I had previously typed up a reply on this and lost it, so I'll post up a short form reply.

I had the benefit of seeing Prof Hendon's presentation of the results of this last year and I think everyone would benefit from the emphasis that he put on it: my impression was that he was saying that the important thing on this study is really the whiskers on the data points that you can see in figure 4. The research stands for the proposition that for a conventional 20g dose, 40g yield shot brewed at their parameters, at the barista-determined grind setting, you are unlikely to be able to reduce the extraction yield margin of error below about 0.75%-1%. At least, not without being a better barista than Michael. The rest of the Matter article is basically about reducing this margin of error. In the matter article, they are very, very careful to make the point that the problem that their methodology is solving is this margin of error for an equivalent extraction yield, but they acknowledge that it will taste different, eg:

The chemical composition of the faster shot cannot be the same as the slower shot owing to molecular differences in solubility, dissolution rate, and resultant molecule-dependent impact on the refractive index.38 This result does not undermine our use of EY, but rather illustrates that the barista indeed needs to taste the coffee, rather than measure its solvated mass.

No doubt their caution in the Matter article is due to the subjectivity of the taste judgment, so it's annoying, but predictable, to see that everywhere it's reported and commented on, people are commenting on the basis that it tastes good. It might, but that's not the point of the article, and that's not what they researched. The scientific and statistical rigour applies to the extraction yield and the error, not how good it tastes.

Having observed that, this is still a useful paper for all of us:

If you like the way that 15g dose/40g yield/6bar shots taste, then you know that you have a very good chance of being able to reproduce them time after time.

If you don't like the way that such shots taste and you're more of the 20g dose/40g yield school of thought, then you know that it's going to be pretty hard to achieve repeatability of greater than a margin of error of 1% extraction yield. So if that is keeping you up at night, maybe this will help you deal with it.

Given the caution with which Prof Hendon and Michael avoided saying that the shots taste better in the Matter article, it's sort of odd to see that, in the press around the article, they seem to be putting out their judgments about taste. It's even more annoying that such judgments are reported without a clear description of how the shots taste different, which is forever the problem with the way that the media reports on coffee. They said that they got a cafe to try switching to faster shots and they had cost savings data from that cafe; it would have been interesting if they had sales data or consumer preference data as well. I wonder if the cafe gained or lost customers because of the switch? We'll never know. I predict that people will try the faster shots, appreciate that they are different, many people will conclude that they don't like the taste of the faster shots and those people will therefore disregard the rest of the matter article, which is a shame.

In terms of faster shots/lower dose, to some extent it's a matter of taste, and it's worthwhile noting that these ideas aren't new. RyanP observed to me long before the article that he liked coarse grind/high weight shots. Jim observed a while back that the move to big 20g type baskets for similar shot weights can lead to taste disappointment. Personally, I'm keen to dabble more with it.

Finally, I'll leave out a little bonus round question. Can anyone explain this to me: In figure 4, you will see that the extraction yield peak data point at grind setting 1.7 has whiskers, so there's a margin of error there of maybe 0.6%TDS. In figure 6, the equivalent extraction yield peak data point has no whiskers. What gives? Doesn't the usefulness of the topmost regime in figure 6 depend on the error bar for that data point being pretty much zero, which is not what their research showed? I'm guessing this is an oversimplification just for illustrative purposes, since presumably they would have used the data point for grind setting 1.9 in figure 4 with the exact same methodology. But could people weigh in, because I was on the verge of emailing Prof Hendon to ask and I don't want to pester him if this is a stupid question!