A Study of Espresso Puck Resistance and How Puck Preparation Affects it

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through the Amazon links identified below.

Before I begin to characterize how different grinders or espresso preparation methods affect extraction and taste, I decided to test how repeatable my shots of espresso are. The DE1 Decent Espresso machine is pretty nice for this purpose because it shows you graphs of your live pressure, the flow of water coming out of the shower head, and how the weight of your espresso beverage changes over time if you connect a Bluetooth scale to it.

To carry this experiment, I bought a large bag of washed Mas Morenos Honduras coffee roasted by my friend Andy Kyres who owns Color/Full Coffee Corp, dialled it in on the Niche Zero grinder, and followed Scott Rao’s puck preparation technique that he posted on his Instagram TV channel. I started on the Niche first because I can compare it to other Decent users more easily, and I plan to do a similar experiment on my EG-1 grinder very soon.

My current grinding corner, where I am preparing a battery of tests to compare the 68mm conical Mazzer burrs in the Niche Zero to the 80mm SSP ultra-low-fines flat burrs in the EG-1.

For this experiment, I dosed 18.0 grams of coffee in the Niche and ground directly in the portafilter, like Scott shows in his Instagram video. I used grind size 14.0 on the Niche for the whole experiment. I made sure to send grounds as much as possible at the edges of the portafilter at first, and then filled up the rest in a circular motion. To do this, I moved the portafilter around in a “nutation motion” with my left hand while the Niche was grinding. Here’s a video of my attempt at this:

I used the tall Decent portafilter funnel to help get all of the grounds in my portafilter, and Levercraft’s WDT tool in its default configuration. Levercraft’s is my favorite WDT tool so far; I find that the large number of needles (8) and their wide angle makes it much easier to distribute the grounds evenly and obtain a level puck.

Levercraft’s WDT tool

I decided to test a few different puck preparation techniques while I was at it. First, I tried using exactly the Weiss Distribution Technique (WDT) method that Scott showed on his IGTV channel, where you only de-clump and stir the top third of the depth of the coffee puck before tamping. You then tap gently and tamp once, trying to get a coffee puck that is as level as possible. I’ll call this method “surface WDT” here. You can see a demonstration in this video:

I also tried another method, “deep WDT”, that is more popular among DE1 users on the Decent Diaspora forum, where the WDT tool is inserted at full depth and used to stir the coffee grounds more vigorously in circular patterns, slowly going up and then finishing on the surface, like I do in this video:

This is followed again by a small tap of the portafilter on the tamping mat, and by tamping. I used The Force Tamper (Amazon Affiliate link) to reduce variations due to non-level tamping. I found that holding the base of the tamper with my left hand and then tamping with the right hand allows me to obtain a more level coffee puck, like I show in this video:

I left the tamper at its default level of pressure. Scott Rao recently popularized the use of paper filters at the top of the puck to help distributing flow evenly across the top of the puck, and I decided to also include that method in today’s experiment. With the first two methods that I described above, I did not use any paper filter. So I added two more kinds of methods where I did either surface WDT or deep WDT, and also added a dry Cafelat Robot 58mm paper filter on the top of the coffee dose. I always added the filter such that the creped side was facing the coffee puck, because this is where I want to maximize the surface of contact. I just placed the dry paper filter on top of the tamped coffee bed like I show in this video:

In order to minimize any systematic effects of my getting tired, slowly improving my technique, or the grinder or DE1 getting warmer, I alternated between methods. I started by pulling 6 shots where I alternated between surface and deep WDT without paper filters, and then I started alternating between all four methods. I used the “Best Pressure Profile” that peaks at 8 bar after preinfusion, and then slowly goes down to lower pressures over time. My goal was to assess how repeatable would be the puck resistance and its decline with time, so I like to keep a fixed pressure for this experiment because I know changing the pressure from shot to shot could affect the physics of what size of coffee fines can migrate down the puck, and at what rate they do so. I also pulled three shots before starting the experiment, to dial in the coffee and ensure that the DE1’s temperature was stable enough.

Note that I did not refract any of the shots in this experiment: doing this properly takes time, and I decided to focus this experiment on repeatability of puck resistance only, as this allowed me to pull and log 24 shots in a bit less than two hours, probably twice as much as if I had refracted each of them.

The DE1 “best pressure profile” I used for this experiment

Before we discuss the results of this experiment, it is worth talking a bit about the DE1 graphs for one of the no-paper, “surface WDT” shots described above.

DE1 graph for a representative espresso shot with the “best pressure profile”

I know the DE1 graphs contain a lot of packed information, especially if you are not familiar with the machine. The green line shows the pressure in units of bar and how it changes over time during the shot; you can see a slow rise of pressure during preinfusion (in the first 10 seconds), which causes the machine to exit the preinfusion phase when a pressure of 4 bar is reached. The blue curve shows the rate at which water flows out of the shower head, as calculated by the DE1, based on measuring the voltage of the pumps and a complicated physical model of the whole machine. As you can see above, I used a 3.5 mL/s preinfusion during the preinfusion. The early 4 mL/s start is related to the firmware set-up of my machine, and I did not bother changing it because I don’t think it matters much for this test.

When the pressure reaches 4 bar, the machine switches from a “flow-controlled” mode to a “pressure-controlled” mode where the pressure is immediately cranked up to 8.4 bar, and then it slowly decreases to about 5 bar at 35 seconds. The red curve shows the temperature near the shower head, in units of degrees Celsius divided by ten. The brown curve shows the gradually increasing weight of espresso as measured by the Acaia Lunar scale that I had connected to the DE1. You can see that the brown curves becomes similar to the blue curve at around 17 seconds: this is when the puck becomes mostly saturated with water, where the flow of water inside the puck becomes the same as the flow of water that exits the puck. Because the DE1 uses a complicated physical model of the machine to estimate the blue curves (the flow from the shower head), it is not atypical to see errors of about 10–20%. It is notoriously difficult to actually measure the flow of water through pumps with a better accuracy in the context of espresso making. This explains why the blue and brown curves are not perfectly on top of each other even late in the shots. The brown curve also includes dissolved solids and should in principle be about 10% higher than the blue curve for a short while, but this effect is dominated by the systematics of the flow calculation at the moment. Even if we had a perfect blue curve, measuring a flow rate very rapidly with a 0.1 gram-precision scale for the brown curve is hard, because the measurement error gets inflated if you try to measure the flow in a very short amount of time. I went around some of this problem by using custom smoothing algorithms I wrote (not those of the DE1), based on running-box local second-order polynomial fitting, a running “median” box followed by a running average-based smoothing box (you can safely ignore all of this sentence if it means nothing to you).

The yellow and orange curves show estimates of the puck’s resistance based on the pressure curve (green) and either of the two flow curves (blue or brown). For reasons that I explained in my last post, the resistance of the coffee puck to the flow of water can be approximated with the flow of liquid through the puck divided by the square root of the pressure drop. In the DE1’s app, these resistance curves actually show the square of the flow (in mL/s) divided by the pressure drop (in bar), and thus estimates of the square of the puck resistance, but through this post I am showing the resistance for simplicity. It makes sense that the yellow and orange curves only meet when the puck is saturated; before that moment, the whole presuppositions of Darcy’s law and the simple characterization of the puck with a hydraulic resistance don’t make much sense because it has not reached the “percolation threshold”, i.e., the point where espresso starts coming out from under the puck.

[Edit Feb. 7, 2021: in the paragraph above I had originally misquoted the puck resistance as flow divided by the square root of pressure, whereas it’s actually the square root of pressure divided by the flow. Thanks to Sam Roesch for pointing this out.]

I believe that the point where the pressure starts to rise during preinfusion corresponds to the moment where the empty headspace above the puck is completely filled with water, and now the machine has to apply pressure to get more water in the system. Water is basically an incompressible fluid in the context of espresso (you need 220 bar to compress water by 1%), but the machine somehow manages to get more water in there before espresso starts dripping out (in all cases, the first drops of espresso appear at 11-12 seconds). This means that something must be compressing, and if it is not water, it has to be the espresso puck. This effect was actually demonstrated quite neatly with transparent portafilters; the coffee puck is compressed by the espresso pump much like a spring, and it decompresses if no more pressure is applied on it. This is why the “puck resistance” as seen from above the puck (the yellow curve) starts to rise at the end of preinfusion. But then, what happens next is interesting: the resistance of the puck gradually decreases, and then stabilizes.

A demonstration of espresso puck compaction with a transparent portafilter

I used to think this was related to the slurry getting less viscous and therefore more easily flowing through the coffee pores as it gets less concentrated. A paper by Sobolik et al. (2002) showed that coffee beverages with concentrations in the range 0—10% show changes of viscosity of about 50% at 80°C (Fig. 3), which would be expected to cause a decrease in puck resistance by 50% as the coffee solubles get depleted. This is not enough to explain the full changes in resistance that we observe (about a factor 2 to 3.5 here), so while it may explain a small fraction of the decrease, it is definitely not the full explanation.

Figure 3 of Sobolik et al. (2002), showing the dynamic viscosity of water (black circles) and concentrated coffee at 10% TDS (grey circles), 20% TDS (white circles) and 30-50% TDS (triangles), as a function of temperature. The symbols show empirical measurements by Weisser (1972) and the lines are best-fitting power laws.

Another potential culprit for this decrease in bed resistance could be related to the migration of fines, a topic I discussed extensively in my upcoming book The Physics of Filter Coffee. However, the change in bed resistance is so strong and abrupt that it made me worry a third cause might be the explanation: the espresso puck might still be partially dry when we are at the end of the preinfusion. I now think that this is the most likely explanation, because adding just a 10 seconds pause after the preinfusion completely removes the peak in bed resistance, and there is no reason for the slurry to have gotten less viscous, or the fines to have moved anywhere, during a 10 seconds pause without flow. This is a topic for a future post, but I believe it will be really important to add a short pause after preinfusion to allow the puck to fully saturate and improve espresso. Scott Rao has done something similar in the past with his “blooming espresso”, but I suspect we may get some of their benefits with a pause as short as 5—10 seconds after preinfusion. Furthermore, I think that the 30 seconds blooming phase of the blooming espresso allows the puck to decompress, and I am not sure that it can be easily compressed again once the puck has been saturated—not that this is necessarily a problem anyway.

DE1 graph for a shot with a modified version of the “best pressure profile”, where a 10-seconds pause was added immediately after preinfusion. The pause drastically reduced the peak in puck resistance that would usually happen in the first 10 seconds of the shot. This indicates that the true culprit for the peak in bed resistance in normal shots might be an incomplete wetting of the puck.
Comparing only the puck resistances (as calculated from the shower head) of two shots with the same puck preparation and grind size shows how the shot with a pause (red) completely skipped the peak in bed resistance that occurred with the no-pause, normal shot (blue). The puck resistance almost caught up to the normal curve immediately after bloom. The smaller puck resistance even after 30 seconds could be caused either by variations in my puck prep, or by the fact that the coffee puck did not compress as much in the latter case. We might also be seeing a slight decrease of 20% in the red curve during the shot that could be explained by a decrease in the viscosity of the slurry.

I know this was a wordy explanation of the DE1 graphs, but I think it will really help us better interpret the results of this experiment, as well as future ones. Now, here’s what I obtained when comparing all of the “surface WDT” or “deep WDT” shots without paper filter:

Combined DE1 graphs for the “surface WDT” shots (top panel), and “deep WDT” shots (bottom panel), both without paper filters.

The top panel groups all of the shots where I used the “surface WDT” method on top of each other, and the bottom panel groups all those where I used the “deep WDT” method. You can see that the exact moment where the preinfusion ends varies a bit from shot to shot; this is related to how fast the pressure ramps up to 4 bar during preinfusion, and therefore probably related to slight differences in the puck’s resistance. I adjusted all resistance curves by small deviations in dose to make them most comparable to each other, but this was a small adjustment of less than 1% because all of my doses were between 17.8 and 18.0 grams. One thing is immediately obvious from the graphs above: the peak of the puck resistance (yellow curve) varies quite wildly from one shot to the next, with either method. The “deep WDT” method got me peaks in resistance that are closer together, and therefore a bit less variable, but there are still a few outliers.

Puck resistances calculated at the shower head for the shots with surface WDT (blue curves) and deep WDT (red curves). The deep WDT method seems to reduce variations in peak resistance slightly.

I made similar figures for the cases with a dry paper filter on top of the puck:

Combined DE1 graphs for the “surface WDT” shots (top panel), and “deep WDT” shots (bottom panel), both with a dry paper filter on top of the puck.

The graphs above show a very similar picture: the “deep WDT” method still seems to get me slightly more consistent peak resistances from one shot to the next, but the paper filters do not seem to clearly help repeatability.

We can compare the exact value of the peak puck resistance as measured from the top of the puck (yellow curve):

In this graph, I show each shot as a filled grey circle, grouped by puck preparation method. The red circle and vertical bars represent the average and standard deviation for each method, and the blue circle and vertical bars represent the median and median absolute deviations, which are similar concepts to the red circles, except that they are less affected by outliers. There are only a few clear things that seem to be happening here. First, “deep WDT” helps reduce the spread in peak resistance, and perhaps increases the average peak resistance a bit, although this effect is quite diluted by variations in my puck preparation technique. Second, the paper filters don’t clearly improve the situation, but they also seem to increase the overall puck resistance by about 7%. This is not too surprising, because the filter acts as an additional percolation layer, with its own hydraulic resistance that will contribute to increase the global resistance just a bit. In the last two groups to the right, I divided the resistances of the “paper filter” cases by 7% and combined them with the “no-filter” cases, to show all of the “surface WDT” and “deep WDT” shots together, whether I used a paper filter or not. Yet again, we see the effect of the drastically reduced median absolute deviation (blue vertical bar), and the slight decrease in standard deviation (red vertical bar). This tells us that the majority of shots are much better grouped together in terms of peak resistance, but that there are still a few shots that were very different.

Viewing the DE1 profiles for all “surface” vs “deep” WDT methods, regardless of paper filters, is also visually compelling:

Combined DE1 graphs for the “surface WDT” shots (top panel), and “deep WDT” shots (bottom panel), combining all shots regardless of whether a paper filter was used or not.

One thing I find particularly interesting about these data is that all of the outliers with the “deep WDT” method—about 15% to 25% of my shots—seem to be chokers (high resistance), not gushers (low resistance). Intuitively, I expect that mistakes in puck preparation will tend to leave low-density regions in the puck, or worse, small channels, which would both favor the occurrence of gushers. But something else seems to be happening here, that causes these outliers. I have two hypotheses for this: (1) my distribution of coffee grounds in the portafilter is generally really bad, and there were only a few cases where I had a very good puck preparation, i.e. the high-resistance cases above were the only great shots; or (2) another random but relatively rare process causes the puck resistance to go up—possibly some larger coffee particles clogging one or more holes in the espresso basket, or some coffee oil that I did not clean up perfectly from the last shot reduced the effective sizes of some of these holes. If the latter explains my observations, then adding a paper filter at the bottom of the puck should help reduce this effect, because blocking any portafilter hole would not affect the hydraulic resistance of the full system as much with a paper filter above it. Water could still go around the clogged hole quite easily by passing through the paper filer.

However, I have a slightly frightening suspicion that the true explanation is my terrible repeatability at distributing the coffee grounds in the portafilter. The reason is that I noted in my logs that some of these shots had a particularly beautiful even appearance of drips at the bottom of the portafilter holes, and a very clean and even flow of water, and they tend to correspond to the shots that had the highest peak resistances in the graph above. The more frightening aspect of this is that all 24 shots from the experiment looked ok visually; we are talking about minor visual differences here. If this is the explanation for the varied puck resistances, then it would mean puck preparation is absolutely unforgiving in terms of how much the peak resistance varies from one shot to the next. While deep WDT seems to improve the general picture by eliminating most of the worst gushers, it did not seem to improve the rate at which I obtained these higher-resistance shots. This would mean that doing deep WDT helps us to avoid the worst shots, but also that no style of WDT really fixes underlying problems in ground distribution during grinding. This also calls for future experiments to test more methods of ground distribution.

Another fact that hints at an imperfect distribution is the following: all of the shots that I pulled for this experiment, including those with a paper filter at the top, yielded spent pucks that were slightly more hollow near the center:

All the spent pucks from today’s experiment had a slight hollow near the center, indicating that water flowed unevenly through the puck.

The fact that this happened even with a paper filter at the top suggests that this was not caused by the shower head damaging the top of the puck. Rather, it might indicate that the way I distribute grounds onto the portafilter with the circular motion leaves a lower density near the center of the puck. I did a quick test of this by grinding one more shot as a mound into the portafilter and then using the deep WDT method, and the spent puck was indeed much less hollow at the center. Another way to completely fix this problem is to add a paper filter below the puck.

Now, the wide variations in peak resistance that I observed above are not particularly great news for the adaptive profile that I created for the DE1 in my last post. These adaptive profiles try to go around imperfect dial-in in grind size by adopting whatever flow occurs at peak pressure, which means that they rely on the peak resistance to decide what is the best flow for the rest of the shot. In other words, all of the shots in this experiment, which used the exact same grind size, would have yielded flow rates that varied between 1 mL/s and 2 mL/s, and generated quite different styles of beverages. A straight flow-controlled shot with a pre-determined flow rate would be even worse, however; these variations in bed resistance would be reflected in variations twice as large in pressure, and would have caused some of these shots to choke or stay at extremely low pressures, both of which seem to yield quite bad-tasting shots, instead of just different styles of shots.

More typical profiles like the “Best Pressure Profile” seem more forgiving in that regard, and this is because even though they start at wildly different flow rates, they seem to converge to more similar flow rates near the end of the shots, as the puck resistances converge to values which do not vary as much from shot to shot. This convergence effect means that the average flow during these pressure-profile shots do not vary as much, and the shots probably taste more similar as a consequence. I have indeed tasted most of these 24 shots above, and while some were definitely better than others, they mostly did not taste wildly different.

One way we can better characterize this convergence of puck resistance is by looking at the stabilized value near the end of the shot. To do this consistently across all shots, I chose the moment where any shot reached a beverage weight of exactly 40.0 grams, and compiled the puck resistance at that moment. You can see that the results, below, show much less variations, and also less dependence on the puck preparation technique:

The figure above lists the end-of-shot resistances as calculated from the yellow curves (based on the flow rate at the shower head). I made a similar figure by using instead the probably more-reliable puck resistances calculated from the Acaia Lunar scale:

Here, we seem to observe a slight improvement in repeatability when using both a top paper filter and the deep WDT method.

I believe that these data are all a strong indication that the standard DE1 preinfusion is insufficient to reach saturation of the puck before we actually start extraction. I think this is the cause for less overall repeatable shots, and probably for much less even extractions as well. This potentially has even more impact on my adaptive profiles because they rely on a moment where the puck is not yet saturated to determine what the optimum flow rate should be for a given shot.

I think that repeating this experiment with a short pause after preinfusion, or even a blooming profile that then mimics the pressure curve of the “best pressure profile”, would yield much more consistent shots.

So, what are my conclusions this unusually long post, even by my standards?

  • My shots show variations of about 5% in terms of puck resistance near the end of the shot. This is comparable to the kind of resistance change one would expect from adding or removing 1 gram from a 18.0 grams dose. I do not know yet what this corresponds to in terms of grind size, but it indicates that I have to be careful in concluding that any small changes in puck resistance was caused by something else than random variations in my puck preparation.
  • When I use the surface WDT method, my shots show significant variations in peak resistance, by about 40%. This means that using surface WDT will cause wide variations in flow rate when using the adaptive profiles. As a consequence, I will keep using the “deep WDT” method for now.
  • Using a paper filter on top of the puck does not seem to significantly improve the repeatability of my shots, nor does it improve the hollows that often form at the center of my spent pucks. It is still possible that the top paper filter improves the average extraction yield and/or the flavour, so I have not yet made up my mind about whether I should keep using the top paper filter.
  • This experiment taught me that I need to keep working on how I distribute the coffee in my portafilter while grinding. As a consequence, I will keep experimenting with other methods in the near future.
  • This experiment tells me that current preinfusion steps in most of the DE1 profiles are insufficient to fully saturate the puck before extraction. I will therefore experiment a bit to determine how I can improve preinfusion and I will revamp the adaptive profiles very soon.

For those who are interested, I made my log publicly available here for the 24 shots used in this experiment. I also zipped all of the DE1 data files from my 24 shots with the “best pressure profile” here. I also generated 7 more figures to diagnose my data and make sure that no further systematics affected my peak resistance curves; for instance to ensure that variations in preinfusion temperature preinfusion duration or the shot order did not cause issues. I made those figure available here for the more technically minded readers. I also posted them at the end of this post below.

Disclaimer: I receive no financial benefits from any of the companies mentioned above, and I have no business ties to them. Decent Espresso generously offered me a 25% discount on their DE1 machine, and Weber Workshops offered me a set of SSP Ultra-low-fines burrs and their glass cellars, without obligations or expectations. All my impressions of the gear that I use are my own and never financially motivated. The owner of color/full is a personal friend.

I would like to thank Johanna “Mimoja” Amélie Schander for having coded the required Bluetooth communication codes on the DE1 and making it possible to pair the DE1 with Acaia scales.

An Espresso Profile that Adapts to your Grind Size

OK, this will seem sudden to almost everyone, and even more so to readers on my regular blog, but I decided to break out the chronology of my next couple posts to talk to you about something that excites me a lot. And from that break in chronology comes an announcement that I would normally have made in an earlier post: I am now the happy owner of a DE1 Decent espresso machine, thanks to my Patreon followers who amazingly already backed me up to this level of coffee geekery equipment. I never expected this to happen this fast. I received the machine just a bit more than a month ago, and it introduced significant chaos into my coffee habits, thoughts and plans. Positive chaos, however. For those who do not know, this machine might be better described as a computer filled with sensors that drives hot water through your coffee. It a weird, and delightful concept. If even an espresso machine was suited to me and my almost unhealthy level of coffee analysis, this is the one.

In just a month of having used the DE1, I think my understanding of percolation has solidified a lot. I am still creeping out of novice territory in terms of the mechanics of actually preparing espresso and tasting it, but this thing is almost live having X-ray vision into the portafilter if you think hard about the graphs it shows you and have the right tools to interpret them (and the obsession to dream about these graphs). One of the frustrations I encountered, however, relates to preparing espresso in general: the amount of coffee that needs to be wasted while adjusting grind size, dose, and perfecting puck preparation techniques can be significant before we obtain a great-tasting, “dialled-in” shot.

This is particularly true when using the DE1 machine in flow mode, i.e., where the user decides on the desired flow rate (or, more generally, a flow profile) instead of a pressure profile. In principle, a flow mode is great because it can prevent runaway effects where the flow becomes too fast near the end of a shot, by instead automatically adapting the pressure profile of the machine to keep a flow of water that is constant in time. It can also reduce the impacts of channels where too much water takes the same path of the coffee puck, by reducing the pump pressure when this happens.

A DE1 shot in flow mode. The green curve is pressure, and the thick blue curve is flow rate (both with the same axes, in units of bar or mL/s). The dotted blue curve is the cumulated volume of water dispensed by the machine, and the harder-to-spot dashed blue line is the flow rate targeted by the machine at each moment. This shot peaked at 8 bars, which is in the range typical of a nice espresso shot.
A DE1 shot in flow mode that was ground too fine. The curves use the same definitions as above (the yellow curve is an estimate of the puck’s hydraulic resistance). The pressure reached very high values here, causing a high level of puck compression that prevents a good, even percolation. This shot tasted harsh, muted, and bitter.

In practice, however, it is harder to dial in a shot when using a flow profile. This is true because the pressure can react very strongly to small changes in coffee bed resistance (i.e., grind size, puck prep, and other factors) when the machine requires a fixed flow rate. Just to clear out any misconception, the flow rate I am talking about here is actually the volume of water per second that the machine is sending into the coffee puck. At equilibrium (after the coffee puck saturates completely), this will be the same as the drip rate out of the coffee puck because water cannot be compressed at 9 bar (very far from it). Bear in mind this is not exactly the same as the microscopic velocity of water around coffee particles.

It might seem surprising that pressure reacts so strongly to changes in bed resistance if you are familiar with Darcy’s law, which dictates that the drip rate of a fluid depends on the pressure drop across a percolation medium, in a linear way, which means that if you double the pressure, the drip rate should also double. However, Darcy’s law make numerous approximations in order to come to this conclusion, and one of them completely fails in the context of espresso making: Darcy’s law assumes that the medium of percolation is fixed, and immovable. This is wrong in many ways when we pull an espresso shot. As you increase the pressure, the puck compresses and becomes shallower (this, taken alone, would increase the drip rate), and some coffee powder gets detached and moves around through the coffee bed (the drip rate would increase when fines are liberated, then decrease if they get captured further down in the coffee bed or a paper filter). But even more importantly in this situation, the coffee particles getting compacted closer together means that the sizes of pores between the coffee particles become smaller. In technical terms, the porosity of the coffee bed becomes smaller, and this reduces the drip rate.

Looking at the classical form of Darcy’s law can be insightful for this:

In the equation above, Q is the drip rate, L is the thickness of the coffee bed, A is the cross-sectional are of the coffee puck, μ is the viscosity of the slurry, k is the intrinsic permeability of the coffee bed (how easily it lets fluids through, i.e. the reverse of resistance), a property that depends on particle sizes, shapes, roughness, and pore sizes, and Δp is the pressure differential across the puck that is applied by the espresso machine. The considerations that I mentioned earlier mean that both L and k are in reality functions of the pressure drop, and k is even a function of time while fines are still moving around. The exact ways in which L and k depend on pressure are relatively complex and not too important here, but in practice, the Decent users community has noticed that the combined effect of all these phenomena is a drip rate that depends roughly on the square root of the pressure drop:

This seems to hold as long as the pressure does not go above 10 bar, where the drip rate becomes extremely small even at high pressures. John Buckman at Decent calls this a “secondary puck compression”, which may be caused by the cellulose of coffee particles deforming and blocking any remaining pores much more efficiently. We try to avoid going above 10 bar because of this reason; it makes it very hard to manage the flow rate efficiently when it happens, and the pressure tends to peak very high and the espresso shots stall completely. If you play a bit more with Darcy’s law and split out the coffee bed’s permeability k into its grind size and porosity components, you would end up with the pressure drop Δp going with the squared power of the parts of the puck’s resistance that have to do with grind size.

I know it is popular to suggest using the Darcy-Weisbach equation whenever the assumptions of Darcy’s law fail, but Darcy-Weisbach is a slightly more general equation that accounts for friction in turbulent fluids, and rule-of-thumb calculations as well as computer simulations by Ellero & Navarini (2019; for those well versed in hydrodynamics, the microscopic Reynolds number associated with water flowing inside the coffee puck is in the range 2—12, far from the thousands required for turbulent flow) indicate that the flow of water inside a typical espresso coffee puck never comes any close to the velocities required for turbulence. Hence, Darcy-Weisbach is of no help in espresso making, perhaps unless a gigantic channel happens that would leave a visible, large hole in the spent puck.

Pardon me for this relatively technical detour—all this was mostly useful to point out that the pressure curve of the DE1 set in “flow mode” will react to the square of any change in puck resistance to keep the drip rate fixed. In other words, a coffee bed twice as more resistant will require 4 times as much pressure. As you can imagine, this volatility in the pressure curve makes it hard to obtain “standard” espresso shots at about 9 bar, and especially in avoiding a pressure that goes to 10 bar and up.

As a consequence of this, several DE1 users usually adopt a more easily approachable pressure profile when they pull espresso shots. One very popular pressure-based profile is to ramp up to about 8—9 bar right after preinfusion, and then let the pressure fall down gently as the resistance of the coffee puck gradually reduces. This reduction in puck resistance is pretty much universal, although not all coffee pucks change at the same speed. I’ll come back to this in a future post, but I believe this is mostly caused by fines getting detached and reaching the cup of coffee, and probably also due in part to the viscosity of the slurry going down as solubles get depleted and also carried in the coffee cup. Thus, slowly reducing the pressure will maintain a more constant drip rate through the espresso shot, and it seems that this is generally preferred in the coffee community. I do not know why a constant flow rate may lead to better taste, and I am definitely interested to hear if some of you have hypotheses to explain it.

If you still follow me, you will notice that neither of these two types of profiles are “ideal”—flow profiles can cause volatility in the pressure curve unless they are perfectly dialed in, and pressure profiles can lead to changes in flow rate if the rate of decrease in pressure does not match the rate of decrease in puck resistance.

A DE1 shot in pressure mode that went relatively well. The exact shape of the pressure decline did not match how the puck resistance went down exactly, and as a consequence the flow rate got a bit faster near the end of the shot, which is not ideal for taste according to Scott Rao. My (very) limited experience seems to match with this.

With all of this in mind, and after reading John Buckman’s reflections on the three (or for) “mothers” of good espresso recipes, and starting to use Rao’s Allongé profile on the DE1, I had a realization: The Rao Allongé is intended for the pressure to peak in the same range as an espresso shot (8—9 bar), with the same goal of a constant (but faster) flow rate after preinfusion that leads to a slow decrease in pressure, and a similar typical brew time around 30 seconds. The only differences are: a coarser grind, a longer ratio (more water per grams of dry coffee), and a faster flow rate. At first, this seemed a bit arbitrary to me, but then I realized that the Allongés seemed to taste noticeably better when I managed to get them to peak near 9 bar — they are built as flow profiles on the DE1, so it requires a bit of fine-tuning of the grind size too.

A DE1 shot of a Rao Allongé, recently shared by John Buckman on Twitter. Notice how similar this looks compared with an espresso shot in flow mode, except for the faster flow rate. The pressure still beaks near 9 bar, and decreases as the puck resistance also decreases. The extraction even lasts for a similar time, despite the much coarser grind size!

I can’t say I’m sure about this, but this seems to indicate something that might be fundamental about preparing coffee beverages with percolation and the use of pressure. Maybe we just want to get a flow rate that does not decrease during the extraction phase, and get the fastest flow rate that won’t enter this problematic regime where the pores of the coffee bed start closing up beyond 10 bar of pressure. I could see how a flow rate increasing during the extraction would be bad: this would cause a more potent extraction toward the end of the brew, where most of the solubles were already removed anyway, and this might be efficient at extracting the larger molecules that do not taste as good, or maybe even some cellulose that make up the walls of the coffee cells in the most extreme case. Maybe at some point in the future we may realize that decreasing drip rates near the end of the extraction are also good (or perhaps even better), but currently no one seems to be exploring this possibility, for an obvious reason: a coffee bed’s hydraulic resistance usually goes down during the extraction, so unless special care is taken, the drip rate will tend to go up, not down as this happens.

If that intuition turns out to be right, it begs another question: why would only typical espressos at 1:2 (1 part dry coffee to 2 parts water) or 1:3 ratios, and then the Rao Allongés at 1:5 to 1:6 ratios taste good? I don’t see an obvious reason to believe that there is nothing else good-tasting in between, or even on either sides of these two possibilities. Taste is obviously subjective, and the perception of taste is well known to depend a lot on concentrations, but what if those are just two recipes along a continuum of good-tasting recipes that peak near 9 bar and maintain a constant drip rate throughout the extraction?

This question led me to think about another way in which we could program the DE1 to achieve good extractions. If what really matters is peaking the pressure near 9 bar and then maintain a constant drip rate of whatever the drip rate happens to be at the moment where the pressure hits 9 bar, why not ask the machine to do just that? Apart from slight differences in preinfusion, this same profile would yield a “flow profile” espresso shot with a fine grind, and a Rao Allongé profile with an appropriately coarser grind. But then… it would also potentially yield good-tasting beverages anywhere in between, assuming that the range in beverage concentrations is also pleasing to the barista.

A visual representation of 1:2 or 1:3 ratio espresso shots and the Rao Allongé, along with a possible family of good-tasting recipes along the blue line. I am hoping that dialing in with the new profiles proposed here will move us along the good-tasting recipes (purple arrows) rather than outside of them (yellow arrows) and toward the less good-tasting regions.

If this holds to be true, it would be even better than it sounds. This would not just open up new interesting beverages, but it would also mean that getting the grind size wrong would yield a good-tasting beverage anyway. Imagine you use such a profile and get the grind size grossly wrong and grind way too coarse, and end up with a good-tasting Rao Allongé. That may not be what you were trying to achieve, but if it tastes good that would already be a start! And now imagine you just get the grind size wrong by 5-10 microns in burr spacing. Instead of getting a shot that peaks at the wrong pressure, or where the drip rate craps out completely and generates a sour or bitter brew, what you would get is a slightly different beverage type, still along the “family” of good-tasting recipes, without all the extraction defects that follow an uneven or otherwise problematic percolation.

As you can probably tell, I’m excited about this possibility. I would love to stop worrying about pulling a shot with a bag of coffee I only have 100 grams of. And I would also love to waste zero grams of good-tasting coffee regardless of the bag size.

In practice, there is still one hurdle in getting the Decent to behave like this (i.e., peak at 9 bar then keep whatever drip rate at that moment constant). There is currently no way to specify a desired drip rate on what is currently being measuring during a shot. When I asked John Buckman about this possibility, he mentioned that they plan to design a whole programming language surrounding the construction of DE1 profiles, and that this would be a great application for it. I was happy to hear this, and then in the following days I kept realizing even more points about how cool such a profile might be. This made me so excited that I decided to try and find a way to bend the current Decent software into doing it anyway, even if it wouldn’t be perfect.

After a bit of fiddling, I realized that there is actually one way to achieve such a profile with the current DE1 software. This method is limited by the finite amount of steps (20) that the DE1 will accept to execute during a single shot in advanced profile mode. Any well-behaved profile should indeed fit within 20 steps, but this one does not belong to this category, because it is trying to work around the current capabilities of the machine. Here’s the idea:

  1. Execute a normal preinfusion. For current DE1 “flow profile” shots, this is a flow profile step at 3.5 mL/s that triggers an exit to the next step whenever the pressuredrop reaches 4 bar, indicating that the puck is saturated with water or close to it.
  2. Rise the pressure and hold it at 8.6 bar for 4 seconds, as one would usually do at the start of a pressure profile on the DE1.
  3. Now, the weird part: Attempt to start a long flow profile step at a flow too fast for typical espresso, in this case 3.5 mL/s, with a trigger that exists the step if the flow ever reaches a value slower than 3.4 mL/s.
  4. Follow with a very similar step that attempts to establish slightly slower, 3.3 mL/s flow profile step that triggers out if the flow reaches 3.2 mL/s.
  5. Keep adding steps like this until you reach step #19, with a value of flow rate that would typically be too slow for espresso — in this case, 1.0 mL/s, and that triggers a skip whenever the flow rate reaches 0.9 mL/s.
  6. Add a final flow profile step that is even slower, in this case 0.5 mL/s, without any trigger for skipping that last step.
Standard preinfusion and pressure rise steps in the new “adaptive” profiles I’m proposing
Subsequent steps that scan for the current flow rate in the“adaptive” profiles

It may seem like every one of these steps starting from the third one are self-contradictory, but in practice this serves a purpose. Imagine you grind a bit too fine, and instead of reaching your goal of, say, a 2.2 mL/s flow rate during the extraction phase, you actually reach 1.8 mL/s. What will happen ? Well, the machine will peak at 9 bar because you asked it to, the flow rate will reach 1.8 mL/s, and then the machine will attempt to run step number 3, and it will ask the machine to up the flow rate to 3.5 mL/s while immediately starting to check that your flow rate is at least 3.4 mL/s. This happens so fast that the pumps won’t even have any time to react and the software will have triggered out of steps 4, 5, 6 and so on until the exit condition is not triggered anymore: this will happen at whatever step has an exit trigger slightly below your current flow rate (1.8 mL/s). In practice, this will happen at whatever step is asking the machine to reach a flow rate pretty close to 1.8 mL/s, and there you have it: you effectively asked the machine to keep whatever flow rate was currently going on.

I was honestly skeptical this would work because it is so twisted, but to my delight it worked exactly as I wanted! Now, the limit of 20 steps required a careful selection of what flow rates we want to test during steps 3 to 19. I asked some long-time Decent users (thanks Scott Rao and Stéphane Ribes) and looked a bit on the Decent user forum for what ranges of flow rates people are usually pulling their shots at. This depends a bit on roast date, and on the desired ratio (as you might expect from the discussion above), but it mostly seems to happen between 2.0 and 2.7 mL/s. For this reason, I decided to scan this range with steps of 0.1 mL/s, and also scan a wider range (0.5 to 3.5 mL/s) with more spacing between the individual steps. With the more flexible coding capabilities of the future Decent software, the machine will be able to keep exactly the flow rate you had whatever its value is, but right now this will cause the machine to adopt something close, unless you are so far from dial in that your flow rate is outside of these bounds a the moment where the pressure profile peaks.

If you think the profile seems complicated, you are right that the means to get there were complicated, but in reality using the profile should not be more complicated than any other ones, and hopefully it will be simpler because it might taste good in a wider range of grind sizes. Just view it like this: the preinfusion and the initial pressure peak are fixed, and the shape of the flow profile is also fixed. All that changes is how fast the flow rate is, starting at the pressure peak. You are dialing in the fixed flow rate with your grind size. In principle, you might be able to just pull shots and stop them at about 26—30 seconds, and then if your flow rate was a bit faster than your usual goal (typically 2.2 mL/s for light roasts), grind a bit finer next time, and vice versa. Hopefully, you’ll be able to enjoy that brew even if it had a flow rate that is slightly wrong. Now, what happens if you grind WAY too coarse ? You’ll get a sudden jump to the fastest flow rate (3.5 mL/s) after the pressure peaks, accompanied by a sudden drop in pressure. In the reverse, if you went way too fine, the pressure will suddenly jump after the ~9 bar peak and flow rate may struggle to reach even the minimum 0.5 mL/s allowed by this profile. Maybe you will even reach the forsaken range above 10 bar where the shot goes to hell. But hey, what other profiles can fix shots that far from dial-in right now ?

I have only tried this on a few shots so far, and the results seem to be promising. I have only been making and tasting espresso for a bit more than a month now, so I don’t put too much trust on my espresso palate. I also wouldn’t trust my puck prep except that I received a significant amount of help from Scott Rao to the point where I think it should now be acceptable. In short, I’d now like to share this profile to see what others think about it. If there’s something wrong about the many assumptions above, then the profile will just die out.

But otherwise, maybe it will help us waste less coffee and explore more recipes, which would be lovely. I also think that, still assuming I’m not suffering from confirmation bias when tasting my shots, there might be ways in which we can improve it in the future.

The first shot I pulled of an Ethiopian coffee bean ever with the Decent was using a prototype of the adaptive profile described here. I knew Ethiopians need coarser grinds, but I decided not to move my grind size and the shot was indeed slower than usual (notice the ~1.5 mL/s flow rate). But also notice how the pressure peaked nicely at 8.6 bar and the flow rate remained perfectly constant. The shot tasted really nice for such a bad dial in, at least according to my limited experience with espresso.

Maybe a linear decrease in flow rate over the shot will taste better? Maybe current preinfusion can also be improved, as pointed out by Stéphane Ribes in one of his amazing reports available on the users-only Decent forum? Maybe we can have a version that automatically stop based on a set shot time, capturing the fact that coarser grinds seem to do better with larger ratios, weaker concentrations, and similar shot times? We’ll see!

While I was at it, I also built two more profiles that I like, both built by Scott Rao on the DE1. The blooming shot is very similar except for a long pause between the preinfusion and the extraction phase (allowing one to reach shockingly high average extraction yields), and the Rao Allongé. Yes, yes, the recipe that “might unite standard shots and Rao Allongés” has a version for either of those two, but that’s only a temporary limitation because it would be impractical to scan all the flow rates typical of espresso shots, as well as those typical of Rao Allongés (~ 4 mL/s) with the limit of 20 steps currently required by the DE1.

I’ll let you know more about what I think when I’ve tried the profiles more extensively, and I’ll also talk about how others have liked it.

You can find the three profiles I’m proposing here:

Adaptive Shot—Intended for 1:2 to 1:3 espresso

Adaptive Blooming Shot—A modified version of Scott Rao’s Blooming Shot

Adaptive Rao Allongé—Intended for 1:5 to 1:6 Rao Allongés

Please note they will probably change in the future! Also note that the profiles above are all intended to be manually stopped after about 26-40 seconds. Do not let them run in their entirety, otherwise you will get extremely long shots that will probably be weak and over-extracted.

Disclaimer: I receive no financial benefits from Decent Espresso machines, and I have no business ties to them. Decent Espresso generously offered me a 25% discount on the machine. All my impressions of the machine are therefore my own and never financially motivated (as goes with all my past posts, and all products so far).

Brewing with the Fellow Stagg [X] Dripper

Photo by Noé Aubin-Cadot.

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through the Amazon links below. I have no association to Fellow and don’t receive any benefits from posting this.

If you are a regular reader, you might have seen me writing a bit about the Fellow Stagg drippers. Six months ago, Fellow offered me their Stagg EKG 0.6L kettle for me to include it in my analysis of kettle streams, and they included an unexpected bonus in the package: their Stagg [XF] dripper. I had been brewing only V60s for more than a year at that point, and I did not really seek to try other dripper geometries. However, the Stagg immediately intrigued me with its design. For starters, it is a flat-bottom dripper, and comes with prepleated filters similar to Kalita filters, but produced by Fellow.

A photo of a prepleated filter in the Stagg [X] brewer.

As you probably know, I like to swirl the dripper between every kettle pour, and the base design of the Stagg drippers doesn’t make this easy. Fortunately, the rubber ring at the bottom of the dripper can be taken off easily, but even without it, swirling is a bit awkward because the dripper can capsize. I found that the base of my Olivewood Hario V60 worked perfectly if I kept the rubber ring on and placed the dripper on the Olivewood base:

I know the Olivewood is quite expensive especially if you are just going to use the base, but I don’t know about a cheaper option currently. I’m sure some of you could fix this with 3D printers!

Despite these minor hurdles, I was immediately pleased when I first brewed with the Stagg. My brews seemed noticeably sweeter than those I made with the V60, using the same coffee and grind size. One thing I did not like about the Stagg [XF] is that I needed to use a lot more water to pre-rinse the filters because they are a lot taller. They are so tall that not pre-rinsing them carefully or with enough water caused them to leave a slight papery taste in the cup. This is the only time I have ever noticed this with bleached filters, but they also weigh a lot more than typical pourover filters. Quite surprisingly, they did not impart a bad taste at all, but I prefer my coffee untainted. The Stagg [XF] dripper is also quite tall and narrow, and this makes it harder to control how and where your kettle stream lands on the slurry. I mentioned that to Fellow, and they immediately sent me their Stagg [X]!

Turns out not only I preferred the Stagg [X] over the [XF], I preferred it so much over the V60 that I only alternated between the two drippers for a dozen brews, and then I switched to using exclusively the Stagg [X] ever since. Six months and more than 400 brews later, I’m finally ready to write about this.

The reason why I prefer the Stagg [X] brews over the V60 is simple: I prefer the taste of the coffee it brews. As I mentioned, I find it sweeter and less sour. I think this is caused by two things: a higher slurry temperature due to the better thermal insulation, and a more even flow through the full coffee bed that leads to a more even extraction. I tested the slurry temperature with my K-type Thermoworks temperature logger, by placing 100 mL of hot water in different drippers and observing how the temperature evolves when the flow holes are blocked:

As the data show, the Stagg [X] slurry was indeed slightly warmer and more stable than the V60 slurry. I tried comparing the Stagg [X] to the V60 with the exact same parameters, namely a 25.0 grams dose, a 1:17 ratio, 4 pours with the exact same timing and flow rate, the same Hario tabless V60 filters and ColorFull Coffee Corp‘s delicious Sun Blood decaf. I used a 210°F (99°C) kettle temperature with the Plastic V60, and a 202°F (94°C) kettle temperature with the Stagg [X], as previous experiments showed that this was the way to obtain an average slurry temperature of 191°F (88°C) during the brew, in both cases. I measured the concentration of the brews using a VST refractometer and found 1.34% TDS for both brews, corresponding to an average extraction yield of exactly 19.9% in both cases (keep in mind that this is not low for a decaffeinated coffee). The V60 brew had quite a longer brew time, however, clocking in at 5:40 compared with 5:01 for the Stagg [X]. This longer brew time is not too unexpected, as the V shape of the V60 makes the same dose of coffee taller. The resulting brews tasted quite different: the V60 brew had almost no sweetness and had a bit of an ashy aftertaste probably due to some regions of the coffee bed getting over extracted, and the Stagg [X] brew was jammy, sweet and fruity. This comparison is interesting to me, because it means there is something more to just a better thermal insulation with the Stagg [X], despite the two brews having the same average extraction yield. I suspect that figuring out what coffee dose gets me the same shorter brew time with the V60 would reveal it to have a slightly lower average extraction yield.

This was good timing, because I was thinking a lot about dripper geometries to write the percolation chapter of my upcoming coffee book The Physics of Filter Coffee, and I was gradually discovering that water does not flow at all like I expected or wished for when brewing with the V60. By simultaneously measuring my pour rate and the drip rate of the V60 with two scales connected to my friend Francisco’s web app, I was able to deduce the height of the water in the dripper at every moment of the brew, as well as the hydraulic resistance of the whole dripper, filter and coffee bed system. Hydraulic resistance is a quantity that represents how much the combined effects of the dripper, filter and coffee bed are preventing the water from flowing down very fast. Without going too deep into the weeds (this is what the book will be for!), these data really convinced me without a doubt that some significant amount of water was bypassing the coffee bed, i.e. flowing around it rather than through the coffee.

This effect is usually called bypass, and has been discussed a lot in the coffee community. I was previously skeptical about it and I thought that very little water bypassed around the V60 bed, because I had never seen hard data to support the idea. I don’t blame the coffee community for that, because it is not easy to measure this effect accurately. Some recommend to always pour at the center of the slurry to prevent bypass, but this is like deciding to fight against the ocean by punching it. Regardless of where you pour water, it will find the path of least resistance and flow mostly there. Pouring at the center of the slurry and hoping to avoid bypass is thus a lot like pouring at the center of a dining plate to prevent water from spilling out. As soon as the water level gets high enough, it will flow around the plate as much as it likes. The same is true with the V60, and most other drippers. If the level of water becomes high enough that it touches some free paper filter that does not stick closely against a dripper wall, some of this water will go through the filter, regardless of where you pour.

Bypass is not necessarily a terrible thing when water flows completely around the coffee bed: it will just make your beverage more diluted. You would need to pour more water to reach the same average extraction yield compared with a brew where water does not bypass, and the resulting cup would thus be less concentrated. The part that I find desirable at all is where water can also flow through only part of the coffee bed, and then escape midway. When this happens, less water will reach the bottom of the coffee bed, and the bottom will remain less extracted than the surface layers of the coffee bed. This is where dripper geometry becomes important: a conical dripper will allow water to escape from the middle of the coffee bed, whereas a cylindrical dripper won’t allow this as easily.

You might think that a solution against this problem would be to agitate the coffee bed very deeply in a conical dripper, to make sure that the bottom is well extracted. You wouldn’t be wrong at all, but this brings out other complications: such deep agitation will allow very fine coffee powder to be dislodged more easily from the larger coffee particles, and this migration of the fines can completely clog the paper filter. Because clogging tends to leave very few spots of lower resistance in the filter, a large fraction of the water will be forced to flow there, and this uneven flow will usually cause small regions of the coffee bed to become over-extracted, and will result in harsher and more astringent brews. I suspect this might also affect brew clarity, for the same reason, although I have not verified this with measurements yet.

Using smaller pours is yet another strategy to minimize bypass. This will definitely minimize the amount of water that completely bypasses around the coffee bed, but it won’t prevent water from bypassing midway through the coffee bed. This strategy will also lower the slurry temperature, and lead to lower average extraction yields and different flavor profiles that I tend to like less.

Another solution could be to remove the ridges in a conical brewer, much like the upper half of the Kono Meimon dripper. However, this creates another issue: water will only flow through a small opening near the bottom of the dripper, which means that any coffee fines liberated with the flow will be trapped in a smaller surface area of paper filter. This will make the dripper more susceptible to clogging, as I quickly discovered when my friend Dan Eils sent me a 3D-printed ridgeless V60. If you are skeptical of this, I suggest to try an eye-opening experiment: take a glass V60 dripper that does not have a base at the bottom, and place a paper filter at the bottom of the opening outside the dripper. Use an elastic to hold it in place, then try brewing coffee with this. You will experience a surprisingly slow flow, if not an immediate clogging of the filter. This is a nice demonstration of how the total surface area of paper filter really matters for clogging. Remember that only the filter surface that is not pressed against a solid surface counts, because water cannot go through the filter efficiently elsewhere.

Placing a filter outside the bottom of a V60 will cause it to clog instantly.

In other words, the V60 sacrifices some evenness of extraction (the bottom of the coffee bed is harder to extract well), in exchange for a more easily maintained flow of water and less clogging. This is true because it has a very large free filter area compared to the volume of the coffee bed. Flat-bottom drippers face the challenge of having less free filter surface for each gram of coffee.

If you think about it, you will realize that this is also why both the Kalita and the Stagg have ribs at the bottom. They are there to lift the filter, and prevent it from collapsing into the dripper holes. If the filter does collapse, water can only flow through a minuscule surface area of free paper filter, and it will immediately clog. You might have experienced this with the Kalita: if you then just gently lift the filter, water will start flowing again because you have increased its free surface dramatically.

The Stagg dripper is definitely an improvement over the Kalita in this respect, because it has more holes and wider, more evenly distributed ridges. However, I don’t think they are deep enough: I have experienced clogging several times when I first started brewing with the Stagg. Ray Murakawa suggested placing a few whole coffee beans at the bottom of the dripper to avoid this. This solution works, but I didn’t really love it because it can easily lead to an uneven coffee bed and makes it harder to get an even flow. Scott Rao had a nice idea that turned out to work much better for me: cutting out a tea strainer mesh, and placing it at the bottom of the dripper.

Although I don’t think this is the ideal solution, it seems to be one of the best options that can be easily done at home. It completely fixed the issue of clogging for me when I used it with the prepleated Stagg filters. However, prepleated filters are also problematic in my view because they still allow for some bypass, and even probably allow for some water to escape from the middle parts of the coffee bed, although this will definitely not be as much of an issue than it is with a conical dripper. This is probably what pushed others like Scott Rao and Matt Leberman to prefer brews made with a V60 filter that is forced into the Stagg dripper after pre-wetting it.

This may look messy, but it is quite clever. It forces the filter to adhere on the walls of the dripper, thus minimizing bypass and giving you more surface for pouring and more volume for water. It’s also not the perfect solution because it will be hard to always place the filter in the exact same way, and thus may lead to different brew times, and different amounts of bypass. It also becomes important to use a tea strainer mesh, because the elimination of bypass means that the dripper will become even more susceptible to clogging. I tried to bring further improvements to the filter mesh, by poking it with a drill bit and causing it to have a “spiky” shape:

The idea behind this is to minimize the amount of contact between the mesh and the filter, and therefore increase the total free surface area of the filter. I found that it helped with flow, especially when the filter is pressed against the dripper walls. However, I did not get as many spikes as I would like, and they are not of the same exact height. This makes it a bit harder to obtain a flat coffee bed and an even flow.

If you prefer to use the Stagg filters and don’t mind the extra minute, you can place one in the dripper as usual, pre-wet it and then press each filter fold against the dripper with a finger until they all adhere to the dripper wall:

I won’t go into filter properties too much here (this will be discussed in great length in the book), but the quality of pores in the Stagg filters is not as great as those of the Hario tabless V60 filters. The V60 filters are also a lot thicker, even though the Stagg filters feel thicker because they are just more rigid. The better pores and increased thickness make the V60 filters more interesting to me, but I’m a bit bothered by the worse repeatability that is caused by the messy filter placement.

Here is the recipe that I currently follow when I brew with the Stagg [X] dripper.  I don’t think it’s the only good way to use it, but this has worked well for me over the past few months:

  • Use between 22 and 26 grams of coffee (I personally like to use 25 grams).
  • Use a coffee-to-water ratio that suits your grinder and personal preference of extraction yield versus beverage strength combination. I usually brew with 17:1.
  • I recommend making a nest shape in the dry coffee bed, and start pouring in the hole during the bloom step. This may seem counterintuitive, but I think this helps getting water through the bottom layers of the coffee bed a bit faster even with flat-bottom drippers.
  • Pour about 2.5 to 3.5 grams of water per gram of dry coffee dose for the bloom step. You want to have enough water that you can immediately give the dripper a good swirl to make the surface of the coffee bed even before all water has dripped out.
  • I use the same grind size as I would for V60, and the same pouring patterns (anything that covers the full surface, for example spirals). I also use the same pour height (just below the point where it splatters), and the same pour rate of 5-6 g/s (as described here, this depends on your kettle). Several people pointed out to me that the breakup length of a kettle is reduced when pouring hotter water. I think this is indeed true, but it doesn’t change my recommendation of pouring just below the height where it splatters.
  • I still use very hot water most of the time (210°F, 99°C). The only reason I don’t use a running boil is to avoid getting an unstable kettle stream, or even spurting. I lower my kettle temperature to 205°F (96°C) when I don’t know a roaster well, because it minimizes roastiness (burnt and bitter taste) when a coffee is roasted darker. I have gotten good results with kettle temperatures as low as 190°F (88°C) when coffee was roasted even darker, but I tend to prefer roasts that are well matched to hotter water.
  • I tend to use about 4 pours when I brew with the Stagg (not counting the bloom), rather than only two as I used to brew with the V60. The better heat insulation makes this possible, and you will be forced to do this if you use the Stagg [X] with non-folded prepleated filters because of the smaller dripper volume. I really like the brews that this produces, but it makes it a lot harder to keep track of pour sizes and the timing of pours, especially given that different coffees tend to flow at different rates. I currently start a new pour whenever the water level goes lower than about one inch above the coffee bed, and I am not particularly great at keeping track of my pour sizes, although I try to keep them around 90 grams each.
  • I swirl the dripper very gently just after every kettle pour for about half a second (I swirl the bloom more vigorously, and for 2-3 seconds).
  • I use prepleated Stagg filters as intended the first time I brew a bag of coffee, with the flat tea strainer mesh. If I don’t get very long brew times (about 3:45 or more for me, but this depends on your grinder), I will tend either press the filter folds against the dripper walls with the spiked tea strainer mesh, or use a V60 filter with the flat tea strainer mesh. Combining the V60 filter with the spiked mesh won’t help as much, because the V60 filter will collapse between the spikes anyway when it is forced inside the dripper.
  • Immediately take out the spent coffee and filter and rinse the dripper and mesh. Avoid to let the coffee steep in there because that will stain the dripper with stale coffee oils much faster.

You can find a video of an example brew I made here. This coffee was a really nice washed Ethiopian Yirgacheffe roasted by my friend Andy Kires (the roaster behind Colorfull). I used a 25.3 grams coffee dose and 430 grams of water for a 1:17 ratio, and I obtained a brew with a 1.53% TDS concentration, for an average extraction yield of 22.1%.

You might have guessed it from the brew guide above: although I prefer the taste of my Stagg [X] brews over the V60, and they tend to reach higher average extraction yields in part due to the higher achievable slurry temperature, I do not obtain the same level of repeatability as I did for the V60. As I showed in a previous post, my 2-pour V60 brews were typically repeatable within 0.2% extraction yield, but my Stagg [X] brews have a worse stability with a larger spread of about 0.3—0.4% in average extraction yield. I’m pretty sure this is mainly due to two things: it is harder to replicate the mesh and filter placement exactly, and it is harder to keep precise track of more kettle pours. This means the V60 was a great tool for experimenting, and it allowed me to generate many figures based on several hundred repeatable brews for my book. But now, I really enjoy this improved taste even at the cost of a little repeatability!

Note: Since I released this blog post on Patreon a few months ago, I have switched to using a coarser mesh at the bottom of the Stagg [X], which I made from a pasta strainer. I also made a version where I imprinted spikes by wrapping the mesh around a nail with pliers. I found that it works best with the V60 filter to ensure that you still get enough flow by minimizing the contact between even the mesh and the paper filter.

Note that a coarse but flat mesh works very well with the Stagg [X] filters too. I find this combination useful with coffees that otherwise choke the V60 filter or otherwise take a lot more than 5 minutes to brew.

I have also received several questions about how I place the V60 filters in the Stagg [X]; I now simply place the dry filter on top of the dripped (with the point of the V shape standing in at the bottom of the dripper) and then run some faucet water in the middle. The faucet pressure collapses the filter, and I help it fall down with my hand until the filter sticks to the walls well. I then press any folds to the walls to, like this:

The Effects of Varieties, Origin and Processing

Over the past year, I have logged the details of just over a thousand pour over brews that I prepared and drank. I had started doing this with several goals in mind, and one of those was to understand whether different coffee varieties behave differently when brewing pour overs. Over this period of time, I have learned how to improve my brew method to obtain more repeatable results and better quality extractions, and I also changed coffee brewing gear a couple times. I also realized several months back that the coffee concentrations (TDS) I was measuring with my reflectometer were initially not as precise as I thought, and I wrote a blog post about how to measure them better when I figured this out.

I decided to have a look at the brews where I used exactly the same coffee to water ratio (1:17), the same dripper (plastic V60), grind size, burr and grinder (the Weber Workshops EG-1 with SSP ultra low fines burrs), brew water (the Rao/Perger composition as described here) as well as method (which I describe here), with doses between 20 and 22 grams. I allowed for a small range in doses because in a conical dripper, this range of doses corresponds to a very small range of bed depth (4.65 cm to 4.80 cm, or a change of only 3%). The detailed list of all coffee gear I use can also be found here. These criteria narrowed the list down to 416 brews prepared in an identical way, but with a range of 192 distinct coffee bags.

One thing that jumps immediately when looking at the data is the wide range of average extraction yields that result from brewing different coffee beans. Even at the stages where my brews got consistent to within 0.2% in extraction yield between brews of the same beans, changing beans caused my extraction yields to vary widely across ~20% to ~23% average extraction yields. This figure shows the average and range in average extraction yields that I obtained when I was brewing coffee from a bag that mentioned each coffee variety:

I found it really interesting that the Kenyan variety SL34 comes much higher than SL28 in the figure above. We only rarely get to taste SL34 alone as it is almost always mixed with SL28, but that trend tells me that SL34 may contribute a lot more solubles because bags with both SL28 and SL34 seem to extract a lot more stuff than those with only SL28 !

As you can also see, varieties typically grown in Kenya (SL34, SL28, Ruiru 11) and Ethiopia (74110, 74112, Kurume and Wush Wush) produced the highest extraction yields, whereas those often grown in South America (Bourbon, Catuai, Pacas, Castillo) tended to yield lower ones. A particularly interesting case is that of Gesha, a variety originating from Ethiopia (as they all do in reality) that is now also grown in South America, often with a different spelling “Geisha”, and likely with a slightly different genetic makeup (Please see Getu Bekele’s excellent book “A Reference Guide to Ethiopian Coffee Varieties” for more on this). Again, the general trend where African coffees generate more solubles is still observed: Ethiopian Gesha extracts about 0.4% higher than South American Geisha on average.

These differences across origin countries can be viewed with the following graph:

Now, the explanation for these trends is unclear: it could be explained by differences in the genetics of the coffee, in how these genes get expressed across different environments, or it could also be caused by different agricultural or processing practices. I doubt that processing alone could explain the differences, as grouping the values along natural or washed coffees shows a weaker trend where washed coffees seem to extraction about 0.5% higher on average:

The possibilities that I just mentioned all assume that the different extraction yields are actually real, in the sense that African coffees bean tend to contain more soluble materials. However, there could be a different explanation; the different chemical profiles that correspond to different varietals might also be throwing off the calibrations of the VST refractometer, which uses the refraction index of the coffee and translates this into a concentration with their proprietary calibrations. I don’t know how sensitive these calibrations are on the exact chemical profile of the coffee, but I suspect there may be some variations in the sense that obtaining extremely accurate extraction yields would require using a different calibration for Ethiopian, Kenyan or Colombian coffees for example.

Probably the most interesting trend that appeared in the data is this one:

Each red circle in the figure above is a distinct bag of coffee. The symbol size is larger for points in denser regions, to give more focus on the general trend and away from lone outliers. The blue ellipses are contours of a 2D normal distribution that fits the data (at 1, 2, and 3 times the characteristic width), and the orientation of these ellipses are caused by the correlation between altitude and average extraction yield – the Pearson correlation coefficient is 25.6%, which means that the angle of the ellipses is significantly off from horizontal, as you can see.

It looks like coffees grown at higher altitudes produce higher average extraction yields ! This would favor the hypothesis that the differences are real and due at least in part to growing conditions, because there would not be any reason for this correlation between extraction yield and altitude to appear if it was purely caused by the VST refractometer calibration getting slightly thrown off by different chemical compositions. Specifically, some research by Vaast et al. (2006) and Worku et al. (2018) showed that higher-grown coffees tend to concentrate more nutrients into their seeds in harsher growing environments such as high-altitude farms. This may be an evolutionary strategy for the coffee tree to maximize the survival rate of its seeds that will face more difficulty in growing into new coffee trees.

Another effect that I find interesting is how varieties and origin countries might affect the total brew time. The reason to expect variations in brew time has to do with how brittle coffee beans are, because their brittleness will affect how much coffee fines are produced during grinding. Basically, more brittle coffees will generate more fines, and will result in longer brew times. This will likely be strongly affected by roasting style too, but the initial moisture, density and hardness of the green coffee or the bean size may all have an effect, which would introduce variations across varieties and origins.

One interesting thing about the figures above is that they confirm some well known facts about Ethiopian coffees: they seem to generate a lot more fines, and clog a lot more easily as a consequence. I often heard that Kenyans also behave this way, but my data does not seem to suggest they do, or at least much less than Ethiopians !

Differences in processing did not seem to affect brew time significantly:

But differences in growing altitude also show a strong correlation (a Pearson correlation coefficient of 36.3%), perhaps indicating that harsher environments may also affect the physical structure of the beans, potentially making the beans harder, drier and/or smaller, which will lead to more fines once they are ground:

Another possible explanation of these trends in brew time could be that it only has to do with roasting, but that these correlations were introduced because we tend to like different coffee beans roasted at different development levels on average. While this almost certainly happens to some extent (e.g., more expensive Ethiopians and Kenyans getting roasted more lightly to preserve more origin characteristics), I doubt it is the only explanation, and I suspect that either genes or growing environment also have something to do with it.

Note that I do not share my complete brew log publicly, but it is available through my Patreon page, specifically here. Keep in mind I do not share a live version even on Patreon, because it requires some data management before I share it. The current version has 636 brews and covers all of 2019, and I plan to update it with the next 400 brews in the coming month or so. 

The Physics of Kettle Streams

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through the Amazon links below. I have no association to Hanna instruments and don’t receive any benefits from them.

For a while now, I have wanted to dig deeper into the physics of kettle streams and how they affect agitation when we brew coffee. It is currently hard to find any information about the effects of pour height and pour rate on coffee brewing, and yet in my experience it can definitely have an impact on the resulting cup of coffee. It probably comes at no surprise if I tell you that pouring way too fast can dig a hollow in the coffee bed and lead to clogging a filter, or to an unevenly extracted and astringent brew. What may surprise some readers a bit more is that measuring my pour rate and pour height to keep them stable is one of the things that really helped me achieve more repeatable brews. I discussed this a bit more in previous posts about what affects brew time and the repeatability of manual pour overs.

Apart from affecting the height of water at any given moment and therefore indirectly affecting brew time, the rate at which you pour water onto your coffee slurry (i.e., the mix of water and coffee grounds) also has an important effect on how much you agitate it, and whether you end up with a hollow in the coffee bed. But this is also linked to how high you pour from, and what kettle you are using ! More specifically, the width of the gooseneck, how smooth it is at its tip, how vertical the stream is, and how fast you move the kettle around will all change the properties of the water stream and how it interacts with the slurry.

When a stream of water leaves the gooseneck, it will usually have a relatively smooth and cylindrical shape. Any tiny imperfections at the tip of the gooseneck will still be imprinted on the shape of the stream, which initially do not show much, but as the stream falls down further, its imperfections will gradually grow in size. This is due to a property of fluids called the surface tension. Molecules of water are slightly attracted to each other because they are polar, which means they have a slightly negative electric charge near the oxygen atom, and slightly positive charges near the hydrogen atom.

Polarity of a water molecule.
Network of water molecules arranged to align the negative and positive charges forming hydrogen bonds, responsible for the cohesive nature of water.

This means that water is slightly attracted to itself, a property called cohesion. It can also be similarly attracted to other materials: we call this adhesion. Water experiences very little adhesion with ambient air, and as a consequence the surface of water can seem to behave like a stretched elastic membrane, for example by forcing the water to shape itself into droplets. This effect is what we call surface tension, and it is thus simply a consequence of water having more cohesion than it has adhesion with air.

An illustration of surface tension. Water molecules near the surface of water are attracted down by forces of cohesion (black arrows), whereas they have very little adhesion pulling them upwards toward the air. This causes water to behave as if its surface was a stretched elastic membrane. Source: Wikimedia commons.

Now back to our slightly imperfect stream of water; because of water’s surface tension, the regions where the stream has a bit more curvature will experience a slightly decreased inward force, and vice versa. This is what will slowly drive the imperfections to grow until they are noticeable by eye, and further cause the stream to break into droplets. This phenomenon is called a Plateau-Rayleigh instability, and can be observed easily by pouring water with a gooseneck kettle from high up above a sink.

An illustration of the Plateau-Rayleigh effect, where an initially smooth stream of water (top) gradually becomes more sinuous and eventually breaks into droplets. Source: NIST.

The length at which the perturbations become large enough to break the stream into droplets is typically called the breakup length of a stream of water. This length is affected both by the smoothness of the gooseneck and the pour rate. As you can imagine, a rougher gooseneck will have larger initial perturbations in the stream shape, and therefore a shorter breakup length. The effect of pour rate on breakup length is however much less straightforward; in fact, for each gooseneck type there is an “optimum” pour rate where the breakup length is very long, and pouring either slower or faster will reduce it.

Illustration of flow regimes and breakup length for gradually increasing pour rates (left to right), based on a simulation by Bertola & Brenn (2019).
Illustration of breakup length as a function of pour rate, with the associated regimes. Typical pour over kettles cannot pour fast enough to reach far into the wind induced regime, or to the atomization regime.

I decided to check this relation with the Brewista Artisan 0.6L kettle I used for the past year or so, with and without the flow restrictor I described in this earlier blog post. For a while now I have suspected that the Fellow kettles actually do better than Brewista in terms of their gooseneck smoothness, and Fellow generously offered to send me their Stagg EKG 0.6L model for me to do a direct comparison. In each case, I poured water from above a tall mason jar placed on an Acaia Pearl S in flow rate mode, starting low and then raising up the kettle while keeping the same flow rate. When I started hearing a splattering sound, this indicated that the breakup length coincided with the surface of the water in the mason jar, so I stopped and measured both the absolute height of the tip of the gooseneck, as well as the height of the water. Subtracting these two values allowed me to estimate the breakup length associated with a given pour rate. By repeating this with several different pour rates, I was able to verify the predictions based on the physics papers above, as well as measure a net improvement in breakup length for the Stagg EKG over the Brewista, as I suspected !

Measured breakup length as a function of pour rate for the Stagg EKG 0.6L kettle and the Brewista Artisan 0.6L kettle, with and without a flow restrictor. Each kettle had a pour rate that maximizes breakup length at about 8 g/s, and the Stagg EKG has a longer breakup length at most pour rates due to its gooseneck being smoother. Adding a flow restrictor to the Brewista reduced its optimum pour rate, but also reduced its longest breakup length.

Understanding breakup length is important because the smoothness of the water stream when it lands on the slurry will completely change how much, and how deeply the slurry gets agitated ! Engineers have already figured out these details for the construction of plunging jet reactors, which aim to maximize the rate of chemical reactions between a gas and fluid with a stream of fluid falling in a still pool and entraining some gas bubbles with it (for example, see Yin et al. 2018).

Illustration of a plunging jet reactor, where gas bubbles are entrained in a pool of liquid with a stream of falling fluid, to maximize the surface of contact between the gas and liquid and optimize their rate of chemical reactions.

Basically, the rougher a stream is where it lands on the slurry, the more air bubbles will be transported in the slurry (see Baylar et al. 2016). The air bubbles will cause friction with the flow of water below the surface, as the bubbles themselves are slowed down by buoyancy forces pushing them upwards. At the depth where water has slowed down enough, only buoyancy remains as an acting force on the air bubbles and they will rise back up to the surface. This rapid breaking of the underwater current can cause the flow to become turbulent, which is a good thing for coffee extraction because turbulence is great at distributing fresh water between all particles of coffee, and make sure that no one particle remains under-extracted.

If the stream of water is extremely irregular where it lands however, a lot of air bubbles will be dragged into the slurry and the breaking will happen so fast that agitation will not reach very deep down, potentially leaving the coffee bed undisturbed. At its most extreme, pouring from higher than the breakup length will cause only tiny droplets of water to fall on the slurry, and no agitation at all will be able to reach the coffee bed and make extraction more even.

Illustration of turbulent flow in the context of Rayleigh-Taylor perturbations, as simulated by Chiel van Heerwaarden

If we take the opposite approach and pour water at a fast rate by holding the kettle just above the surface of the slurry, the stream will be very smooth while it enters the slurry and almost no air will be dragged into it. This means that the underwater current will not slow down as much before it reaches the coffee bed, and you will instead slowly dig a hollow in the coffee bed using a laminar flow (i.e., a “not turbulent” flow). This is not very efficient at separating all coffee particles to extract them evenly, and the hollow will potentially later cause a channel and local over-extraction unless you swirl the dripper to make the coffee bed flat again.

Now that we saw how breakup length affects agitation, it should be clearer that in order to obtain a turbulent agitation of the coffee bed, we need the stream of water to be a little rough as it enters the slurry, but not too much. To little will result in non-turbulent flows, and too much will only agitate the surface of the slurry. In practice, I found best results by pouring from the height that will place your breakup length just below the surface of the slurry. You can achieve this by pouring just a bit lower than the point where you can hear a splattering sound as the stream enters the slurry. In other words, you do not want to hear that splattering sound, but you want to be pouring just below the point where you would hear it.

Illustration of different types of agitation imparted to the coffee slurry by different shapes of water streams.

Depending on how fast you are pouring and what kettle you are using, this pour height that almost causes splattering sounds will be different, because the breaking length of your stream will be different. These effects were explored by Qu et al. (2011) and Baylar et al. (2016) in the context of plunging jet reactors, and are illustrated below:

Depth of agitation and bubble entrainment as a function of pour rate. Choosing a pour rate that falls just below the slurry will maximize agitation; higher pours will cause only shallow agitation, and lower pours will cause laminar currents in the slurry. Based on experimental data from Qu et al. (2011).
Depth of agitation and bubble entrainment as a function of nozzle roughness (i.e., gooseneck roughness in the context of coffee brewing) and the impact velocity of water. Based on experimental data from Baylar et al. (2016).

The longer your breakup length is, the more velocity your stream of water will have as it enters the slurry, because it had more time to be accelerated by gravity. This means that selecting a longer breakup length will cause stronger and deeper agitation, and will agitate your coffee bed more. Unfortunately, I do not think there is a “best answer” when it comes to what should be your breakup length. If you have a breakup length that is very long, and end up pouring from very high up (without splatter !) as a result, you might impart so much agitation to the coffee bed that you could end up clogging your filter because very fine coffee particles will more easily fall at the bottom.

In my experience, I like using a breakup length of about 18-19 cm, which I achieve by pouring at 5 g/s with my Stagg EKG 0.6L kettle, and pour from a height just below that which causes splattering sounds. This works very well with my current set-up (detailed here on my blog), but you may want to select shorter break-up lengths if your grinder or coffee generates a lot of fines. I sometimes reduce my pour rate to 3-4 g/s with some Ethiopian coffees that clog easily, and crank it up to 6-7 g/s with some naturals or South Americans that have insanely fast drawdown times. I recommend experimenting with your own kettle and determine the pour rate that suits you best.

There is another detail we did not discuss that will affect the breakup length of your kettle. How vertical your falling stream of water is important – generally, I found that the most vertical streams are easier to control, preserve a good breakup length, and cause less hollows in the coffee bed. To make your stream as vertical as possible, I recommend moving your kettle relatively slowly to leave the stream undisturbed, and keep it at a relatively stable height during your brew. One nice advantage of having a high quality kettle is that they are designed to minimize the teapot effect. This is another phenomenon caused by the surface tension of water, causing it to want to stick to the outer wall of the kettle, and make the stream of water less vertical especially at slow pour rates.

Illustration of the teapot effect, causing the stream of water to bend towards the kettle especially at slow pour rates. A hydrophobic coating completely eliminates the adhesion of water to the gooseneck surface and therefore eliminates the teapot effect completely. Source: phys.org.

Now I know we discussed a lot of different things in this blog post, and it can be a little hard to keep all these variables in mind when you try to apply these concepts to your brewing method. To help you a little, I created some 2-dimensional maps of how much agitation you will cause in the slurry, and whether that agitation is turbulent or laminar. Your exact map will depend on the kettle you are using, but this should give you a good idea of how to explore your options. As I mentioned, I personally found the best results by maximizing turbulence by pouring just below the splattering point, and finding the flow rate that causes the right amount of agitation to avoid clogging and maximize evenness of extraction.

Illustration of an agitation map based on different pour heights and pour rates. I normally aim for the “sweet spot” where agitation is turbulent and significant, i.e. when pour height is just below the breakup height that would cause a splattering sound. I select the highest pour rate that does not result in filter clogging, typically 5 g/s.
Illustration of my pour strategy placed on top of the previous agitation map. I normally start at 5 g/s, and if a given coffee draws down very fast because it has a small amount of fines, I go to faster pour rates and pour from slightly higher to maintain a breakup length just below the surface of the slurry. If on the other hand a coffee generates an unusual amount of fines, I will pour slightly slower, and adjust the pour height lower so to avoid the splattering sound.

I would like to thank Fellow who generously sent me a Stagg EKG 0.6L gooseneck kettle for free so that I could compare its performance with the Brewista artisan. This gift came without any obligation on my part regarding the content of this or further blog posts.

Why Can’t we Grind Coffee Finer for Pour Over ?

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through the Amazon links below. I have no association to Hanna instruments and don’t receive any benefits from them.

When I started reading about the craft of pour over coffee, I learned that it’s important not to grind too fine, because past a certain point the coffee tastes harsh and astringent (i.e., a drying sensation of the tongue). This was very much in line with my experience, and on top of that, grinding too fine would result in a completely clogged V60 filter that would make it impossible to get all the water through without absurdly long brew times. For these reasons, I didn’t question the concept and I was happy to grind as coarse as everyone else.

The Barista Hustle cupping experiment, showing how average extraction yield evolves with time with fine (green), medium (orange) and coarse (blue) coffee grounds.

The first thing that made be re-think this was an experiment performed by Barista Hustle, where they measured the extraction yield versus time for different cuppings made with narrow particle size distributions of different sizes, which they obtained by sifting ground coffee. If you have not already watched this video, you should definitely take a look; it provides quite compelling evidence that we are never really extracting the core of particles larger than about 100 microns within reasonable brew times and water-to-coffee ratios. To convince myself that their experiment was consistent with the idea of under-extracted particle cores, I created a simple model and fitted it to their data and found that water only seemed to extract down to a characteristic depth of about 40 microns, i.e., this is the depth where extraction is about 60% slower than surface extraction. This means a very large fraction of the coffee mass is almost never extracted at typical filter grind size !

Just the idea that we are wasting that much coffee is already not great, but I think the problem goes even further; as I argued in a previous post, the cells that are hidden deep under the surface of coffee particles still extract but slower; this will contribute flavors to your cup of coffee that have a very different profile compared to the cells near the particle surface. In other words, we are not doing a great job of extracting all the coffee cells evenly when we brew pour over. Does it mean even extractions will necessarily taste better ? No, but I think we should try it and see what happens.

Following this, Matt Perger was shouting from the rooftops about the virtues of grinding finer, and there was even talk of using a chemistry Büchner funnel for filter coffee. I thought he potentially had a point, and I became very interested in the idea of getting past the astringency limit. I started experimenting with Turkish ground coffee brewed on the siphon with limited success; it only worked well with a couple coffee beans. I also tried the Büchner funnel; my friends Matt Leberman and Mitch Hale seemed to like it, but I never managed to brew something that I did not find very astringent. Dan Eils also sent me a nice 3D printed V-shaped vacuum brewer, which he called the Vac60, but I was still unable to brew coffee free of astringency if I used it with very finely ground coffee.

Dan Eils’ Vac60 brewer.

I thought maybe agitation was the problem, so I tried using the Melodrip with finely ground coffee and a couple different brew techniques, ranging from my usual two pours to Ray Murakawa’s full two stirred blooms, multi-pour method. Again, I always found the results too astringent, and worse than what I was obtaining with a coarser grind size. These poor results eventually pushed me away from experimenting with very finely ground coffee.

The Melodrip, photo by Elika Liftee.

As I started to suspect I was not gonna be able to obtain great brews with the methods above, I decided to instead improve my grinder’s particle size distribution as much as I could. After some research, this is what led me to purchase a Weber Workshop EG-1 with the ultra low fines SSP burrs. Using my own imaging software, I verified that it indeed produced the narrowest particle size distribution I had seen amongst all the grinders I tested, ranging from several hand grinders to my previous Baratza Forté and even to industry standard Titus-aligned Mahlkönig EK43 grinders. I expected that dialing in the grinder would allow me to push my extraction yields very high, and to grind extremely fine.

The SSP Ultra low fines burrs.
Filter coffee grind size with the SSP ultra low fines burrs. Fines are nowhere to be seen !

To my surprise, this is not exactly what happened; I gained about 1% in average extraction yield compared to the Forté, but my brews suddenly tasted much cleaner, and the tasting notes were a lot more obvious, much like what you get from cupping. I attributed this to the quasi absence of fines produced by the SSP burrs that I installed on it (I still think this is the explanation). I was barely able to choke my V60 brews, even if I ground my coffee insanely fine, yet there was still a definitive limit in grind size below which my coffee tasted very astringent. Even more confusing, I did not seem to be observing a decrease in average extraction yield when I ground fine enough for my brews to taste bad, as I kept hearing should happen. I attributed this to my poor skills at brewing consistently, or measuring my coffee concentration accurately; I still think it was wiser not to trust myself too much back then, especially given that I struggled to brew two V60s in a row with brew times within 20 seconds or concentrations within 0.05% total dissolved solids (TDS) of each other.

The coffee station I assembled to brew consistent V60s.

So I spent the next several months significantly upping my consistency game. I insulated my kettle better with a Brewcoat (see my previous post about this); I found an obscure flow restrictor only available in Asia; I started logging every detail of my brews with two scales and a brew stand to record complete brew curves and see my kettle pour rates in real time; I significantly tightened my refractometer measurement procedure; I improved my measurements of brew water composition (see this post); and I even installed a measuring tape on my wall to always pour from the same height and impart the same amount of agitation to the V60 slurry. I also started weighing my dose with a precision of 0.01 grams for good measure; the effect of a 0.1 grams dose accuracy on calculated extraction yields is about the same as that of a 0.01% TDS measurement accuracy.

Consistent V60 brews across days.
My kettle pours for the consistent brews.

I don’t think measuring my coffee dose to 0.01 grams had a big impact on my repeatability, but it elucidated something that had been bothering me: sometimes my dose suddenly weighed 0.1 or even 0.2 grams less if I was very slow at preparing it. I now know that this is caused by water evaporating from the wet V60 filter, which means you will be mis-calculating your doses slightly if you take too much time to weigh your dose and compensate for that loss. I now take about 5 to 10 seconds to weigh my dose within 0.01 grams, which is fast enough to avoid any issue with evaporating water.

About 8 months ago, I passed the point where I was getting brews consistent to within about 0.4% in average extraction yield, which corresponds to differences of about 0.02% in TDS concentrations with my typical brew parameters. But only very lately, I finally passed the point where I am able to replicate brews to within 0.2% in average extraction yield across five days. Now, the fun is really starting because I can finally start to investigate how many variables influence extraction yield in a much more accurate way.

Just as I was getting these results, a new science paper about espresso rocked the media, which essentially argued that most third wave espressos are brewed with a very uneven flow, and that we might want to consider grinding espresso coarser for a more even extraction. That’s nice and all, but I don’t really drink espresso. I was still excited when I saw this figure:

How extraction yield changes with grind size for espresso brews. Figure from Cameron et al. (2019).

This shows exactly the “common knowledge” curve I was talking about earlier; if you grind too fine, average extraction yield goes down because your extraction becomes more uneven. As I had just achieved the kind of precision that allowed me to do this, I decided to build a similar curve for filter coffee. I brewed seven V60s in a row using the same coffee and grind settings ranging from 3.0 to 9.0 on the EG-1 and tried to replicate my brew technique as well as I could. Here’s what my kettle pours looked like across the six brews;

My kettle pours compared during the grind size experiment.

Not too bad, except for one brew (grind size 6.0) where my first kettle pour was a bit ahead. More interestingly, here are the TDS normalized by beverage weight tat I obtained. They are simply the average extraction yields divided by 14.4, as I normalized my results to a 20.0 grams dose and a 288 grams beverage weight:

How my beverage concentration (normalized to the same beverage weight) changed with grind size on the EG-1.

This result actually confirmed the impression I had even a couple weeks after obtaining the EG-1; I am not brewing near the maximum average extraction yield… not at all ! A blind tasting revealed the best brew in this case was at grind size 8.0, and all those high-extraction brews at grind sizes below 6 did not taste great at all !

I do not know whether this would happen with different grinders, or if this is a weird feature about a very uniform particle size distribution. Regardless, I think it might be further indication that I’m not using my grinder to its full potential. One hypothesis that immediately jumped to my mind was that the deviation from a linear trend might indicate uneven flow. The mathematical model of Cameron et al. (2019) based on the diffusion of chemical compounds in a two-sized bed of coffee particles predicted a linear trend for an espresso brewed with even flow.

I could very well be wrong about this, but I don’t see an obvious reason for a similar model to produce a very different behavior at coarser grind size. The difference between filter and espresso brews might be that the regime over which flow becomes uneven happens on a very short range of grind sizes for espresso, whereas for filter coffee it is happening over a gigantic range. Because of this, my figure may be showing an increase of extraction yield that is due to the significant increase in available coffee particles surface as I was grinding finer. Thus, I thought the deviation from a linear trend may tell us about flow evenness:

An estimation of how uneven my extraction was if I assume extraction yield should go up linearly as I grind finer.

But then I sat on this for a while, and thought more about it. Another hypothesis came to my mind, this one much simpler. What if uneven flow is not the main culprit for astringency when I grind just a bit too fine (say, at 5.0) ? What if I am just locked into this grind size restriction because of my a priori hypothesis that I should brew with a 1:17, or even 1:16 coffee-to-water ratio ?

When I brew tea (I sometimes do !), I can make almost anything astringent if I wait and steep it for too long. I think this might also be true of coffee, except for some unicorn exquisitely roasted high-quality green coffee (i.e., those coffees that tasted great with my Turkish siphon recipe). Only, percolating coffee is much more complicated because we don’t directly control steep time; we are under the tyrannic reign of water flow, unless we cut the brew with water still in the V60 – which would feel so wrong.

Locking my ratio to 1:17 means that I am using a fixed amount of solvent to extract flavors from the coffee particles. It is a relatively high ratio, which means that I am extracting aggressively fast. When I grind at my usual setting, I might be over-extracting just a small fraction of the coffee cells near the outermost surface of coffee particles, contributing a small amount of chemical compounds that cause astringency, but not enough for me to really taste that. But as I grind finer, this fraction of exposed coffee cells goes up, while my flow of water gradually goes down, causing an even more aggressive total extraction which leads to astringency.

In other words, the fine grind size is not necessarily the unique source of the problem; it might be the combination of long brew time and finer grind size that leads to a worse cup of coffee. So here’s an idea; what happens if we use less water, and then bypass the resulting cup of coffee with fresh water to obtain the preferred concentration ? This is very similar to what Scott Rao suggested in his books when dealing with very large doses in batch brewers; we end up not having enough flow, so we have to limit brew time or grind absurdly coarse. Hence the suggestion to use a smaller ratio and then bypass the result with brew water to reach the desired beverage concentration. We might want to try that even for single-cup pour over coffee, or at least when we are using a grinder that produces very little fines. If you grind so fine that your dry coffee grounds start to clump, or your filter clogs, we should not expect an improved taste regardless of the brew ratio ! It is also possible that fines will migrate into your cup, muddying the flavors if there are too many of them.

It is not guaranteed that the finer grind size will make up for the smaller amount of water; maybe our average extraction yields will go down or the brews will not taste as good, but I have started trying this out on a few coffees and my (very) preliminary results seem positive. I will definitely start investigating this more, and see how fine I can get without hitting astringency, and I’ll report back with some results ! Maybe I’ll finally get a Büchner brew I like, only with a ratio as small as 1:10 ?

I’d like to thank Matt Perger, Scott Rao and Chris Hendon for useful discussions.

Measuring Brew Water Properties

Photo header credit: Kathy Gagné

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through the Amazon links below. I have no association to Hanna instruments, API or Red Fish and don’t receive any benefits from them.

Most of the questions I receive about coffee brewing are related to the craft of brew water, especially when it comes to verifying that its properties are consistent with one’s goal. If you are not familiar with how brew water affects your coffee, I highly recommend reading this previous post I wrote on the subject before you read what follows. I will assume that the readers are familiar with the concepts of alkaline buffer and total hardness, both of which seem to have the most impact on taste. If you have the chance to grab one of the elusive copies of the Water for Coffee book, I also highly recommend it if you would like to learn more about water.

Both my blog and the Barista Hustle website contain various useful tools (e.g., here and here) when it comes to selecting your brew water recipe and crafting it, but here I want to focus on how we can precisely measure the total hardness and alkaline buffer of any water. This can be useful to verify your water recipe came out good, or to check whether your tap water is good for coffee. If you have tap water soft enough to add minerals until you reach a desired composition, this is yet another situation where you will want to know the properties of your original water precisely.

A conductivity meter used to estimate the total dissolved solids in water.

In the past, a lot of focus has been placed on measuring the approximate total dissolved solids in water using electrical conductivity meters like this one. These devices only measure the electrical conductivity (EC) of water as their name suggests, and then try to guess the total hardness of water assuming a specific composition. In general they are not very accurate, but changes in EC measurements can be useful to quickly notice a change in water composition. Most of them claim to apply an automated correction for temperature, required because the temperature of water also affects its conductivity. Unfortunately, I have come to stop trusting any of these claims, and I suggest making sure you always measure your water EC at a fixed temperature (25°C, i.e. 77°F, is typical).

As EC meters are not very accurate, it is natural to wonder if there are better methods available to diagnose your water, and there are ! It is much more useful to directly measure the alkaline buffer (KH*) and total hardness (GH) of your water. If you don’t want to measure both, I recommend going with KH because a bad buffer can really destroy a good cup of coffee. After looking for a while, I found two sets of tools to measure KH and GH, both based on the concept of titration. We will go deep in the rabbit hole of what titration means further down, but for now let’s just focus on the titration tools I found.

*KH is not always equal to the alkaline buffer, but for most water compositions they are the same (put simply, as long as GH is larger than the alkaline buffer we can call the latter KH).

Colorimeters and Photometers

A Hanna Instruments HI-775 colorimeter for total alkalinity.

Those are definitely the nicest options if you are able to find them and are willing to spend a couple hundred dollars for your measurements. You only have to add some reactant to a sample of your water, place it in a small colorimeter or photometer device, and you will get the composition of your water as a number on a screen. It is almost as easy as using a refractometer to measure coffee concentration. With typical water recipes, they can reach precisions of about 5-6 ppm as CaCO3 in both GH and KH. This is a very good precision, and should be better than what we can actually taste in the resulting coffee. It’s also very hard to make a human mistake during the measurements, which is not the case with the other tools described below. There are some down sides however; they are usually hard to find and you will need to order reactant chemicals every once in a while. Those that I could find are manufactured by Hanna Instruments. The most relevant ones are:

  • HI-775 colorimeter for KH measurements*.
  • HI96735 photometer for general hardness (expensive !).

*Make sure to use the freshwater model, not the saline water model.

If you get tempted by the calcium and magnesium hardness colorimeters like I did (HI-719 and HI-720), don’t get them because their range is only useful for extremely soft water. I personally use the HI-775 for quick KH measurements. It provides measurements in units of mg/L HCO3-, which you need to divide by 1.219 to obtain a measurement in our typical units ppm as CaCO3.

A KH colorimeter device (left), next to a sample of water after adding reactant (middle) and a cuvette of clean water for calibration (right).

Aquarium Titration Kits

The other cheaper and more widely available option is to go to your local aquarium store and buy titration kits for GH and KH, which are sometimes sold as a single package. The big issue with these kits is that the smallest increments they can measure by default in both GH and KH is about 18 ppm as CaCO3. Maybe this is fine for aquariums, but it’s a bit sloppy for coffee brewing because we generally like to use soft water (for example, a lot of popular recipes have KH at ~40 ppm as CaCO3). They are a bit more difficult to use and prone to human error compared to colorimeters, but they can definitely do a much better job than EC meters.

These kits come with a user manual explaining how to perform the measurement; you usually need to put 5 mL of water in a small cuvette, add a single drop of reagent, shake it, then look at the color. Depending on the brand and type of titration kit you are using, different colors will indicate that you are done with your measurement. For the popular API brand, the GH kits go from orange to dark green and the KH kits go from blue to green and then yellow. Once you reach the color turning point, you can look up a table in the user manual to convert the number of reactant drops to a GH or KH measurement – typically, every drop counts for 1 german degree, or 17.85 ppm as CaCO3.

What is Titration ?

Titration is a method widely used in chemistry to measure the concentration of a specific compound in a solution that we’ll call the test solution. The idea is to use another solution, called a reactant, which has a known concentration in another chemical agent that can react with the chemical we want to measure in the test solution. That reactant also contains a color indicator that will change color quickly if none of the compound to be measured is left in the test solution.

Then, the idea is to slowly add a precisely measured volume of reactant in a known volume of test solution, until the compound we want to measure has all reacted away, at which point the color indicator will suddenly cause a change of color. You can then do some maths to calculate how much of the original chemical you reacted away before none was left, because you know exactly how much of the reactant you added.

A titration curve (red line) with its turning point (vertical dashed blue line) and the zone where the color indicator shifts (green to yellow shaded region).

What I call a “trigger” in the figure above can be the pH of the solution, or other technical properties I won’t get into. The important point is that the color indicator is sensitive to that triggering property. In the example image above, the chemical that serves as a color indicator is sensitive early in the transition, and this indicates that the best measurement will be obtained by using the reactant volume corresponding to when the color transition is completed. Some other reactants and indicators would however have the color transition happen right in the middle of the turning point, which would make the color transition so fast that there would be almost no ambiguity, and therefore the titration would be more precise.

In a laboratory environment, chemists will use very precise tools like burettes in a very clean and controlled environment, allowing them to reach very accurate measurements with precisions as good as about 1% of the measurement. We should not expect to be able to do this with aquarium tests, which are way less precise than volumetric tools. If you’d like to learn more of the technical details behind titration, I highly recommend this website.

What I found about the API GH titration tests is that they use a reactant called an EDTA tetrasodium salt, a large and complicated molecule that can capture calcium and magnesium ions, and triethanolamine, a buffering component that keeps the pH very high to make the transition point very steep. I wasn’t able to find out what color indicator they used however, which makes it unclear where exactly the color transition takes place with respect to the turning point. The KH titration test information is even more sparse; my feeling is that API only publishes as much information as they have to for safety measures, and they keep the rest as intellectual property. The best we can do in that situation is follow their instructions, which don’t clearly specify whether the correct measurement happens at the start or the end of the color transition.

Stretched Titration with Aquarium Kits

One of the most straightforward ways to obtain a better precision with an aquarium titration kit is to use a larger sample of water, and put more drops of reactant in it. You will obviously need a clear container larger than what is provided by API kits for this. Don’t use any dining glassware, as the reactants are not safe for consumption ! If you double your water sample size, then each drop of reactant will only represent a measurement of half a German degree (i.e., ~9 ppm as CaCO3). This is already much better, but it can quickly require a lot of reactant especially with harder water. I built a Google spreadsheet to help you determine your measurement and precision with this method. If you are pretty sure that you added the exact number of drops required for the color transition, leave the “Precision for number of drops” to one, but if you are not entirely sure which one of the last two or three drops caused it, you can use errors of two or three drops accordingly; the spreadsheet will include this in your final KH or GH measurement error.

I purposely set a maximum achievable precision of 5% of the measurement and 5 ppm as CaCO3 in the spreadsheet (this is true of the other methods below as well) – I don’t think it is likely we will get better measurements in a non-laboratory environment and with the specific chemicals that the API aquarium kits use for titration. You would hit a couple systematic problems if you tried to obtain a better accuracy. For example, the titration curves of the API kit have a smooth part near the transition point which will make the point where the color changes less obvious in those situations.

If I did not put these limits on precision, you may use a large sample and think you have measured KH with a 0.1% precision, when in reality you may be 5% off. Another example of a systematic error is the fact that magnesium reacts slightly differently than calcium with the API test kits, and because they do not know your water’s Mg/Ca ratio, they use the properties of calcium only to translate your reactant volume into a GH measurement. Other problems could include using a cuvette that is not perfectly clean or dry, not having shaken the reactant properly, having left water sample to evaporate a bit, or even having lost a couple milligrams of your sample when you shook it. As you can imagine, those are all things you should try to avoid !

Here are step-by-step instructions for a stretched titration:

  1. Thoroughly clean and dry the glass cuvette.
  2. Wear clean and dry protective gloves.
  3. Measure a precise volume of brew water (API’s default is 5 mL).
  4. Shake the reactant thoroughly and open it.
  5. Add a single drop of reactant and stir/shake well (depending on your cuvette).
  6. Look at the solution color in front of a white wall in a well-lit environment.
  7. Keep going until you see a color shift happen, and make sure you don’t lose the count of reactant drops.
  8. Stop counting when the full solution color has noticeably shifted even after you stir/shake the cuvette for a dozen seconds.

Back Titration with Aquarium Kits

There is one neat trick called back titration that we can use to reach a better precision without needing all the additional reactant. The idea is to put a fixed, small number of reactant drops in the cuvette, and then add your brew water one drop at a time until the reverse color shift happens. Because we reversed things, now a KH kit would go from yellow to blue when you reach the transition point. Translating your experiment into a measurement is a bit more complicated in this scenario, so I also included a back titration calculator in the Google spreadsheet to help you doing it. The more drops of reactant you started with, the better precision you will obtain, up to a limit. However, you will also need to add a lot more drops of brew water especially when measuring soft water, and it can become easy to miscount them or place a drop outside or on the sides of the cuvette if you are not focused enough.

One thing you will need to do before you can use this method is calibrate your particular pipette with a milligram-precision scale like this one so that the spreadsheet knows the average weight of one of its water drops. There is a space just for that in the Google spreadsheet; simply fill it up with the weight of 25 drops obtained with your pipette in milligrams, and the spreadsheet will do the rest. Otherwise, you could use the exact same pipette as I used (sold by Ronyes Lifescience) and leave the default values in the spreadsheet. Make sure you put the rubber cap on the size with the larger hole.

Here are step-by-step instructions for a back titration:

  1. Thoroughly clean and dry the glass cuvette.
  2. Wear clean and dry protective gloves.
  3. Shake the reactant thoroughly and open it.
  4. Put a single drop of reactant in the cuvette.
  5. Add a single drop of water and stir/shake well (depending on your cuvette).
  6. Look at the solution color in front of a white wall in a well-lit environment.
  7. Keep going until you see a color shift start to happen, and make sure you don’t lose the count of water drops.
  8. Stop counting when you first see signs that the color of the full solution has started shifting noticeably. It will keep getting more obvious if you add more water, but use the moment where you first noticed it clearly as your number of drops.
  9. If you went past the color shift or want a better precision, add another drop of reactant and then keep adding water again. You will just need to update the number of reactant drops in the calculator, and don’t forget to always count the total drops of water that you added since the very start.

Scale-Assisted Back Titration with Aquarium Kits

As it can be hard and annoying to count many drops of water, I thought of a better way to do this, but it requires using a milligram-precision scale. The idea is to use weight instead of volumetric measurements, and translate those back into GH and KH using known properties of the reactant and water. Having only access to API-brand kits, I made the required measurements for their specific types of reactants only.

To perform a scale-assisted back titration, first place your titration cuvette on the scale and note the total weight. Then add the desired number of reactant drops (remember that more drops will get you more precision), and note down the updated weight. Now start adding drops of your brew water one at a time. You can pick up the cuvette, shake it, and place it in front of a well-lit white wall to make things easier. I recommend noting the total weight at every step (without taring) in case your scale turns off automatically when you take too long. Put the cuvette back on the scale when you reached the color turning point, and note down the water weight that you needed. From these measurements you can deduce the total weights of water and reactant, and enter them in the spreadsheet to obtain your KH or GH measurement, with its measurement errors.

I highly recommend using this method at a room temperature in the range 18-25°C, because otherwise the mass density of the reactant drops and water could change too much, which would falsify the mass to volume relations that I hard coded into the spreadsheet. This method is a lot more robust against human error compared to a regular back titration. The step-by-step instructions are the same as a normal back titration; you just need to keep track of total water and reactant weights as you go.

Testing the Different Methods

I decided to measure my Rao/Perger brew water with each method to verify that the expected accuracies were reached. Here are visualizations of each measurement with its error bars obtained with the web tool, and compared to the expected value for the water recipe I used, shown as a vertical blue dashed line (40.5 ppm as CaCO3).

Measurements of total alkalinity made with the different approaches explained here (red circles with measurement errors), compared to the expected KH of the brew water recipe I crafted (vertical blue dashed line).

When to Use Which Method

In order to illustrate when each method is preferable, you need to have an idea of whether your water is very hard or soft, and what type of precision you want to achieve. To help you decide, I built a figure illustrating the fractional error you would get for KH measurements with the different methods mentioned here, depending on your water’s total alkalinity:

Fractional measurement errors  in total alkalinity for different methods discussed here.

The case for GH is very similar so I didn’t show a separate figure for it. As you can see, stretching the API titration kit to a 20 mL sample instead of 5 mL will easily get you a very good precision, but it can be very costly in terms of reagent, especially with harder water. To illustrate this, let’s look at how much reagent a stretched normal titration requires compared with back titration:

Number of reagent drops required to reach a fractional error of 10% (black)  or 20% (red) with the stretched titrations (full lines) versus back titrations (dashed lines). The back titrations require a lot less reactant.

Here we can really see the strength of back titration: it requires way less reactant to reach similar precisions. This is especially true in the case of hard water, where a lot of reactant can be otherwise needed. If you like to see the absolute error for different methods rather than the fractional errors, then I’ve made an alternate figure for you:

Red Sea Titration Test

After I wrote this post, a friend of mine Victor Malherbe told me about another nice option: the Red Sea Fish titration tests. They have titration kits for total alkalinity, calcium and magnesium hardness with the following precisions; they seem especially interesting for their KH precision:

  • KH: 2.5 ppm as CaCO3 (i.e., 0.5 meq/L),
  • Calcium hardness: 12.5 ppm as CaCO3 (i.e., 5 mg/L),
  • Magnesium hardness: 82.4 ppm as CaCO3 (i.e., 20 mg/L).

As a reminder, general hardness (GH) is the sum of calcium and magnesium hardness. I suspect their calcium and magnesium hardness tests could also be stretched and reversed !

Hopefully this post will make water measurement simpler and clearer for some of you !

The Physics of Fines Migration

Jesse Lambeth, a member of my Telegram channel, recently drew my attention to a really interesting physics paper by Knight, Jaeger & Nagel that I now think may be relevant to our understanding of fines migration in coffee brewing. There are some considerations that could complicate its application to coffee, but I’ll come back to those at the end of the post.

First, let me summarize my view of fines migration before I had read that paper. You might remember that when we discuss the size distribution of coffee particles, we often call the smallest ones fines and the largest ones boulders. When we use the word migration in this context, we usually refer to fines preferentially moving towards the bottom of the coffee bed during a brew. This is also sometimes called the Brazil nut effect because fragments will tend to accumulate at the bottom of a container where dry food is stored.

The Brazil nut effect observed in a jar of cat food. The smaller fragments tend to accumulate at the bottom of the jar.

This phenomenon of fines moving toward the bottom of the coffee bed is mostly relevant to percolation methods, i.e.those where we pour water on top the coffee and use the grounds themselves as part of the filtration system. We worry about this because the accumulation of fines near the filter that holds the coffee bed together might get enough of its pores blocked to cause clogging. When a filter clogs, it will slow down the flow of water in a poorly controlled way, and this will also result in a less evenly distributed flow and therefore extraction through the coffee bed. If you’re interested to read more about this, see this previous post I wrote on the subject.

My previous understanding of how fines migration occurred was that, in the presence of vibrations strong enough to lift coffee boulders, some larger gaps between them would appear momentarily and allow fines to fall through toward the bottom. This video shows this mechanism in action:

While it can still happen, the key finding of the Knight paper is that this effect is not required for fines migration to occur ! Even if the vibrations are too weak to lift the boulders or cause them to jump around, a whole different mechanism can still cause the fines to clog the pores of your coffee filter, and its details are incredibly interesting.

To figure this out, the team of scientists led by Knight placed identically shaped glass balls in a cylindrical container, and added a single larger ball with the same mass density as the other ones. They painted the large ball and a couple of small ones with a bright color so that they could easily track their displacements, and they attached a device to the container that can imitate a short tap in a very controllable way. This way, they could artificially tap the container as many times as they like, always with the same exact force and duration. All they then had to do is turn on the tapping and observe how the colored balls moved around. To the scientists’ surprise, no ball actually needed to be lifted for anything to happen. Instead, the balls near the edges of the container were pushed down by the frictional forces of the walls vibrating against the balls, which started a cyclic flow in the whole container that looks very much like convection in a hot liquid.

The image above displays a temporal sequence of motion in a cylindrical container as it is tapped many times. Some of the balls were colored to keep track of their motion. The balls against the walls flow downward, and push the central particles upward, starting a convection-like cyclical flow. Source: Knight et al. (1993), modified.

As you can see in the image above, the glass balls that are incoming from above are pushing the bottom layer of balls toward the center of the container, and those are in turn pushing the bottom center balls upward. When they reach the surface, they migrate outward to the container walls, and that completes the cycle where they get pushed down again by the vibration.

In nature, convection can be observed in all kinds of places where a fluid is quickly heated; at the surface of the Sun, in a boiling pot of water, in a hot Miso soup especially just after you remove its lid, or even in a small cup of tea in a cool environment with the right lighting. I don’t think the motion of the balls can be called proper convection, but they sure move in a very similar way.

The image above shows convection cells at the surface of the Sun. Large bubbles of (brighter) hot gas flow upward and thin layers of (fainter) cool gas flow downward between the bubbles. The cyclical nature of this flow is similar to what is observed in a tapped cylinder of glass balls. Source: University of Wisconsin, Madison.

In the image above, you can see convection taking place at the surface of the Sun; the cells of upward-moving hot fluid and the interstices of slightly cooler, downward-moving fluid look similar to the container of glass balls. In fact, the latter almost looks like a large convection cell going up at the center of the cylinder.

If all balls had the same size, the result would be a slow and cyclical motion around the container. But something goes haywire when the larger glass ball gets near the cylinder wall at the top layer. It is too large to fit in the downward flow, and it gets stuck at the top layer. If you had many large balls, they would eventually all get trapped near the top of the container. If you imagine a large amount of large nuts with some amount of nut fragments, you would end up with all the powder at the bottom after a little while, and therefore observe the Brazil nut effect.

To verify their hypothesis that the whole flows were driven by friction with the container walls, they repeated the experiment using a container with a rough and a smooth side. The rough side provides a lot more friction, and as they expected it drove a much more important downward flow near the rough edge of the container wall:

In the image above, the right side of the cylindrical container is rough and provides friction, whereas the left side is smooth and provides almost no friction. As a consequence, the downward flow only happens on the right side. Source: Knight et al. (1993).

The team of scientists did not stop there. They decided to test another container shape, and quite amazingly they chose to study the behavior of a cone, making this very relevant to V60 brews ! They repeated the same experiment, and observed something shocking. The flow reversed entirely, and this caused the larger balls to get trapped at the bottom, instead of the surface !

A conical container reverses the direction of the convection-like flows, causing the larger balls to get stuck at the bottom. Source: Knight et al. (1993), modified.

They also observed that the thickness of the upward edge flow was a bit larger compared to the cylinder case. In principle, this could mean slightly larger boulders can complete the full cycle rather than get stuck somewhere, compared to the cylinder case.

Another science paper by Hejmady and co-authors showed this beautiful sequence of how layers of colored spheres evolve with vibration to bring out the details of the flow structure:

Top layer of photos: temporal sequence of how vibrations affect layers of colored balls. Bottom layer: similar with the exception of a large ball that is included to show how it gets trapped at the surface. Source: Hejmady et al. (2012).

They also show a nice visualization of the flow direction in a superposed image similar to a long-exposure photo:

This video shows a great visualization of the phenomenon:

With all these results in mind, you might think that a V-shaped container might work against clogging a filter because it would concentrate the boulders at the bottom instead of the fines, but that’s besides the point because the fines circulating around the edges of the paper filter would contribute to clogging it, regardless of the direction in which they are flowing. Fines small enough to penetrate the pores of the paper filter will get stuck in it, so any kind of motion that brings more fines in the vicinity of the filter will contribute to reach the point where the filter clogs. In other words, tapping either a conical or cylindrical filter will contribute to clogging the filter, even if they trap the boulders in different regions of the coffee bed.

These concepts could also be applied to coffee brewing in a couple more ways:

  • A vibrating cylinder could be used to lift boulders to the top of a dose of ground coffee to scoop it out and make the particle distribution narrower. This would be similar to sifting, but might be a bit more convenient and faster. It would probably share some of the disadvantages of sifting however, for example it would probably be messy and hard to replicate exactly.
  • The wall angles of a conical brewer could probably be chosen to minimize any displacement of particles in the brewer even in the presence of vibrations. As the Knight paper discusses, the fact that the very slanted walls reverse the flow probably means that there is one geometry with less slanted walls, in between this particular cone and a cylinder, which would stop all ball motion. Very smooth edges could also achieve this, but remember that the edges of a coffee bed are the paper filter itself, and it seems implausible to have a very smooth, frictionless paper filter.
  • If we had a filter that cannot be clogged, we could use this re-organization of the coffee bed particles to make extraction more even. Normally, the bottom of the coffee bed only comes in contact with concentrated water, and this causes the bottom to extract less and in a different way: it will preferentially extract chemicals that are not already dissolved in water, see this previous post I wrote for more about this. Conical brewers partly make up for this by having a larger amount of water pass through the bottom of the coffee bed because of their geometry, and this prevents the bottom from being under-extracted. This is an imperfect solution however, because the bottom of the coffee bed still extracts in a different way. Using vibration to re-organize the coffee bed without clogging the filter would be amazing, but a real challenge: it would require both the coffee particle distribution to be very even, and the filter pores to also be very even and smaller than the coffee particles.

It’s important to keep in mind that these papers are based on idealized scenario which could make their application to coffee brewing less straightforward. Here are some caveats I could think about, but there may be more:

  • The presence of water in the brewer provides additional forces (dragging along the currents and upward buoyancy) that may introduce different displacements of the coffee particles. In the presence of vibration, the flow of particles described above probably still happens, but it could be washed away by stronger effects.
  • The scientists that carried these experiments used a very specific type of tapping with a single 30 Hz frequency vibration. It’s possible that changing that frequency might affect the strength or even the direction of particle flows. In practice, finger tapping might cause vibrations of different and even varying frequencies.
  • The shapes of coffee particles are far from being spherical, which could affect these results. Similarly, in real scenarios we have a wide variety of particle sizes, not just uniform small particles and one large particle. I think those effects are less likely an issue, because convection-type flows are observed even in containers of unevenly shaped and sized nuts and dry foods.
  • It’s important to keep in mind that strong vibrations or upward forces can still lift boulders and cause fines to fall in between the cracks. This means that even in a scenario where the geometry of a container prevents convection-like displacements, this alternative type of fines migration could still happen.

I’m hoping you found these results as interesting as I did ! I’d like to thank Jesse Lambeth for digging up these papers too ! The header image is from Gsrdzl at Wikimedia Commons.

The Struggle for a Steady Kettle Flow

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through some of the links below.

Recently, I started attempting to modify my Brewista Artisan 0.6L gooseneck kettle to reduce its maximum flow rate and make it easier to keep a constant kettle flow of about 5 grams per second during my brews.

The Brewista Artisan 0.6L has a gooseneck with an inner opening diameter of 5.5 mm at the exit spout, which results in a maximum flow rate of about 24 g/s. This is a bit more than I would like, as I rarely ever need to pour that fast, and this makes it a bit harder to pour consistently at only 20% of the maximum flow rate. I asked someone on Instagram who has a Fellow Stagg EKG 0.6L to measure its maximum flow rate, and it is apparently a bit lower, at 20 g/s. This is quite nice, and adds another reason why I would have chosen the latter kettle if I was to do this again. Other reasons include not having to press re-heat again every time I pick up the kettle; the nice counter-weight in the handle which makes it easier to pour consistently; and the much better and more responsive customer service of Fellow versus Brewista.

If you wonder why I’m worrying about kettle flow speed, the consistency of my swirls and my accuracy in measuring TDS, it’s all because I’m trying to make every aspect of my brewing as consistent as I can from one brew to the next. There are a lot of things I would like to investigate which will require my brews to be very stable, including but not limited to:

  • The effect of filters on flow and clogging.
  • The effect of coffee aging and hardness on flow (through its shattering properties during grind).
  • The effects of bed preparation (nest shape, etc.) on extraction evenness.

Those are just a few examples in a very long list; it’s possible that getting a Decent espresso machine in pour over mode may allow me to do this, but saving up Patreon funds for it will take a long time, so I think it’s worth improving my manual repeatability in the meantime. I want to improve my repeatability not only in terms of extraction yield, but also brew time and flow of coffee coming out of the V60.

The first way in which I attempted to reduce my kettle flow was to obtain a Mandritech Hand Drip Water Valve, which you may have seen on my Instagram or Telegram channel. This product is only sold in Asia, but Edo Rezaprasga, a reader of my blog, was kind enough to mail me one so that I could test it.

The device ended up working very well in terms of getting a more constant kettle flow, as can be seen on this graph where I tried to maintain my flow at 5 g/s with and without the valve:

There is however one big drawback in using the Mandritech valve; it ended up making the flow of water much less cohesive, which resulted in less agitation, and lower average extraction yields. You can see how the flow breaks up after a very small vertical distance at 5 g/s in this video:

I could even see that the layer of water on top of my coffee bed was super clean, which is a telltale sign that very little agitation is being applied on the slurry:

For example, brewing Passenger’s Los Yoyos 2017 Harvest coffee without the Mandritech with my usual recipe (22 g dose with 1:17 ratio, Rao/Perger boiling water, 5 g/s kettle flow) yielded 1.49% TDS with a 320 g beverage weight for an average extraction yield of 21.7%, whereas brewing the same coffee in the same way with the Mandritech yielded a 328 g beverage weight at 1.28% TDS for an average extraction yield of 19.1%. This is quite a heavy loss !

Fancy gooseneck kettles are slightly twisted near the tip, and that makes their flow much more cohesive. As a result, the stream of water falling from the kettle can penetrate deeper in the slurry, and can introduce turbulence all the way down to the coffee bed instead of having all its kinetic energy dispersed near the surface of the water. This is an important aspect of gooseneck kettles that I had not appreciated before.

This does not mean it’s impossible to brew well with the Mandritech, but optimal pour over recipes for its use would resemble those that work well with the Melodrip, and this is yet another rabbit hole I have not explored well. Judging from Ray Murakawa’s brews, it seems that Melodrip-style pour overs work better with smaller doses, much finer grind size and more agressive blooms to get rid of dry pockets despite the finer grind size.

Just when I was about to leave the possibility of flow restriction aside, Edo wrote back to me and told me he found a set of flow restrictor plugs small enough to fit my Artisan 0.6L model ! Those are very similar to the Brewista-branded ones that only work with the larger 0.9L kettles, except with way more choices in terms of outer and inner diameter sizes. This is yet again another piece of geeky coffee gear that is only available in Asia, so I could only get my hands on one thanks to Edo again ! I asked him to get the model that has an inner diameter closest to 4.1 mm, as I calculated that this would result in a reduced flow of about 10 grams per second according to Sampson’s law. Edo found one with an inner diameter of 4.0 mm (the Bonavita number 5) and sent it to me.

We were not sure if the outer diameter of the restrictor would correspond to whatever the inner diameter of my kettle is at the base of the gooseneck, but I couldn’t verify whether it was 5.5 mm wide with my digital caliper anyway, because it doesn’t fit inside the kettle. We thus decided to try ordering it anyway and see what would happen.

When I received it, I measured its outer and inner diameters with my caliper and found that the inner diameter was 4.0 mm as expected, but the outer diameter was 6.9 mm, a bit larger than the exit spout of the Brewista Artisan gooseneck. Fortunately, 6.9 mm seems to be exactly the size of the gooseneck entrance opening at the base, because the flow restrictor fitted perfectly ! I have no idea why they are called Bonavita flow restrictors; I am not aware of a Bonavita kettle with such a small gooseneck diameter.

The maximum flow rate with the restrictor installed turned out to be about 9.5 g/s, very close to the expected value ! When I saw that, I went ahead and tried brewing with it, which resulted in a total mess. I realized mid brew that the boiling water seemed to create some bubbles near the smaller entrance of the flow restrictor, resulting in a very uneven flow. The kettle flow kept interrupting, or spitting high-velocity drops of water which created bad splattering. Here’s an example of such uneven flow in this video:

In order to test whether the boiling water was really the cause of this messy flow, I tried pouring with water at 210°F instead of 212°F, and the result was much better ! Here’s what this looked like:

So I went ahead and made another brew at 210°F, and this time the kettle flow went super well. Here’s a video of that brew:

I was brewing Heart’s Colombia Decaf Platino which is a washed mix of Caturra & Castillo decaffeinated with the ethyl acetate process. I used a grind setting of 7.3 at 700 RPM on the EG-1 with a 22 g dose, Rao/Perger water, a kettle flow of 5 g/s with a 1:17 ratio that resulted in a brew time of 4:50, a 323 g beverage weight and a 1.31% TDS concentration. This corresponds to a 19.2% average extraction yield. Decaffeinated coffee always extracts much lower than usual so that number is quite normal. Decaf coffee beans are also very brittle and therefore generate a lot of fines. This is why I used a coarser grind size than usual, and still ended up with a long drawdown time of 4:50; I should have ground even coarser than that. The brew tasted quite good; Heart are one of the rare roasters that roast their decaf coffee well. It was not astringent despite the long drawdown time, but I actually never tasted an astringent decaf, so I would not be surprised if the ethyl acetate process also removes whatever is responsible for astringency. Note that both of my swirls were quite bad in this brew.

You might think that it’s not worth losing 2 degrees Fahrenheit to get a more reliable kettle flow, but I actually think it’s worth it especially in my case; the loss of temperature in the slurry will be smaller than 2 degrees because water temperature drops faster from 212°F than it does from 210°F, but more importantly it will make my brews more repeatable which is very useful for science !

Here’s a comparison of how steadily I was able to pour with the non-upgraded Brewista Artisan kettle, with the Mandritech valve, and with the flow restrictor:

You can see that either the Mandritech or the flow restrictor allowed me to reach a much better stability, but only the flow restrictor allowed me to do this without losing the cohesion of my kettle flow ! To quantify this a bit better, here are the averages and standard deviations of the kettle flows I obtained during these practice pours:

  • Default kettle: 4.6 ± 0.6 g/s
  • Mandritech: 5.0 ± 0.2 g/s
  • Flow restrictor: 5.1 ± 0.2 g/s

Thanks again to Edo Rezaprasga for making all of this possible ! I’m super happy I finally found a solution to make my kettle flow more stable.

The Hardness of Green Coffee

Reminder: as an Amazon Associate I earn small commissions from qualifying purchases made through some of the links below.

When reading about coffee roasting, I have often seen bean density being discussed as an important variable that affects how beans will behave during the roasting process. The term bean density is a slightly vague way to refer to its volumetric mass density. One thing I have rarely seen discussed is the green coffee’s surface hardness and how it may affect roasting. In this post, I’ll show that the hardness of different green coffee beans can vary quite a lot, and even seem to vary more than their mass densities across different origins and processing methods.

Lately, I obtained a Shore D durometer with the goal to investigate how roasted coffee hardness correlates with pour over drawdown time through the beans’ shattering dynamics when grinding. This is an experiment I’m in the process of completing, but while I was doing this my friend Andy Kires at Canadian Roasting Society generously sent me samples of seven different green coffee beans so that I could try measuring the surface hardness of green coffee directly. So, I decided to compare the hardness and volumetric mass densities of each of these beans. Here are quick descriptions of the seven coffees:

  • Washed Colombian
  • Natural Colombian
  • Washed Ethiopian
  • Natural Ethiopian
  • Natural Brazilian
  • Washed Kenyan
  • Ethyl acetate decaffeinated washed Colombian

A more detailed description for each of them is given at the end of this post.

To measure the mass density, I followed this method by Green Farm Coffee where 25.0 grams (let’s call this M) of green coffee is placed in a 100 mL glass graduated cylinder pre-filled with 50 mL of water (this is V1).  After gently knocking the graduated cylinder to remove all air bubbles, I measured the new total volume V2 which can be used to determine how much displacement the green coffee caused. The volumetric mass density can then be calculated from this simple equation:

In the image above, some green coffee beans are floating because I have not yet gently knocked the graduated cylinder to remove air bubbles

The measurement errors on both volumes were about 0.5 mL as my cylinder is graduated to the mL, and the measurement error on the mass was 0.1 grams given the scale that I used. The results surprised me quite a bit: after knocking out the air bubbles which could seriously alter the results, all coffee beans had almost exactly the same volumetric mass density !

The next step was to measure the surface hardness of the green beans with the durometer. I tried both a Shore A and Shore D durometers, and it turns out that only the Shore D had the correct range to take useful measurements. I placed the green coffee beans on a flat surface, with the flat side of the bean down, and placed the durometer’s needle near the center of the rounded back of the coffee bean. I then pushed straight down on the durometer relatively hard until the measurement stabilized, and noted it down.

As I expected hardness to vary from one bean to the other, or even with the exact needle placement, I repeated this on at least 17 beans for every origin. One could look at the average of these measurements as the most interesting variable, but in these situations I tend to prefer looking at the median because it’s more robust against outlier measurements. The statistics geeks will know that the measurement error on the average is the standard deviation divided by the square root of the number of beans; an analogy can be made for the median, but the median absolute deviation is what must be used instead of the standard deviation. This is very convenient because the median absolute deviation is also more robust against outlier measurements.

Here are the median Shore D measurements I obtained for the seven different green coffees:

The individual raw measurements are available here. The Shore D hardness is a number that goes from 0 to 100 and goes up with hardness.

I still do not know exactly how these will affect roast behavior, but this seems very interesting to me, because we can tell different origins apart much easier than with the volumetric mass density. It might be worth the effort of taking 17 hardness measurements on green coffee beans if that allows a roaster to predict how hard a bean might crash, especially if it’s an expensive microlot !

Furthermore, it’s nice to see that the washed Kenyan lived up to the common folklore which says that Kenyans are particularly hard coffees ! However, it did not show any noticeable increase in mass density. This may be further evidence that mass density, or at least the Green Farm Coffee method of measuring it, are not that useful to characterize how green coffees behave during roasting !

If any of you roasters out there try measuring your green coffee hardness and notice a correlation with roast behavior, I’d love to hear about it; otherwise, Andy and I will keep you posted about what we find out.

For those interested, here’s the detailed information about each of the green coffees:

Washed Colombian: Laderas Del Tapias

Location: Neira, Caldas, Colombia
Altitude: 1650 to 1950 m.a.s.l.
Varietals: Caturra, Castillo, Bourbon, Catiopes
Owner: Rodrigo Alberto Pelaez

Natural Colombian: Villa Clabelina Natural

Cropster link
Location: Ciudad Bolivar, Antioquia, Colombia
Altitude: 1510 to 1800 m.a.s.l.
Varietals: Colombia, Caturra
Owner: Finca Villa Clabelina

Washed Ethiopian: Suke Quto

Trabocca link
Location: Guji, Ethiopia
Altitude: 1800 to 2200 m.a.s.l.
Varietals: Kurume, Welicho
Owner: Suke Quto farm

Natural Ethiopian: Chelelektu Natural

Ally Coffee link
Location: Yirgacheffe, Ethiopia
Altitude: 1800 m.a.s.l.
Varietals: Kurume, Dega, Wolisho
Owner: Chelelektu washing station (several small farms)

Natural Brazilian: Pedro Humberto Veloso

Location: Carmo do Paranaiba, Santa Cecília, Santa Catarina, Brazil
Altitude: 980 to 1120 m.a.s.l.
Varietals: Catucaí 785-15
Owner: Pedro Humberto Veloso

Washed Kenyan: Kii AA

Cropster Link
Location: Kirinyaga, Kenya
Altitude: 1600 to 1800 m.a.s.l.
Varietals: SL28, SL34
Owner: Smallholders at Kii wet mill factory

Decaffeinated Washed Colombian: Cauca Inzá 

Location: Cauca Inzá, Colombia
Altitude: 1750 to 1900 m.a.s.l.
Varietals: Caturra, Colombia, Castillo
Decaffeination: Ethyl Acetate
Owner: Asorcafe

I’d like to thank Andy Kires from Canadian Roasting Society for allowing me to run this cool experiment by providing the green coffee !