Having Ponie and Parrot as two deliverables means we can have twice the faith in the correctness of the code. Without this setup, we might not know if a problem lies with Parrot or Perl 6. Now, we'll be able to isolate some of our problems.
Once I started thinking about the project (and talking with Leon Brocard over gin & tonics), I realized how much automated testing would be required to verify that Ponie works as advertised. I put on my Testing Evangelist hat and started hatching plans for how to make use of the existing Perl 5 tests, as well as the tests for the top 20% of CPAN (following the 80/20 rule), to help verify the correctness of Ponie. Not all modules have great test suites, so I'm going to work with those authors to see how we can beef them up, and increase the coverage, both for the module and for Ponie.
Automated testing is a low-cost, high-payback benefit. Cycles are cheap, and it makes sense to let the machine be watching over your shoulder as much as possible. I'd like to see daily, or even hourly, builds of Ponie to make sure that nothing's been broken.
(Mirrored at http://www.oreillynet.com/pub/wlg/3520)