Just solved a nasty algorithmic problem with TCR by making sure (0,0) worked, then (0,1), then (1,2), and on to (n-1,n). Kept me from a) being overwhelmed by too many constraints at once and b) breaking things that were already working. Complexity partitioning rules!
I wish we had PBT back when we were exploring XP/TFD at IONA... the shrink phase is inspired ... the one insight we missed that makes PBT scale a few orders better. If only the Smalltalk/Haskell folk had more overlap we'd've saved a decade of dead ends as an industry...
