Sort:  

That's only good when it shows us that "normal everyday data" seems to work.

I was hoping we deliberately send "bad and tainted" data to the nodes, and eonwarped makes a good point.. Having a mixture of v20 and v19 nodes on testnet is best, especially to duplicate the transition period.

On a testnet, you really do need to go wild, and test every possible scenario when possible.

It's impossible to do that... but if you can get "close to impossible", you'll get "close to seeing if the code is nearly perfect" too. :)