While modern agriculture relies on genetic homogeneity, diversifying practices associated with seed exchange and seed recycling may allow crops to adapt to their environment. This socio-genetic model is an original experimental evolution design referred to as on-farm dynamic management of crop diversity. Investigating such model can help in understanding how evolutionary and adaptive mechanisms shape crop diversity submitted to diverse agro-environments. We studied a French farmer- led initiative where a mixture of four wheat landraces called 'Mélange de Touselles' (MDT) was created in 2001 and circulated within a farmers' network. Fifteen of these MDT subpopulations were sampled in 2008. Each of them were simultaneously submitted to diverse environments (e.g. altitude, rainfall) and diverse farmers' practices (e.g. field size, sowing and harvesting date, etc.). Twenty-one space-time samples of 80 individuals each were genotyped using 17 microsatellite markers and characterized for their heading date in a 'common-garden' experiment. Gene polymorphism was studied using four markers located in earliness genes. An original network-based approach was developed to depict the particular and complex genetic structure of the landraces composing the mixture. Rapid differentiation among populations within the mixture was detected, larger at the phenotypic and gene levels than at the neutral genetic level, indicating potential divergent selection. We identified two interacting selection processes: variation in the mixture component frequencies, and evolution of within-variety diversity, that shaped the standing variability available within the mixture. These results confirmed that diversifying practices and environments maintain genetic diversity and allow for crop evolution and adaptation in the context of global change. Including concrete measurements of farmers' practices is critical to disentangle crop evolution processes.