Neal,
Milton's scenario is not correct, as you pointed out, but since
the NotA votes are a strict threshold (i.e. votes for a candidate
must be greater than the sum of NotA), the problem persists.
Please notice that you have 100 voters but max 300 votes (and
some of those are "supervotes" witth ambiguous values from -1
to -3). Let's make a twist to Milton's example:
10 voters vote A|B|C
70 voters vote A|B|N
10 voters vote A|N
10 voters vote N
Tally:
Voters=100
Votes= 270 < 300 → OK
N=90
C = 10 < 90 → not elected
B = 80 < 90 → not elected
A = 90 = 90 → not elected

Let me get back into the mock example above:
* A|B|C votes have _positive_ effect for A, B, C.
* A|B|N votes have _negative_ effect for C and _neutral_ effect
  for A, B (they add 1 to A's and B's count, but also add 1 to the
  threshold)
* A|N votes have _negative_ effect for B, C, and _neutral_ for A
* N votes have _negative_ effect for A, B, C.

As discussed elsewhere (and I think Milton agrees with this), the
underlying idea seems to have some kind of "trust|distrust|neutral"
system, i.e. the voter has 3 options vis-à-vis each candidate =
{1, -1, 0}. Only candidates with a _positive_ outcome are elected.
However, the implementation of the idea became a fiasco. The
simplest solution would have been to redesign the ballot.

Regards,

Enrique