Stavroulaki and colleagues shift pedestrian analysis from explaining observed movement to predicting flows in streets that may not yet exist. Their methodological strategy is deliberately parsimonious: a small set of spatial predictors derived from street-network modelling is combined through LASSO regression, trained on Stockholm data and tested in central Gothenburg. The iconic contribution lies less in machine learning itself than in the insistence that a planning model must remain usable when detailed behavioural data are unavailable. Prediction becomes an operational test of whether spatial knowledge can travel from one urban context to another. This raises a crucial epistemic issue: portability is not identical to universality, and predictive success depends on disciplined reduction of the city to variables that remain calculable across sites. The study bridges space syntax, predictive modelling and urban design, demonstrating how abstraction can become an instrument for scenario-making while simultaneously revealing the selective character of every modelled urban future.