Getting a new medical diagnostic to market — LessWrong