The Predictive Realistic Image Synthesis project is a testbed for global illumination algorithms. The goal of the project is to build an infrastructure capable of handling complex models and materials and allowing the exploration of both finite element and Monte Carlo algorithms.
Originally, the goal of PRIS was a system capable of handling many (well 3-5) developers working on a wide range of rendering research. This meant trying to make everything modifiable and to make these modifications transparent to the other users. This strategy added a large amount of complexity to the system with minimal payoff unless those areas were actually modified. The flexibility and generality of the interfaces still weren't flexible enough in the areas we were doing research on, yet and the complexity increased the overhead of using other parts of the system.
The new goal is a skeleton that can handle the IO and other basic requirements of graphics, and gives reasonable functionality for exploring primarily global and local illumination algorithms. The resulting system is less able to handle multiple people making modifications, but much more able to support major changes to a few core components.