Why TDS/extraction yield measurements are not a good way to evaluate taste profile/quality? - Page 3

Want to talk espresso but not sure which forum? If so, this is the right one.

Postby RyanJE » Mar 13, 2019, 9:39 am

samuellaw178 wrote:
For the second part, I was working under the assumption that, different particle distribution would contribute to different levels of EY%, at the same extraction parameters (same extraction time, same brew ratio). High uniformity=higher extraction yield and low uniformity=low EY%. That's why EY% may be used to indirectly access how *uniform* the particle distribution is.

By extrapolating that concept, similar EY% = similar particle uniformity, which should in theory taste similar (because we are extracting from the same coffee that has the same set of chemical molecules). At least that was what I thought before this thread.

I am by no means an expert but have played around a bit with this using a variety of grinders and testing EY. I think its plausible that the average extraction could be the same especially at lower EY%. The issue is that then the better grinders dont taste their best at that same extraction. For example, I had a vario with ceramic burrs and also (have) a MK Tanzania. I could easily extract using the same parameters to a 19% EY for example. The vario would taste OK but not the Tanz, it tasted gross. Numerically I have to push the Tanz much further to get it to shine often requiring a much finer grind. At times I wish I never bought a refractometer, for a long time I was trying to "fit" my extractions into a numerical window (the various brew charts). It was frustrating and I was basically limiting myself because of the numbers.

Once I started relying on taste for a while I went back to test just to see. Regularly the EY exceeds or is at the upper limits of the "optimum ranges of the SCA", it has to or else it tastes weird (presets at zero). I think Jim S. has used the term "cartoonish" to describe this before.
I drink two shots before I drink two shots, then I drink two more....

User avatar
Team HB

Postby another_jim » Mar 13, 2019, 10:41 am

RyanJE wrote:IOnce I started relying on taste for a while I went back to test just to see. Regularly the EY exceeds or is at the upper limits of the "optimum ranges of the SCA", it has to or else it tastes weird (presets at zero). I think Jim S. has used the term "cartoonish" to describe this before.

Yeah, I associate underextracted coffee with cartoons: primary colors and outlines in heavy black, in your face, without subtlety or balance. Overextracted is like a pale watercolor or pastel drawing, often lovely, but so subtle, you wish it would speak up. (this presumes roughly equal TDS levels)

This is why Sam and I were surprised that Sam was getting these (shy/brash?) differences from the same coffee at the same EY measure. I assumed when the same dose and grind level from different grinders gave me this difference, it was due to EY differences. That may not be the case.
Jim Schulman

User avatar
Team HB

Postby Jake_G » Mar 14, 2019, 8:12 pm

Well, I can't taste colors or shapes yet, so my input here will be slightly muted in comparison to some :roll: .

But I do have some thoughts on what TDS and EY actually tell us regarding the extraction and why it is that some grinders need more extraction than others to taste good, so I'll give it a go. I'm going to use "well-aligned" or "precision" to describe grinders with a tight particle distribution spread and "misaligned" or "poorly-aligned" to describe grinders with a wider spread. Really, the grind distribution is all that matters, whether it be burr geometry or alignment that gets you to one extreme or the other.

Below is a random bunch of plotted numbers that vaguely represent a particle distribution plot of a very well aligned flat burr grinder, with a little pictogram of "perfectly aligned" burrs to drive the point home.
In this perfect world, the burrs are 400 um apart and they generate relatively few fines in the 50 um range. You build a puck with these grinds and pull a shot. You get some TDS at some brew ratio and it corresponds to some EY%. All clear?

Then we have some misaligned burrs.
With these burrs, we have the lower burr cocked off to one side in the rotating burr mount.


Looking at our pictogram, the "center" of the lower burr is at 400 um and the left side is maybe 300 and the right side 500. This widens the particle distribution and greatly decreases the amount of 400 um particles. Even so, on average there are twice as many 400 um particles as there are 300 or 500. This is because if we were to look at the burrs from a top-down view, the front (6 o'clock) and the back (12 o'clock) of the burrs (the center in our pictogram) would both be 400 um apart. The left (9 o'clock) and the right (3 o'clock) would be at 300 and 500 um, respectively. Two points of contact at 400, one each at 300 and 500. Make sense so far? I left the fines alone because I'm not sure how alignment affects fines so I won't make assumptions suggesting that I do...


Brewing a shot with this batch will result in some TDS at some brew ratio and another corresponding EY%. These values may or may not equal the values we got above. But here's why they will taste different even if the TDS and EY% are the same...

Extraction is all about contact time, concentration gradients and surface area. You create a puck, with soluble compounds clinging to the surface area of the ground coffee particles. You introduce a solvent into the puck and it strips the soluble compounds off of the exposed surfaces. The process is similar to heat transfer through a heat exchanger. Too fast, not enough contact time to transfer the compounds into the solvent, so the concentration gradient remains high (in an HX, this means the cooling water doesn't get hot and the thing being cooled doesn't cool down). Too slow, and the concentration gradient gets too low and the solvent is just as concentrated as the puck and it can't extract any more (in an HX, the cooling water gets hot, but the thing being cooled doesn't cool off...).

The trick in both cases is to find the flow rate where the contact time is sufficient to strip the desired compounds from the available surface area, and the replenishment rate is high enough to bring new lower concentration solvent into the puck to keep the extraction progressing (in the HX like an engine radiator, the goal would be to cool the engine coolant by optimizing this rate).

Why do I bring up this?

Well, if you look again at the particle analysis, it's clear that since smaller particles have a higher surface area to volume ratio than larger particles, the better aligned grinder has a more uniform amount of surface area available for extraction than the misaligned grinder. So what does that mean?

Well, the smaller particles in the misaligned grinder will extract more solids than the target size and the larger particles because they (the smaller ones) have more surface area. The target particles will extract roughly the same as the target particles coming from the well aligned grinder. Finally, the larger particles will extract less solids than the target and smaller particles.

Begin clarifying edit:
So, when the flow rate is optimized for the average particles, you get a mix of extraction yields from the various particle sizes.
•The smallest particles, which have the most surface area, release the most solids per unit volume, but do so without being over-extracted, because they have a lower contact time per surface area than any of the other particles. This is because the flow rate is faster than what would be ideal for particles with this much surface area since it is optimized for the average particles.
•The target particles behave as expected. They have a medium surface area per unit volume, and a medium contact time per surface area. So, fewer solids per unit volume are extracted, but what is extracted is extracted more than the smaller particles.
•Finally we have the large guys, or boulders. They have the lowest surface area per unit volume, and the highest contact time per unit surface area. The boulders contribute the least solids per unit volume, but because the flow rate is optimized for the average particle size, the contact time on each particle is huge relative to its surface area. Because of this, the boulders give you over-extracted flavors first.
•All together you have an average extraction yield that is based on some portion of under-extracted small particles and over extracted boulders mixed with the target particles that extracted at the level that you are shooting for.
End clarifying edit:

What does this mean?

Well, I think it means that over-extracted flavors will show up faster with misaligned grinders and they will taste better at lower extraction yields than their well-aligned brethren. Conversely, with fewer small sized particles and boulders (for a given target grind size) the well-aligned grinder will taste better at higher extraction yields than the misaligned grinder.

Put all of this together and even though the higher extraction of smaller particles and lower extraction of larger particles may even out with the extraction of target particles on a well-aligned grinder, they will taste different. Likewise, a burr geometry that yields a tight distribution (such as "unimodal" burrs) will taste better at high extraction yields than a burr geometry that yields a wider distribution (or "bimodal" burrs). Conversley, the wider spread will taste better than the tight spread at low extraction yields and the one constant is that they will taste different from each other at the same extraction yield...

That's all.




Postby happycat » Mar 14, 2019, 9:28 pm

another_jim wrote:No need for a contest between scientism and subjective taste swamps. If you salt your food, it tastes saltier; if you add sugar, it tastes sweeter. So what happens to the taste of coffee when you grind finer? I call it "softer" Call it "finier" if you like; but it's just as consistent as adding salt and things tasting saltier. There's nothing scientistic or subjective about it; it's a repeatable action leads to the same change each time.

Of course, it helps to know what change to look for when doing a particular action. Here's my suggestion for TDS and EY (after playing with the brew ratios of well extracted shots).

Bitter/sours diminish more quickly than sweetness as the concentration of the coffee is lowered. This makes evolutionary sense: our starving ancestors would go for sweet in any concentration, while bitter and sour were toxicity warnings that could be ignored at lower concentrations, but not higher ones. This means a poorly extracted coffee might taste great at low TDS, and be a sink shot at high TDS, while a highly extracted one (with lots of buffers and starchy compounds mellowing the bitters and sours), will be much the same at any TDS level.

See, this idea is really easy to falsify, therefore eminently scientific, yet all about "subjective taste."

Thanks for the reply. I don't really want to pick at your statements. I suppose I would just leave off that science is not just repeatable in a subjective microcosm, it's repeatable within the community of practice, is grounded in what has gone before, and is communicated in a repeatable and observable manner. Having ground through a recent peer review, I'm sensitive :D
LMWDP #603

User avatar
Team HB

Postby another_jim » Mar 14, 2019, 10:09 pm

happycat wrote: ... science is not just repeatable in a subjective microcosm, it's repeatable within the community of practice,

Really. The salted soup only tastes consistently saltier at home; it's hit or miss in restaurant kitchens? Also, while meters readings may be objective, the experience of reading it is subjective. So my EY of 20% may not be be yours, since I'm reporting the subjective experience of reading the meter.

Every "community of practice" is based on a trained intersubjectivity. Interpreting X-rays is a good example of a highly trained and almost completely intersubjective sensory judgments on which lives depend. An X-ray reading AI may make the process even more consistent (as soon as someone gets a clue of the criteria the AI uses), but it wouldn't suddenly turn unscientific guesswork into hard science.

I might have R and Red you as an anonymous reviewer as well. :wink:
Jim Schulman

User avatar

Postby Almico » Mar 15, 2019, 12:23 pm

My answer to the OP query: Simply because TDS/extraction yield only tells you the concentration of particles in the brew. It does not tell you what those particles taste like.

I happily jumped off the coffee minutia train a while ago. There are just too many uncontrolled/uncontrollable variables in a handful of roasted coffee to bother.

For example, in two handfuls of "single origin coffee", how many beans are Typica, Caturra, Catuai, Bourbon, Colombia etc. What percentage of those beans were picked at the peak of ripeness? How many were over ripe and under ripe? Are they all the same density? Are all those percentages exactly the same in each handful? If not, they will taste different. And one funky bean can spoil a whole cup, if not a whole pot.

So unless the raw materials are exactly the same, what point is there is testing the subtlest of flavor nuances from one to the other. You would have to do multiple tests over a very large sample groups to even begin to start tracking useable data and that's just too much work for me.

So just because the EY percentages might be the same, the dissolved solids are likely not.

User avatar

Postby Balthazar_B » Mar 15, 2019, 12:46 pm

samuellaw178 wrote:So in my opinion, TLDR (or if it got too technical): Refractometer measurement is not sensitive enough to account for subtle taste difference, and the measurement doesn't account for the chemical composition. Most of us are not trained enough (if it is at all possible to achieve zero variation) to do the measurement with enough precision.

This is not to say refractometer is useless (far from it, they can definitely be useful in some scenarios), but I think it should not be the only guideline to evaluate anything, especially not without a proper context.

That's just my way of rationalizing what I am seeing and by no means tested or proven, so take it with a big grain of salt! But I am keen to hear what you think about it. Have you done EY measurements across different grinders/machines using the same coffee and notice the same observation?

You're right, Sam, and to use an analogy, using a refractometer as a tool for measuring flavor is like using a barometer to measure humidity. There'll be an indirect relationship, but not the best choice of tool (maybe not the best choice of analogy, either, but you get my drift). As you point out, a refractometer can measure TDS, though a bit crudely. You could get a better idea of the composition of those solids through mass spectrometry. And to quantify actual flavor -- which is arguably the ultimate object -- you'd need to use the right tool for the job. I'm not well-versed in that area of instrumentation, but my suspicion is:

  • The tools and methodologies are pretty crude right now, in relation to where they will be someday;
  • That someday could well be fast approaching, since there's a lot of money to be made matching products to mass tastes. Human labor is expensive and inefficient. And the nexus of instrumentation, big data, and AI is likely to be -- eventually -- less expensive and more efficient.
Perhaps you've run across this document already?
- John

LMWDP # 577


Postby Mrboots2u » Mar 15, 2019, 1:08 pm

Measuring extraction yield was never meant to be a replacement for tasting .....


Postby OldNuc » Mar 15, 2019, 3:56 pm

It is highly unlikely that an AI human taste bud replicator will ever be developed that could even remotely be considered universally acceptable. The Star Trek replicator will never make the perfect cup of coffee.

Team HB

Postby samuellaw178 » Mar 15, 2019, 4:15 pm

Balthazar_B wrote:Perhaps you've run across this document already?

No John, I haven't seen that but that looks very interesting. I will have a read, thank you!

Mrboots2u wrote:Measuring extraction yield was never meant to be a replacement for tasting .....

Now I see, 'replacement for tasting' is a poor choice of word on my part. :oops: I definitely meant using TDS/EY as a predictor of taste quality, relative to a comparison standard (eventually we all still want to taste/drink the coffee right? :P But how do you communicate that taste quality to someone across the globe on the internet?).

Same/similar grinder, same TDS/EY, same coffee, same extraction metrics (contact time, brew ratio, and what have you), and the TDS/EY is likely still not a good predictive indicator that the shots will taste the same (in relation to the 'standard/control shot' with the same parameters).