Which 5% of projects are really great? In my experience, presuming you have tight filters such that all of your projects are plausibly potentially great, you really don’t know until you try. That’s the point of an incubator.

It’s not that hard to evaluate when something is working (ie the hard part in evaluation is false negatives, not false positives).

In Area 120’s case there was no coasting - if anything there was a hair-trigger standard to shut down underperforming projects.