Three, from 2012=(11111011100)2, we know that the numbers satisfying the conditions have at most 11 digits in binary representation.
The first digit must be 1, so the number of d+1-digit numbers with exactly k+1 digits being 1 is Cdk, and condition (2) is equivalent to
k+1⩾d−k+2⇔k⩾2d+1⇔d−k⩽2d−1.
First, consider the numbers in [1,2048) that satisfy condition (2)
There are
∑d=110∑k={2d+1]dCdk=∑d=110[2d−1]=∑a=05(∑i=0u−1C2a−1i+∑i=0a−1C2ai)=∑a=15[22a−2+21(22a−C2aa)]=43∑a=154n−21∑a=1aC2aa=(45−1)−21⋅(C21+C42+C63+C84+C105)=1023−21(2+6+20+70+252)=848 (numbers),
where ⌈x⌉ represents the smallest integer not less than the real number x, and [x] represents the largest integer not greater than the real number x.
Since there are 35 integers in [2013,2048), and only
(11111100000)2=2016
does not satisfy condition (2), we have
N=848−34=814.
Therefore, the sum of the digits of N is 13.