I'd tweak it to more like this
28nm Maxwell GM204 Q4 2014 (GTX870 /GTX880)
20nm Maxwell GM210 Q2/3 2015 (Titan)
20nm Maxwell GM210 Q2/3 2015 (GTX880 Ti)(it's a Titan core, it will just be held back 6+months, but timeline of manufacturing this should be here
20nm Maxwell GM204 Q3/Q4 2015 (GTX 870/880, or a gtx965ti boost mega super uber evga black 28gb) + 15% clocks or something
It's highly likely that GM204 is effectively a 20nm part but being made bigger at 28nm until it can be shrunk. IE it's a circa 300mm^2 20nm part, but it will be a 550mm^2 or so 28nm part till the shrink.
Difficult to say if they'll rebrand it or do it sensibly. IE a 285gtx, or release it like a new part, even then they could call it high end or midrange, who knows.
You can almost always gain some efficiency over 2-3 years so even at the same die size it should be faster, not least if it has a smaller memory bus and takes less die size it will have more raw power(or they might make it smaller, depends on the shrink I guess, could be 300mm^2 @ 20nm and 580mm^2 @ 28, or because it has to fit both nodes it might be smaller).
The one trouble they'll have is the bus, it could be faster than a Gk110 in some circumstances, and significantly slower in others. 1080p/ultra settings, it might be say 10-20% faster, but at 4k it might be 30% slower despite having more actual power.
It's not surprising, a midrange card won't be aimed at solid 4k performance for another couple generations. People usually underestimate how much of the power usage is down to accessing the memory. For 99% of buyers who want solid fps at 1080p(currently) then a 384/512bit bus would be a waste, use more power, cost more die space. Trying to make 4k better on mainstream hardware will likely arrive with HMB on package solutions where the bandwidth/memory cost will be drastically reduced.