Data shows, as you say, that small scale performance is no indicator of large scale performance. How then do you decide which projects are worth trying on the

Data shows, as you say, that small scale performance is no indicator of large scale performance. How then do you decide which projects are worth trying on the large scale?