Managing the unpredictable is itself predictable. A startup is defined by extreme uncertainty (A startup is defined by uncertainty, not size) — it would seem that no theory is possible here: what can be taught where outcomes are inherently unpredictable? Yet Ries demands testable predictions from a theory of entrepreneurial management — and supplies them.
The paradox is resolved by separating levels. The theory’s predictions are not about business outcomes (whether it will take off or not), but about an organization’s behavior in the process. Ries’s example: the theory predicts that cross-functional teams switched to learning milestones will complain about declining productivity — the familiar measure of progress (features, releases) has been taken away, while the new one (knowledge) is not yet felt as work. The predicted resistance is testable in any team adopting the method. Market uncertainty does not contaminate the process level: people’s and organizations’ responses to the method’s discipline are reproducible even when the product’s fate is not.
Popper’s criterion (Scientific theories are unprovable, but falsifiable): Ries’s theory can be falsified. Teams adopt learning milestones — and no complaints about productivity arise; the method is applied with discipline — and no knowledge is produced: such observations would count against the theory. This distinguishes a theory from a set of recommendations compatible with any outcome — cf. Deutsch’s apt explanation: in a bad explanation, details can be changed without changing the predictions, so testing is useless.
A response to the genre critique. Lean Startup is based on a genre it itself debunks states that the method’s key evidence is the success story of IMVU — that is, a genre Ries himself debunked. Process-level predictions are what the method can offer in response to this reproach: the prediction about teams’ resistance is testable independently of the IMVU story. The rehabilitation is partial: the predictions support the theory of the process, but the book’s promise is business success, which remains at the outcome level. And there is a second boundary: the falsifiability of the process does not extend to the method’s apex — vision remains an untestable core (see The vision-strategy-product hierarchy is built along a gradient of testability).
The Cynefin parallel (02. Connectivity and Complexity - Cynefin): in a complex domain, outcomes are unpredictable, but the working protocol is prescriptive — experiment, small stakes, retrospective sense-making. Process predictability amid outcome unpredictability is a general property of the complex domain; in Ries it is carried through to falsifiable formulations.
How it supports Startup success can be constructed - the right process can be learned
It removes the objection that “there is nothing to teach in a domain of uncertainty, because no theory is possible there.” The teaching is not to predict the market — it is to run a predictable process in an unpredictable domain; what is transferable is precisely what is predictable. Without separating the levels, the claim that success can be constructed is vulnerable: any startup failure would look like a refutation of the method.