Thanks Kendall for this great opportunity.

Jim, I have a couple Qs. You know much better than I. Hope the Qs don't sound too critical.

1 Are SOs better suited for this type of testing (as opposed to a blend)?

2 How would people deal with storage and humidity during the 8 waiting days? I ask this because beans could pick up humidity. And drier beans are be more brittle (more fines?).

If it were me, I would vote for the 2x6 samples form instead of 1x12. The results would be more solid as the condition of the burrs could play a role. If possible, I would chose two people with the same grinder.

I think it would be important if all beans came out of the roast batch (just in case).

Cheers

another_jim wrote:Here are the main questions:

• One sample from 12 grinders or two doses from 6 grinders?
• Restrict it to one basket and machine or allow variants?
• Which coffee?

Thank you to Kendall and yourself for the effort to pull this together. To address only the most obvious questions: To be meaningful, there should be an effort to eliminate as many extraneous variables as possible. All grind samples from the same batch of coffee. And a coffee with a reputation of being competently roasted so that we're not distracted by unusual roasting artifacts. I'm sure that this would be satisfied by any number of brand name coffees such as Black Cat. No one coffee will have universal acclaim, but Black Cat may be one good candidate in so much as it is widely distributed and well known to a lot of people.

It may not make a difference, but I would suggest an effort to restrict this to one basket and machine. It would be aggravating if the results were tainted or muddied by differences between baskets and machines that are not obvious at this time.

Most of all, I would suggest that two samples from 6 grinders would be more informative. I'm sure a lot of people would like to see grinder X (probably the one they own) included. But two samples per grinder would provide a second level of information. Not just how distributions vary between grinders. But also whether "coarser" vs. "finer" grinds affect distributions in a consistent manner across many grinders or if each grinder is a story unto itself. At least two large conical and two large planar grinders amongst the six? Plus a bulk grinder?

Open ended questions: Agree/Disagree? Any important considerations that haven't been raised yet?

.

First, thanks to Kendall for his efforts.

But, does anyone here seriously think that anything useful can come of this? Finding a needle in a haystack would be far easier.

By way of biases, let me state that I think we are barking up the wrong tree. But let's say, for the sake of argument, that there was really something simple to find in this sort of analysis that would answer a useful question or questions. Now how would one set about trying to find this piece of information?

I think that one would simplify the "experiment" substantially in order to give it some chance, however small, of answering at least one question. And how would one do that?

(1) Use a single batch of a simple coffee consisting of one type of bean. This would be the easiest. If one chose to complicate matters than use a single bag of a single blend that had been freshly roasted.

(2) Use a couple of grinders, one that could be considered to be mediocre and one that could be considered to be excellent. Perhaps one could choose a Rocky or a Mazzer Mini for the "mediocre" grinder, and a Robur or a Compak K10 or something else in that class as the "excellent" grinder.

(3) Have the two grinders in one place, preferably where the test is to be performed, and have them grind the coffee within a relatively short period of time before the coffee is analyzed. Set the grinders to produce an equivalent, acceptable "pour" time and volume on an "acceptable" reference espresso machine.

(4) Analyze the resultant grounds.

If you found something with this sort of a simple test, then one could consider broadening the test, however the sort of stuff that is being discussed in this thread is almost certain to produce no useful results, whatsoever.

ken

I'm not certain that grind distribution is well enough understood that it would be easily reduced to a true single variable that can be rigorously tested. Or that such a narrow focus would even be the best approach at this point. There is obviously value in the rigorous approach that Ken advocates. But I was looking at this as an opportunity to look for trends and flesh out general principles. This might not be as definitive but we would probably gain a better overall understanding of what is going on.

Two different approaches. And either way, the methodology deserves discussion and careful planning.

.

I agree with whats been stated above, limit this test to a few grinders (say a Rocky, a Super Jolly and a Robur) all dialed in by the same person on the same machine. Ideally the coffee should be cupped by more than one person and the notes available alongside the test results.

I'm a bit skeptical to sending ground coffee in the mail though... If the samples are provided enough energy (shaking, changes in temperature etc) the particle distribution could change. Perhaps this should be tested first? if Kendall would grind up two samples on his grinder and leave one undisturbed but put the other one through rigorous shaking, into the fridge and back to room temperature etc this issue could be settled.

If we're just exploring, and not testing statistics, a pair of samples from six grinders would tell more that one from each of twelve. If each sample from a pair looks roughly the same, then a "good enough" level of confidence would be achieved that the particle distribution on X grinder is Y.

Now, if the three flat burrs all end up looking like Y, and the three conicals look like Z, and Y tastes like "#"% and Z tastes better, then you're on to something.

PC

Interestingly, I linked the original thread to a local coffee forum and we ended up in a similar argument...

Both points of view have their merits. Controlled experiments are not the only solution and make little sense if we do not understand the context well enough to design a good experiment. On the other hand, exploring plain noise does not increase our understanding.

To get some measures of the variance of "the testing machinery", multiple samples per grinder are needed. At least two, preferrable more. At least I would be more interested in seeing a robust result of two representative grinders (Mini and Robur, perhaps) than having more grinders tested. Of course it would be nice to have both, but it will be quite a lot of work.