The spirit of this prediction is for assessing how critical it is to reduce emissions. If carbon capture is expected to get cost-effective enough then reducing emissions now is less critical. Presumably. So anything that permanently 

 carbon counts. Anything that just prevents emission in the first place, including things like smokestack scrubbing that capture carbon before it's actually emitted, do not count.

"Eventually." See the next section. But this will resolve on January 1, 2050 at the latest based on scientific consensus.

That carbon removal will end up being responsible for an impactful decrease in atmospheric CO2 concentration -- enough to yield half a degree Celsius less warming than we'd otherwise see.

The idea is to predict whether carbon capture will get cheap enough to do at a big enough scale to matter and be worthwhile.

The atmospheric CO2 decrease doesn't have to have happened by 2050.