Let's Talk Grind Retention
With more and more high end, single dosing grinders landing on people’s countertops, espresso enthusiasts are looking for ways to differentiate them from one another. As the “third wave” of coffee is characterized in part by a heightened attention to detail, quantitative analysis simply comes with the territory. But to what degree do properties like grind retention really matter, and how can we pick apart the concept to truly understand how they might apply to the espresso we make at home?
I recently had a conversation with @meticulist about this, which inspired me to flesh out my thoughts a bit more clearly. Also, I always type way too much in my Instagram messages. I’m sure some will appreciate a more legible format.
First things first, let’s align on what retention is. When people talk about retention, they’re referring to ground coffee that stays in the grinder after you grind it. This usually occurs somewhere between the burr set and the end of the chute, and it occurs because friction, static electricity, and adhesion are simply unavoidable (though static is reducible with techniques like RDT). We’re interested in minimizing retention because retained grinds from one dose can contribute both stale coffee and mass variability to the next dose.
There’s really two things people are actually talking about when they mention retention, and they’re related but different from one another. The first is average retention mass and the second is retention variation. Average retention mass is how much ground coffee is being retained on average. Retention variation is how the amount of retained coffee varies from grind to grind. Both are measured in grams, and together they form the format I’ll be using to describe retention as a whole: average retention ± retention variation.
Quick examples.
If you have incredibly small retention variation, you might always see 18.00g in and 18.00g out, but that doesn’t tell you anything about the average retention mass - your grinder could be retaining 0g or 0.1g or 1g or 10g every time and you’d be none the wiser, because all you know is that in = out, meaning the retention variation is ±0.00g.
Likewise, you might see anywhere from 17.90g to 18.10g out of the grinder for your 18.00g in, but again, this is only informing us about the retention variation. Best case scenario, you’re looking at a measly retention of 0.05±0.05g. At its extreme ends, this means your grinder is occasionally retaining 0.1g from one dose (netting you 17.90g in your basket) and depositing that same 0.1g for the next dose (netting you 18.1g to your basket) while retaining 0.0g.
Now, it’s worth pointing out that the opportunity for a large absolute (not relative) retention variation goes up with higher average retention mass, because there’s simply more retained coffee to vary. You can have 3g±1g of retained coffee, but you can’t have 0.1g±1.0g of retained coffee, because negative coffee is thankfully not a thing. I’m assuming a symmetric variation here, which isn’t necessarily true. In practice the retention distribution could be asymmetrical about the average retention mass, but I’m not going to touch that today.
How much do you care?
Before we go on a misguided experimental journey, let’s set some expectations about what different average retention masses and retention variations will mean in the context of an actual espresso. We’re here because we want good coffee, right? Okay, average retention mass first. This determines how much stale ground coffee ends up mixed in with your fresh ground coffee in the basket. Stale coffee will have a negative effect primarily on the taste. Staleness is not a particularly binary concept, but I imagine it exhibits asymptotic behavior. A lot of off-gassing is happening in the seconds and minutes after grinding, but after a day your espresso-ground coffee is probably maximally stale and it can’t get much worse. This implies that if you’re making shot after shot after shot, you might not care a lot about average retention mass, so long as you purge the grinder once before your first shot. But for many of us, an hour or more between shots is probably enough to make staleness a factor. Stale coffee will affect extraction as well, but if it’s always the same amount of stale coffee, my intuition says that you’ll naturally compensate for this when tuning in your shot. Someone can let me know if this sounds wrong, though.
So how much stale coffee makes a noticeable difference? That’s obviously a very subjective question, but if we start to think of staleness primarily as the lack of flavor and thus lack of contribution to the shot, we might be able to get in the ballpark. Do you think you’d notice if your coffee were 99% fresh instead of 100% fresh? Probably not. 95%? Maybe. 90%? Sure, I’ll buy that. If 10% of your dose is stale, and you’re grinding 18g, then we’re talking about 1.8g of stale coffee, aka 1.8g average retention mass. I’d love someone to do an experiment on this to see if I’m right or wrong. I’ll put guidance on how I’d run it at the bottom of this post.
Now let’s dig into retention variation. This determines how your dose might vary from the dose you wanted. You were planning on an 18g dose but got 17.9g or 18.1g out. Do you really care? If you’re aiming for a 36g yield, how close do you usually get to your target yield? Probably within a gram pretty reliably, but not much better. I’d wager that yield variation is ±1g, but not on a normal distribution. Likely a random one. That means that even with a perfectly consistent dose, your brew ratio with fall anywhere between 1.944:1 and 2.055:1 when you’re aiming for 2:1. By contrast, if your yield were perfectly consistent and you were dealing with ±0.1g variation on your dose, that would result in a brew ratio between 1.99:1 and 2.01:1, which would be on a normal distribution, not a random one (more right, more of the time). Not so bad, then, in comparison. Since reality is reflected in neither of these cases, though, let’s combine them and see what our distribution of brew ratios looks like. I went ahead and simulated 1000 shots of espresso, with randomly generated yields between 35 and 37 grams, and randomly generated but normally distributed grind retention of 18±0.1g. With the perfect dose and a randomly distributed yield of 36±1g, 90% of our shots are falling between 1.95:1 and 2.05:1. With our more realistic doses influenced by grind retention, and the same yield distribution, 90% drops to more like 88%, and the worst you can get are 1.934:1 and 2.067:1. I haven’t yet run a full sensitivity study on brew ratio to dose variation, but this should at least tell you that if you’re seeing dose variation anywhere around the 0.1g mark, you can probably sleep okay tonight.
random distribution of yield, 36±1g
Experimentation.
Great, so now we have a system for describing retention, but how do we measure it to actually compare grinders? This part’s a bit tougher. Measuring retention variation is the easiest. We can grind a lot of doses (like a hundred, so buy some cheap coffee) back to back to back, and measure the delta between mass in and mass out for each. The average of these deltas approaches zero since mass is conserved through the system on average. We can’t just take the standard deviation of these dose deltas and call it the standard deviation of the retention, though. That would be too easy. This is because each dose delta we’re seeing is in fact the sum of the retention from the current grind (a negative value, since the grinder is keeping it) and the retention from the previous grind (a positive value, since the grinder is generously giving it to us ahead of the fresh coffee we’ve just ground). Therefore, while the retention variation we’re after is (assumed to follow) a normal distribution, the dose delta variation follows the root sum square (RSS) of two of these normal distributions. I was just doing an RSS tolerance stack at work and I have to admit, I didn’t think I’d be doing another one for fun over the weekend. Typically we wouldn’t be able to make any assumptions about the inputs of an RSS stack based on the output alone, but here we can, because we know it’s just the same one twice! One equation, one unknown. In case you can’t follow along with my furious scribbles below, the bottom line is that we have everything we need to find the standard deviation of our grinder’s retention. And I’m going to call three times that our grind variation, which will describe our retention 99.73% of the time. That feels sufficient.
Measuring average retention mass is the harder one. On paper, the simplest way is putting the whole grinder on a scale and measuring how the overall system’s mass changes from dose to dose. In practice, a load cell with the range required to capture the total grinder weight would lack the sensitivity to reliably detect the retention mass by an order of magnitude or so. The simplest “practical” way would be to dismantle the grinder after each dose, sweep out the retained grinds, and weigh them. This, of course, is quite tedious depending on the grinder, particularly given how many data points we want to collect, right? Well, we actually don’t need nearly as many data points to form a reasonable picture about average retention mass as we do for retention variation (I glossed over it earlier, but the larger dataset required for high certainty of retention variation is why it’s nicer to derive it from dose deltas rather than from individual retentions(?) swept out of the grinder). You could do this once and get a reasonable idea of how much retention you’re working with, and do it ten times to get a very good idea.
Now let’s go back to that staleness taste test I was mentioning. If we want to run test shots with different controlled levels of stale vs. fresh coffee, how can we do that if our grinder is already putting out an unknown mix?
Grind enough coffee for all the stale mass you’ll need in testing and leave it out overnight.
For each portion of fresh coffee per dose, let the first bit of coffee fall out of the grinder and onto your counter instead of into your dosing cup. The stale, retained coffee comes out first, and the fresh coffee follows — all we want is the fresh part.
Build your doses by combining the right amount of stale and fresh coffee just before each shot.
Closing thoughts.
I realize this is all theoretical, and based on a number of assumptions about distributions. My intent isn’t necessarily to provide an answer to the question of “how much retention is ok?”, but to first correct it to “how much average retention mass and retention variation is ok?” and then provide a reasonable framework with which we might answer it. The answers will be different to different people, based on their personal sensitivities to taste, workflow, time, and money, but I’ll be curious to know if this helps any of you answer it for yourself.