We use observations of heavy elements in very metal-poor stars (Fe/H < −2.5) in order to place constraints on the viability of collapsar models as a significant source of the r-process. We combine bipolar explosion nucleosynthesis calculations with recent disk calculations to make predictions of the observational imprints that these explosions would leave on very metal-poor stars. We find that a source of low ( 0.1-0.5 ) Fe mass, which also yields a relatively high (>0.08 ) r-process mass, would, after subsequently mixing and forming new stars, result in r/Fe abundances up to three orders of magnitude higher than those seen in stars. In order to match inferred abundances, 10-103 of Fe would need be efficiently incorporated into the r-process ejecta. We show that Fe enhancement, and hence r/Fe dilution from other nearby supernovae, is not able to explain the observations unless significant inflow of pristine gas occurs before the ejecta are able to form new stars. Finally, we show that the inferred Eu/Fe abundances require levels of gas mixing that are in conflict with other properties of r-process enhanced metal-poor stars. Our results suggest that early r-process production is likely to be spatially uncorrelated with Fe production, a condition that can be satisfied by neutron star mergers due to their large kick velocities and purely r-process yields.