Interestingly, if you assume these are the only ways a startup can fail, and that for a typical startup creator (who is self-selected as smarter and more conscientious than average) each of these has an independent 90% chance of working out (what seemed to me like a reasonable estimate), then the overall chance of the startup succeeding is 0.9^21=0.11, remarkably close to the actual base rate. This particular model is unlikely to be accurate, but it is another way of illustrating your point.