StoppingRule

Mechanism

To make a design robust without predicting the future, imagine many possible ExogenousStress events, including absurd ones, and change the design so it survives each, integrating the changes into one structure. The question is when to stop. A statistical stopping rule assumes a distribution and stops when estimates are accurate enough; here no probabilities are assigned, so that rule cannot apply. Instead, after integrating one round, test the design against a new round of unseen simulated stressors. Complex systems tend to settle into a small number of stable states that each survive many different inputs, so a design that survives enough stressors starts surviving ones it has never met: it has generalised, or reached criticality. When most of a fresh round is survived, stop. Cost is the second dimension: integration stops earlier if adding more costs more than it protects, and building partly integrated functions, as a FractionalMetaModel does, lowers that cost so the generalisation point can actually be reached.