Hello,

In EB Garamond I’ve implemented decomposition for all composed glyphs. 
For uppercase greek I had to use contextual kerning rules: accents like 
grave or acute and aspiration marks are placed left of the letter, so 
the glyph has to be placed more to the right and its advance increased. 
The code in the feature file looks like this:
    pos [ Epsilon Eta Iota Rho ]' <79 0 79 0> @grk_tlAcc1;

Iota adscriptum needs extra space on the right of the base glyph, so 
there are position rules for that too:
    pos Eta' <0 0 200 0> uni0345.cap1;
and
    pos Eta' <79 0 279 0> uni0345.cap1 @grk_tlAcc1;

Context (and LuaLaTeX too) doesn’t do the positioning correctly. I’ve 
prepared a minimal example of uppercase etas. The first row without 
decomposition, the second decomposed, the third with alternate 
uni0345.cap1 (iota adscriptum). The first two should look exactly the same:

\usemodule[simplefonts]
\setmainfont[ebgaramond]
\definefontfeature[decomp][ss20=yes]
\definefontfeature[iota1][ss20=yes,cv81=1]
\starttext
ἨἩἪἫἬἭἮἯᾘᾙᾚᾛᾜᾝᾞᾟῌ \par
\addff{decomp}ἨἩἪἫἬἭἮἯᾘᾙᾚᾛᾜᾝᾞᾟῌ \par
\addff{iota1}ἨἩἪἫἬἭἮἯᾘᾙᾚᾛᾜᾝᾞᾟῌ \par
\stoptext

I use ss20 for full decomposition because it should be selected by the 
user explicitly only when needed, while ccmp should be enabled by 
default. The attached ctx_test.pdf shows its output. The font used can 
be downloaded at 
http://georgduffner.at/ebgaramond/fonts/EBGaramond12-Regular.otf. For 
comparison, there’s a test page at 
http://georgduffner.at/ebgaramond/etatests.html – current Firefox and 
Chromium show what is expected.

There’s a second testfile at 
https://github.com/georgd/EB-Garamond/blob/master/tests/context_greektests.tex 
with the complete Unicode block of greek polyphonic characters, its pdf 
output is attached to this mail too, a comparison html page is at 
http://georgduffner.at/ebgaramond/greektests.html

Best regards,
Georg