We investigate the main limitations imposed by existing provenance systems in the development of provenance aware applications. In the case of disclosed provenance APIs, most of those limitations can be traced back to the inability to integrate provenance from different sources, layers and of different granularities into a coherent view of data production. We consider possible solutions in the design of an Improved Provenance API (IPAPI), based on a general model of how different system entities interact to generate, accumulate or propagate provenance. The resulting architecture enables a whole new range of provenance capture scenarios, for which available APIs do not provide adequate support.