I've looked at this really closely and thought about it from a Geo player's perspective.
The troll incident:
This is nothing but a series of mistakes. Geo was playing passive at this point in the game and didn't want to engage in the team fight, so he went for rosh. When the dust settled, he ran away. The team fight ended at the exact moment troll was going toward rosh, so it's possible geo was reacting to either of those facts. Geo should have joined the fight earlier, but that's all there is to this incident. Mistake on his part.
The WW incident:
I had the same reaction as Frank, but maybe even stronger. I actually think I would have made the same play, in general. WW runs South, so geo follows. After about 10 seconds of following, Geo realizes that WW's arctic burn is now available, so he would fly over some trees. Where is he most likely to go? Right? No. That's too accessible for Geo. So he goes left, but does he go toward base or away from base? From WW's perspective, Geo is most likely to guess toward base, so WW goes away from base. That's how I usually plan blind blinks on any hero. Keeping one mental step ahead of a typical strategy. Sometimes it pays off, sometimes it doesn't. There are two additional facts about this incident that suggest to me that MH is unlikely:
1. When blinking into the left-side trees, as Geo did here, there are a limited number of "open" spaces that units will fit into, so if you click your blink on a random space, you usually don't end up exactly where you targeted, since, if you clicked on a bunch of trees, the game will autocorrect and put you into an open space. So we don't really know exactly where Geo meant his blink to land. All we know is that he landed in one of the few open spaces.
2. Not throwing additional nets doesn't tell us anything. If you watch the rest of the replay for context, he sometimes throws random nets in different directions and sometimes doesn't. I suspect he's just undisciplined in his control of the grouped Meeps and doesn't always have them properly selected or queued. Frank suggested something to this effect.
My understanding after reading countless decisions on maphack bans on our forums is that ENT has a blanket policy of not banning for isolated incidents unless they are 100% proof of maphack. I've seen much, much more blatant plays not result in a ban, because they were isolated incidents. And the rosh incident adds literally nothing, as explained above.
@nabo. I would appreciate your thoughts on this issue, as I am still in the process of learning about the standard for judging a maphack ban.