It's actually the percentage of (+1) minus the percentage of (-1). 7-8's are literally not being counted (as you said 0's).
Imagine if 990/1000 give 7 or 8.
Out of the remaining 10, 7 give a 9 and 3 any number below 6.
The count would be: 70%-30%=40 (which would be a good outcome).
But only 1% of everyone that participated ended up determining the actual NPS Score.
I don't think that's right, and the example given in the article seems to agree with me. The 7s&8s are indeed counted, indirectly: the score is the percent of all responses (including 7s&8s) that are promoters minus the percent of detractors.
So, using your example, 990/1000=99% of people are passive, 7/1000=0.7% are promoters, and 3/1000=0.3% are detractors, so the NPS score is 0.7-0.3, or 0.4 (not 40). Passives aren't a part of the last calculation, but they affect the percent values for promoters and detractors.
