Abstract Layered mafic–ultramafic intrusions are the fossilized remnants of magmatic plumbing systems and provide excellent natural laboratories to investigate the processes of magma differentiation and solidification. The Rustenburg Layered Suite is the plutonic mafic–ultramafic part of the Bushveld Complex of South Africa and it has traditionally been assumed to have formed from an upwardly aggrading (and in-sequence) crystal pile in a melt-dominated chamber. In this study, we present field and petrological observations, complemented with detailed plagioclase mineral chemistry [molar An, light rare earth elements (LREE) and strontium isotopes] for the first stratiform anorthosite layer (MG3F anorthosite) at the Lower–Upper Critical Zone boundary (LCZ–UCZ) in the eastern limb of the Bushveld Complex. We use these data to test the overarching paradigm of a melt-dominated chamber for the magmatic evolution of the Rustenburg Layered Suite. The MG3F anorthosite is immediately overlain by the MG3 chromitite and both are surrounded by pyroxenite. A distinctive ‘egg-box’ structure, consisting of round pyroxenite blocks mantled by chromitite, marks the LCZ–UCZ boundary, and represents an erosional disconformity at the base of the MG3F anorthosite. The MG3F anorthosite is laterally continuous for hundreds of kilometers in the eastern limb. In the northern–central sector of the eastern limb, the 1·5 m thick MG3F anorthosite is characterized by non-cotectic proportions of foliated plagioclase and chromite chains that lie parallel to the foliation. The MG3F anorthosite is divisible into two sub-layers on the basis of (1) a compositional break in plagioclase molar An, LREE and strontium isotope composition and (2) a peak in chromite mode (up to 12 vol%). In the lower half of the layer plagioclase LREE concentrations increase upward, molar An shows a marginal decrease upward and strontium isotopes are relatively homogeneous (87Sr/86Sr2·06Ga 0·7056–0·7057). In the upper half of the layer, plagioclase LREE concentrations decrease upward, molar An shows a marginal increase upward and strontium isotopes show strong inter- and intra-grain variability (87Sr/86Sr2·06Ga 0·7053–0·7064). Strontium isotopes in interstitial plagioclase in the immediate footwall and hanging-wall pyroxenites show similar 87Sr/86Sr2·06Ga values to the MG3F anorthosite and decrease with distance from the MG3F anorthosite. In the southern sector of the eastern limb, the 4 m thick MG3F anorthosite exhibits identical stratigraphic compositional trends in terms of molar An in plagioclase. We infer that the MG3F anorthosite formed by two successive sill-like injections of magma into a resident viscoplastic pyroxenitic crystal mush. An initial pulse of plagioclase-saturated melt underwent in situ fractional crystallization, manifested as upwardly decreasing molar An and upwardly increasing LREE in plagioclase in the lower half of the MG3F anorthosite. Sill intrusion caused deformation of the viscoplastic pyroxenite mush and vortices of superheated liquid generated by frictional viscous heating caused disaggregation of the footwall pyroxenitic mush. Disaggregated blocks of pyroxenitic mush reacted with the superheated liquid (a hybrid chromite-saturated melt) to produce chromite-rich rims at the base of the MG3F anorthosite (egg-box structure). A second sill-like injection of magma then entered the chamber that halted in situ crystallization. This sill was a plagioclase slurry that contained isotopically distinct plagioclase laths compared with those present in the previous sill. The upward increase in molar An of plagioclase, and decreasing LREE, may be explained by the slurry becoming more primitive in melt composition with time. The second sill also caused mush disaggregation and renewed the production of a hybrid chromite-saturated melt. Chromite crystals were then mobilized and injected as slurries at the interface between the sill and resident mush towards the back of the flow, culminating in the development of the MG3 chromitite. Our model for the development of the Lower–Upper Critical Zone boundary questions the existence of a melt-dominated chamber and it has implications for the origin of stratiform anorthosites (and chromitites) in crustal magma chambers.