Ok, the 0.85 approximation is probably the "issue". I'll see if I can find the VST patent which has a more accurate estimation, especially over the espresso range.

At 10% TDS the Brix to TDS factor is closer to 0.83. So the correct converted TDS values will actually be lower, not higher.

Eyeballing the two points on that graph as 10.2% and 8.75%, the corrected values would be 10.0% and 8.54%. So the extractions would go from (eyeballing again) 21.9% and 19.7% down to 21.5% and 19.2%.

That calculation comes not from the VST patent alone since it has a formula that relates %TDS to nD. However a second formula relating Brix to nD can be used in conjunction with the VST one to get a Brix to %TDS formula:

%TDS= 0.003228 * Bx^2 + 0.7953 * Bx + 0.002352


Unfiltered samples tend to read high which makes the actual concentration lower still. But since when is a 19% extraction yield "bad"?

I think you're going down the wrong path if you're focusing on EY for milk drinks. I think TDS is a better target, as with EY you hit diminishing returns and start to get a weaker espresso. But even still, I don't think there's really enough info to say what a proper tds is for a milky drink? That's going to depend on each coffee and you'll have to figure it out by dialing in anyway. I can really only see that being beneficial in a cafe setting when you're using the same coffee constantly.