The modern synthesis of evolutionary biology uses a classical model of causality as constraint. Populations develop as they do because failures are either suppressed or eliminated by natural selection. The result is a negative feedback loop; a trial-and-error dynamic that means the only mechanism available to generate novelty is chance mutation. Darwin understood that that evolution could not have occurred without a complementary positive feedback system; a 'trial-and-success' dynamic (Lorenz 1977) capable of generating new patterns and amplifying them to create new attractors. Purposeful action and selective co-operation were key features of Darwin's theory. The modern synthesis struggles to accommodate concepts like agency and emergence, not because they are difficult to understand, but because of an informal consensus that the purpose of scientific research on human evolution is to make the past seem predictable. Innovations are irreducibly unpredictable.
In mathematical systems-modelling, the words 'validation' and 'verification' are used in ways that clarify this expectation. To validate a computer model, for example, would be to ensure that it actually simulates the causal mechanism the scientist has in mind. With complex, multi-agent models, validation is a non-trivial task because these operate at or near the theoretical limits of computability. Once the model has been validated, the next step is to verify it empirically by using it to 'predict the past', as it were.
Verification is relatively straightforward in situations where one is simulating the workings of a well-designed system that has been created for a specific purpose. The concept of verification becomes a philosophical minefield, however, when one is studying ecosystems where social learning, habits and purposeful co-operation shape system dynamics. Accidents of history and geography and expedient, small-scale patterns of co-operation can generate new patterns of learning and forgetting that change the course of history in an utterly unpredictable way. These self-organising events can be explained ex post, but not predicted ex ante because they are contingent on knowledge we do not yet have and habits we have not yet begun to acquire.
Any systems model that slavishly predicted the same historical trajectory every time we ran it could reasonably be said to have been invalidated - it has been over-constrained in a way that forces it to simulate one historical trajectory (the history that actually happened) when we have strong reason to believe this time-series was one of an unbounded set of possible histories, many of which we cannot even imagine.
Plesionic complexity places scientific anthropologists in a difficult bind. Either they repudiate conventional, constraint-driven models of causality and abandon hard-science method, or they push plesionic complexity beyond the scientific pale. There is scant help for them in pop-science literature about chaos and fractals or critical humanism because so much of that literature treats narrative explanation as if it were a causal mechanism. Butterflies do not cause hurricanes. If they did, scientists could predict hurricanes by monitoring butterflies and prevent hurricanes by controlling butterfly behaviour. In the same way, individual human beings do not cause genocidal wars or social exclusion. Institutional constraints and power-relations constrain human agency in ways that cause both the social exclusion and the tenor of contemporary discourse.