They actually used the testnet a lot, but did it in such a way that it wasn't a completely accurate representation of what would happen once the hardfork was deployed.

They need some kind of parallel blockchain that can function autonomously, not just feed in sample transactions from the real chain, because those were being generated by a different version.

At least, that's the confounded understanding I gleaned by talking to a few witnesses earlier. So testing was there, but wasn't enough, and perhaps not in the right manner.