A paper on Espresso Extraction

[another_jim]

I mentioned this paper in the sticky thread on dosing. I now have a completely rewritten version that I think is clearer and more thought out. It's still on the heavy and pedantic side, but I hope more readable. I'd appreciate any comments or people giving the shot tuning suggestions a spin.

For those who want it short and sweet, there's a forthcoming, more how-to focused article coming up right here

[LeoZ]

wow. my kind of paper..

a few questions for you.

do you have the list of coffees grouped by molecular weight and flavors? how many coffees did you test and how does it become relevant when blending? will the flavor profile of blends transfer over as scientifically?

next, you are recommending reducing doses from (im guessing here from what most recommend, i dont weigh) a 16-18 gram double to 13grams. wont the shots blond faster, or do you now grind even more fine to compensate?

from what i recall in italy, the shots drip out as opposed to flow, and its rare to see someone tamp with anything but the ones built into the grinders. so, id assume that grinds are really fine and distribution and tamps arent all that precise. i guess with a smaller amount of coffee in there, and a really fine grind, tamping to perfect distribution becomes sort of null. we also have different coffees here, as you mentioned, so doesnt that come into play, or shall we just let our coffee age a bit more, which from what ive seen, takes off that initial bite.

[another_jim]

It's not possible to group coffee by their full chemistry. There's three or four labs, using very expensive equipment, trying to find new compounds in coffee. Most of the new ones (roughly chemicals 200 to 1100) are quite exotic, occur at very low concentrations, and nobody is sure when in the roast they occur. Occasionally they score a home run, and find a really low concentration chemical so odorous that it contributes to the overall flavor.

On the more workaday chemistry.
-- The fruit acids are there from the start: citric, malic (apple), and acetic are the biggies, along with longer carbon chain versions of these. In the roast they both break down, and combine with sugars to form alcohols and esters. This taste family corresponds to the acidity in coffee, and is strongest in high grown, washed arabicas
-- The light Maiilard cmpounds are the "baking" flavors -- they develop in the 300F to 390F range of the roast as amino acids react with sugar -- think of browning bread, meats, nuts, wood smoke, malt, etc. Unsweetened, these are sharp bright bitter flavors like lemon peel, ginger, cinnamon, quinine, unripe nuts. They mellow out in the presence of sweetness, becoming toasty, malty, and nutty. These are the hallmark of Indonesian and Indian coffees, as well as some Brazils and South American ones
-- Caramels you know - they develop from around 385F and up from the sugars in coffee. Since dry processed coffees collect sugars from the cherry, they are sweeter and caramelize better.
-- Distillates are the caramels and Maillard compounds as they start searing -- think of your broiler or grill. Good dark roasting is mostly a question of the beans surviving long enough. The highest grown Indos and Indian coffees are best, as well as Mexican coffees. Caribbean coffees are so low grown, they can be used to develop these flavors at medium roasts.

The work on dosing versus yield presupposes that you are always setting the grinder to get actual espresso shots, not something else: roughly 20 to 40 second, .5 to 1 ounce singles, or 1 to 2 ounce doubles, no channeling, and blonding only at the end. If it's not this, it's not espresso.

[cpl593h]

Jim, I could have missed something in the article, but it seems to me that it would be helpful if you mentioned that generally, solubility increases with degree of roast. See Espresso Coffee section 8.1.2.5; there aren't any solid numbers, because this is sure to vary according to origin factors. It would help to clarify why lighter roasts benefit from lower doses.

Otherwise, this is great. It's really changed my understanding of espresso. I'll keep rereading it, keeping my eyes open for opportunities to give specific feedback.

[another_jim]

Jim, I could have missed something in the article, but it seems to me that it would be helpful if you mentioned that generally, solubility increases with degree of roast. See Espresso Coffee section 8.1.2.5; there aren't any solid numbers, because this is sure to vary according to origin factors. It would help to clarify why lighter roasts benefit from lower doses.

Otherwise, this is great. It's really changed my understanding of espresso. I'll keep rereading it, keeping my eyes open for opportunities to give specific feedback.

I was not able to replicate this. My dark roast solubles yields were no different from the light roast ones. This may change when the coffee stales and the light roast's solubles disappear into thin air.

[cpl593h]

I'm unconvinced that there is no difference between the solubility of a light coffee and a dark coffee. Pyrolysis reactions degrade more and more compounds as the roast progresses, creating more soluble matter. A very general rule for espresso extraction is to increase brew temperature for lighter roasted coffees. Why? To increase the efficiency of soluble extraction to compensate for the lower solubility of lighter roasts.

Perhaps the thinner pucks of the light roasts introduced a higher margin of error, obscuring the differences between light and dark roasts.

[another_jim]

I didn't do a lot of dark roast observations, and none further than a rolling 2nd crack. There are reactions in this range that go in the other direction. Dark caramels and maillard reaction compounds polymerize into more and more insoluble melanoidins. The acids simply vaporize. Also, the pyrolysis compounds you mention are mainly the cellulose reducing and becoming soluble. This may only become a large factor at deeper roast levels.

[cpl593h]

Interesting. The difference in solubility between light and dark might not be that big of an influence as I thought it would be. Nevertheless, the fine grind + low dose/shallow puck formula is maximizing extraction of "balancing" flavor compounds that are present in lighter and darker coffees in similar amounts, but more need to be extracted to compensate for the acidity of lighter roasted coffee.

[LeoZ]

doesnt pyrolysis occur in high (above 800F) temps and in the absence of oxygen? i suppose it could be argued that while roasting there is some sort of boundry layer around the beans that prevents oxygen from getting too close to the surface, but i cant see how, if at all, this small amount of a reaction would affect the outcome of the bean, except at the infamous 3rd crack, which could correlate to carbonization.

[dsc]

Hi Jim

Great work on the paper, really helped me to understand the whole idea of espresso, how it works, brews and what happens with the coffee.

So I guess overdosing is not a good idea right? Cause it's not about packing as much coffee as you can into the basket. Is this the end of the "pucks with shower screen impression" era?

Cheers,
dsc.

[cafeIKE]

doesnt pyrolysis occur in high (above 800F) temps and in the absence of oxygen? i suppose it could be argued that while roasting there is some sort of boundry layer around the beans that prevents oxygen from getting too close to the surface, but i cant see how, if at all, this small amount of a reaction would affect the outcome of the bean, except at the infamous 3rd crack, which could correlate to carbonization.

pyrolysis :
n. Decomposition or transformation of a compound caused by heat.

InterFIRE definition:

Pyrolysis

A chemical process that occurs when heating in the absence of oxygen causes the chemical decomposition of a compound into its component substances. Pyrolysis often precedes combustion and the gases given off by it are the materials that burn in the flame. Pyrolysis takes place just below the char layer of the material (i.e., wood).

This is an overly restrictive definition. The chemical decomposition process takes place with or without oxygen present. I'm not sure what is meant here by "component substances." Once a molecular bond is broken, what remains is a different substance. Butane is not a "component" of octadecane, but a butane ion may be produced when octadecane pyrolyzes. Heat causes large molecules to break into smaller ones as a result of the breaking of bonds between atoms. If oxygen is present, combustion might occur, assuming there is an energy source that will heat the vapors to their ignition temperature. Further, pyrolysis usually takes place in virgin material in front of the char layer, as well as under it. Finally, stating that pyrolysis "often" precedes combustion fails to discuss the state of the fuel. While liquid fuels can vaporize and burn, pyrolysis (almost) always occurs in the combustion of typical solid fuels. The only exceptions are metals and substances like pool chemicals.

Accurate definition

The chemical decomposition of a compound into one or more other substances by heat alone; pyrolysis often precedes combustion.

[another_jim]

Interesting. The difference in solubility between light and dark might not be that big of an influence as I thought it would be. Nevertheless, the fine grind + low dose/shallow puck formula is maximizing extraction of "balancing" flavor compounds that are present in lighter and darker coffees in similar amounts, but more need to be extracted to compensate for the acidity of lighter roasted coffee.

Is this the end of the "pucks with shower screen impression" era?

In the dark roast, bitterness rose faster than sweetness as one reduced the dose. In a very dark roast, the sweetness of the caramels is reduced, and the amount of distillate bitters increased. It could be that very dark coffees need to be overdosed and underextracted, since there's so few light compounds left. Fingerstrike dosing, which leads to low extractions, was probably introduced on the West Coast, by Peets and early Seattle roasters. They roasted very dark in those days (except for Starbucks, everyone there is gradually lightening it up), and perhaps in that context, the technique was optimal.

My object is always the same. If you don't like a shot of espresso; do you know how to change it? I don't care how you change it; only that you know what options are available, and which directions are most promising. Maybe someday we'll see no more posts of someone making a shot that didn't taste right, and asking which new machine they need to buy.

Right now, I'm doing 4 different coffees roasted to 430F (a scootch lighter than Illy, about like Metropolis's Redline): Kenya Mamuto, Rwanda Musasa, Bolivia Cenaproc, and Yrgacheffe Idido. They are all working fairly well on balancing sweetness versus acids and bright-bitters at around 13 grams in an E61 style basket. I slowed the Cenaproc down before the first crack to get a lot of bright-bitter almond, and it needs a very cool shot temperature. The Mamuto likes it hot. Surprisingly, for the Musasa and Idido, the bright-bitters dominate the acids, so I'm going fairly cool.

This was also true of Terroir's Northern Roast, as Peter reported, which also requires fairly cool shot temps after one down doses to get more of the florals and less of the woody and rooty flavors and balance everything out properly.

So roughly speaking, I'm following this overall diagnostic model for lighter roasts:
-- First Balance Sweetness versus Sour and Bright Bitter: dose less for sweet, more for sour and bright-bitter
-- Then balance Sour versus Bright Bitter: cooler for more sour, hotter for more bright-bitter.

[another_jim]

The final version is up; mostly unchanged except for endnotes, spiffed up html, and a reiteration that low dose shots are not weaker (i.e. less coffee solids in the cup) than high dose ones if pulled properly.

[espressoperson]

The final version is up; mostly unchanged except for endnotes, spiffed up html, and a reiteration that low dose shots are not weaker (i.e. less coffee solids in the cup) than high dose ones if pulled properly.

Jim,

Thanks for this work! You have shown an amazing ability to select a few of the key critical variables and ignore the nonessentials. I look at the problem and see a million and one issues too complicated to try to make sense of. Then you produce a set of meaningful findings that we ourselves can immediately put to the test. Here I was trying to find ways to stuff more grinds into the basket. After reading the original paper, I've reduced my dose by close to 20% and am producing richer, tastier shots. I don't think that would have happened anytime soon without your paper.

[another_jim]

Jim,
I've reduced my dose by close to 20% and am producing richer, tastier shots. I don't think that would have happened anytime soon without your paper.

Thanks for giving it a try.

In hindsight, it's odd we weren't playing with dose a long time ago; we've certainly been varying everything else. But it's easy to develop blindspots in a routine; it took a very flukey chain of coincidences to even get me looking at dosing variations.

[AndyS]

[This post edited several times on 2-11-07]

Hi Jim:

I still don't understand it when you say that changing the dose is the only practical way to affect solubles yield, and that changing shot time and/or shot volume is fruitless.

In my solubles yield testing, I kept the dose fairly constant and varied the volume (ie, mass) of the resulting shot. Below is a scatter graph of the data (you have the original numbers). To me, it sure looks like there's a correlation between brewing ratio (dry coffee grams/liquid espresso grams) and solubles yield. Then again, I've never taken a statistics course and I don't understand how to manage the data like you do.

As I've mentioned to you in email, I don't see a definite blonding point. Instead I see a blonding range. Therefore, there's a range of reasonable shot volumes where the shot can be terminated.

I guess the point I'm making is that keeping the dose constant and varying the volume (mass) of the shot seems like another, valid way to manipulate solubles yield.

[another_jim]

Hi Andy,

Looks that way to me too. But ...

Flip your axes, you have them reversed (it's easier to read a relation if the cause is on X, and the effect on Y). When the points are more scattered at one end than the other (heteroskedasticy, try saying that three times), use an exponent of one variable, or a log scaled axis, to get them more evenly scattered around the trend (homoskedastic -- in stats, homo is good), in this case use a log-scale for the brew ratio,

Finally a bit of algebra: you're saying yield ~ dose/shot.weight. If you are using the same basket throughout, then if you're varying mostly dose and less the shot.weight, your posited relation is the same as mine. If the variation in brew ratio comes from mostly changed shot.weights, then the relation is different.

No statistics required, just eyballs and knowing what variables are varying.

[another_jim]

The plot thickens. The logs of shot time and weight correlate very nicely to the logs of yield, virtually as strongly as the logs of dose and filter.area; while they don't correlate very well when using linear terms. The dose/filter.area relation is much stronger on the linear scale. Basically the two processes take place in completely different ways.

There's no easy way of combining the best models for both so one can tell how much each is really doing overall. I'll need to read up on non-linear methods before I get started, since my knowledge of these is severely out of date (non-linear methods are all brute force computations, and require gobs of computing power. Since there's more of that every month, these methods change really fast). It's a sweet problem.

[another_jim]

My apologies for thinking out loud and 1 step at a time in these three posts. But I flashed on a really old-school way to parcel out the relative effect of dose and basket versus shot weight & time & concentration -- rank correlation. In ranks, you substitute the rank, 1st, 2nd, 3rd, etc for the actual values. This means if the relation is monotonic, it doesn't matter whether its linear, powers, exponential, or whatever. Moreover, the coefficients tell you directly what the story is (a coefficient of one means the rank of the cause is perfectly related to the rank of the effect. Anyway, here goes:

Table:

Coefficients:
                           Estimate Std. Error t value Pr(>|t|)   
rank(filt.d^2/dose)          0.5155     0.1120   4.601 3.43e-05 ***
rank(sh.wt * sh.time/dose)   0.3502     0.1119   3.128  0.00308 **
---


The rankings of filter.area/dose explain 52% of the yield ranks (the "estimate" column times 100); while the rankings of shot.time*shotweight/dose explain 35% of the rankings. You can ignore all the other statistical stuff, it's meaningless. This simply is a rough answer, prior to a full model, on how much dosing versus flow explains solubles yield. If the second term is not divided by dose, it drops to about 25%

I guess I was wrong in my paper about the effect being minor, it's about 70% of the effect of dose, and at least 50%.

[AndyS]

Jim, here's another graph with the axes flipped. Although I, too, abhor heteroskedasticy, I left the axes linear for the time being.

These data points were all generated with the same basket, an LM ridged double.