Title: A flexible modelling of galaxy assembly bias

Abstract: We use the {\sc Illustris TNG300} magneto-hydrodynamic simulation, the {\sc SAGE} semi-analytical model, and the subhalo abundance matching technique (SHAM) to examine the diversity in predictions for galaxy assembly bias (i.e. the difference in the large scale clustering of galaxies at a fixed halo mass due to correlations with the assembly history and other properties of host haloes). We consider samples of galaxies selected according to their stellar mass or star formation rate at various redshifts. We find that all models predict an assembly bias signal of different magnitude, redshift evolution, and dependence with selection criteria and number density. To model these non-trivial dependences, we propose an extension to the standard SHAM technique so it can include arbitrary amounts of assembly bias. We do this by preferentially selecting subhaloes with the same internal property but different {\it individual} large-scale bias. We find that with this model, we can successfully reproduce the galaxy assembly bias signal in either {\sc SAGE} or the {\sc TNG}, for all redshifts and galaxy number densities. We anticipate that this model can be used to constrain the level of assembly bias in observations and aid in the creation of more realistic mock galaxy catalogues.