Darwinian

1. The key issue: needs complete test-suite to rule out incorrect candidates
2. Another issue - is fitness function reasonable?
   1. Results will only apply to production use cases to the extent that your test suite mirrors production usage
   2. Is it easy to setup up for arbitrary project?
3. Looks like any sub-class could be a DSS for an interface ADT? Will that cause trouble (combine with point 1)
4. If the gson etc. is successful, then why will there not be any actual feedback from developers?
5. NSGA-II genetic algorithm-based optimiser
6. Object-Oriented Genetic Improvement for Improved Energy Consumption in Google Guava https://news.ycombinator.com/item?id=18678803
7. Compared w. Cozy
   1. Spec to source vs. source to source
   2. Based on solver vs. based on testing/profiling
   3. Take spec as oracle / take original implementation as oracle
8. I would say it is a low-hanging fruit type of work, but still useful