this consistent and repeatable side of the peak though was the shot times. These were racing out—10-16 second shots, at a wildly coarser grind setting than I’d ever used on an EK. Yet I was able to achieve the same extraction yield as those finer—but more variable—grind settings, at both 6 and 9 bars of pressure.
Dusting our hands off thinking we were done, we were incredibly excited about the findings, and believed we had a damn good paper on our hands. We even had a draft lined up ready for peer-review.
Then Chris met some mathematicians and things went a little crazy.
THE MATHEMATICAL MODEL (fig. 3)
No matchy, matchy.
William Lee and Jamie Foster are mathematicians based at the time at the University of Portsmouth in the UK. The enthusiastic reader may have already noticed the name William Lee, as he’d already made the limelight earlier by describing how layers formed in Guinness. They had previously published a mathematical model for filter coffee and were now working on a second model for espresso. Having been introduced to Chris, it was clear with our experimental data we’d have an even better paper when matched to their new espresso model. As Chris said at the time, it would turn this paper into a powerhouse.
Things immediately went a little awry, though. The mathematical model William and Jamie built did not completely match our experimental data—when it should have. Sure, for the majority of the 9 bar data it matched, and for most of the 6 bar data it matched. But at the point where our experimental data showed a drop in EY when the grind started getting finer, the mathematical model showed EY going up.
The mathematical model based results calculated on the population of fines and boulders, and the knowledge grinding finer changes the size of the boulders and the population of fines. The model was showed if we ground finer from a “middle” peak grind setting at 1.7, EY should have gone up fairly linearly. This meant the slope of the line from 1.7 to 1.1 should have continued a consistent upward trend. Yet with our data, it didn’t. That line sank. In other words, the ‘more finer/more tasty’ hypothesis was bunk.
We did particle size analysis, so we knew the grinder was behaving. With a San Remo Opera, gear pumps, PuqPress, and Acaia Scales, we also knew we had our variables under control. And repeated experiments in different cafe settings continued to replicate our findings—we achieved a volcano effect as we went up and down in grind size. Perplexed and confused, going back to the mathematicians, William and Jamie simply pointed out what we were seeing was inefficient extraction from the entire bed. We were over-extracting in some areas, under-extracting in others, with some areas just completely untouched. Overall, at a certain point experimentally, there was a lower mass of coffee in the cup compared to the mathematical model.
“I do not think you are measuring what you think you are measuring.”
All our experimental samples were measured using a coffee refractometer. This device measures an average extraction yield of all the extractions the tens of millions of coffee granules could have given us. Technically, it’s not even an average. All it does is estimate the dissolved mass with some degree of certainty, with no definitive statement about where that mass came from, or what that mass is. It’s possible at a fine grind setting some of our coffee would have extracted at 28%—and at the 1.3 grind setting, the mathematical model told us we should have. We only measured it at 19% EY because the rest of the measurements within that extraction had dragged the average down.
From a physics point of view, we were only effectively extracting from 15-16g of our 20g dose. But the measurements we were taking from the refractometer assumed the original 20g dose—again, dragging the average down. We had a ‘chemistry vs physics’ problem—the maths said we should extract x amount based on the chemistry occurring (and it is!). But the experimental data showed we wouldn’t. Our coffee was “choking”—there was no physical pathway for the fully extracted coffee to make it through the puck into the cup. What eventually did make it through to the cup was channelling—with lazy water finding the most accessible point (or multiple points) through the puck. What we had measured with the refractometer were all the over- and under-extracted portions of the 20g dose, including parts that hadn’t even extracted at all.
This is a fundamental problem with espresso: the actual coffee puck itself needs to work as a restrictor to promote extraction. At a certain point, it restricts the flow of water through—and the flow of espresso out—too much. The shot either chokes, channels, or both, and you end up with a highly variable coffee you cannot replicate consistently. At least, that is, when using the traditional parameters of 9 bars, large doses, and 25-30 second extraction times.
TURBO SHOTS (fig. 4)
Just hit the accelerator.
As we put these findings together, one of the final steps for us was to lower the dose. We weren’t using all of it anyway, so why not just use less? In doing this, we could achieve a bunch of things: less coffee would mean less resistance, so we could create more resistance by grinding finer—but not so fine as to reach the choking point. This finer setting at a lower dose would still be a coarser setting than at a higher dose, so if shot time were kept constant, the line on the volcano graph would move up. If we kept the beverage weight the same, we could target the same EY on the consistent right side of the volcano.
However, these shots come out fast. But sweet. And with less strength—as measured by a refractometer.
There are plenty of articles out there that’ll tell you the difference between strength and extraction yield. While there’s a relationship between the two, they do not mean the same thing. For me, I’ve always looked at strength as just that—how strong is my coffee, measured by my me as I taste it. Or, if measured by a refractometer, as ‘total dissolved solids’—basically how much ground coffee have I dissolved into the cup. Extraction yield is an equation expressed here as a percentage of the ground coffee initially used, interpreted as how efficient I was at getting the ground coffee into the cup.
It stands to reason then if we use less coffee (15g vs 20g) keep the same brew weight (40g) and achieve the same EY; mathematically we’d have less dissolved coffee in the smaller 15g coffee than the 20g coffee.
But this is as measured by a refractometer. And you do not need a refractometer to figure out what tastes good.
“In the end, the tasty point was inside all of us.”
In the paper—by absolute design—we never objectively stated whether you could make better-tasting coffee by either lowering your pressure, grinding coarser, or lowering your dose. We based our objective statements on measurements made by the refractometer, all of which were in relation to what we coined ‘the barista defined tasty point’. What we did objectively show, was how you could improve on this tasty point through raising extraction yields and vastly improving consistency. You, the barista, defined that tasty point, and we hoped we had illustrated a range of strategies available to reach that point, and reproduce it every time.
You could keep your dose at 20g, beverage weight at 40g, pump pressure at 9 bars, grind fine, and hit a 30-second extraction. And if that coffee tasted good to you—great! You’re just not going to make it taste great every time—it just won’t happen. This is traditional espresso, and it’s where we’ve been stuck for a long time.
You could keep your dose at 20g, beverage weight at 40g, pump pressure at 9 bars, grind coarser, hit a 20-second extraction time (or even 10 seconds!) and if that coffee tasted good to you—great! Even better, you’ll achieve this great taste more consistently than at the finer grind.
You could then maybe decide to lower your pump pressure. Same 20g dose, same 40g beverage weight, fine grind, 30-second extraction time—and if it tastes good to you, great! You’ll achieve this tasty coffee more consistently than you could at 9 bars, and even better, you’ll achieve higher extraction yields too.
You could then keep all variables the same but grind coarser. You’re now hitting higher extraction yields and consistency than with any of the previous three options.
You could then lower your dose using a 15g basket, keep your beverage weight at 40g, grind fine, and see how you go with a traditional 30-second extraction. Taste good? Again, great! You’re almost at the peak efficiency and reproducibility sweet spot. (And just to be clear—we used a 15g basket and dosed 15g into it. We did not use a 20g basket and dose 15g. I’d highly recommend using a basket suited to the dose).
Now just hit the accelerator, grind coarse, and see how it tastes. If you like the flavour, then compared to any of the previous strategies you’re now going to hit higher extractions, and repeat that flavour shot after shot—it’s just science.
THE ‘PERFECT ESPRESSO’
No one makes espresso anymore.
“As historically defined by the Specialty Coffee Association, an espresso is a 25–35 mL (ca. 20–30 g) beverage prepared from 7–9 g of ground coffee made with water heated to 92C–95C, forced through the granular bed under 9–10 bar of static water pressure and a total flow time of 20–30 s.” Cameron, Michael I., et al. “Systematically improving espresso: insights from mathematical modeling and experiment.” Matter (2020), pp. 631–48.
No one makes espresso anymore. Not like the SCA define anyways. We don’t measure our beverage weight in millilitres, we don’t use 7-9g of coffee, and it seems a lot of people aren’t brewing at 9-10 bars anymore. And for a turbo shot, 20 seconds is the max time you’d aim for.
In the paper, I defined the initial tasty point. Through our experimental data and the mathematical model, we were able to apply the results to a real-world setting and consistently replicate that tasty point. You don’t have to lower your pump pressure, grind coarse, reduce your beverage weight, use less coffee, measure with a refractometer (or combine two shots!)—you don’t have to do any of that. You can combine as many of these strategies as you want or none at all. You define the tasty point, and if you do combine these strategies, then we’ll show you how to hit it every time.
I never said this was the perfect espresso—the media did, and it made for a damn good hook. In interview after interview I was asked plenty of times “if I made the best coffee in the world” and every time I politely side-stepped the answer. But for the last four years, this is how I’ve been making coffee—low pressure, 15g doses, and fast shot times—and to me, it is perfect. Sweet, balanced, and delicious—that’s my definition of espresso.