Stages:

1. plug all empty spaces surrounded only by color C with new stones also of color C.
2. Remove stones from anywhere on the board in pairs of one black, one white, until no such pairs exist.
3. The winner is the color left on the board.

I'm using the TrompTaylorRules as a metric. Under those, stones in seki count towards the score (they "reach empty"). Some empty instersections in seki may be territory, and get plugged (they "reach only one color"). Others aren't, and don't. Then the remaining stones are counted off pairwise as above.

So, I don't see how the count procedure fails?