I can't believe Blizzard doesn't automate testing of this stuff. When we can hit target dummies and see changes, they must be able to automate the same thing... take every class and cookie cutter spec - have them dps against a target dummy for long enough that random crap like crit strings average out. Apply a reasonable set of 5, 10 and 25 man buffs - rinse and repeat. Do some variations on the specs... rinse and repeat. Do some variations on gear (or put people all in T7 and T7.5)... rinse and repeat. This can all be scripted and run in parallel on multiple machines - it's not like they need hundred of live bodies to do this.

It's silly that a player can hit a target dummy and see the difference yet Blizzard actually releases this and then goes... oops, our bad, we missed that.

Its not inefficient, its a common programming problem-solving approach. If you think a number is too high, you reduce it by 50%. If the result seems too low, increase it by 50% of its new value; too high, reduce it by 50%. This approach reduces the probability space of the "right" value by half each step, and is known to be the best way to arrive at an unknown value from an arbitrary starting point.

It's known as a binary search:
http://en.wikipedia.org/wiki/Binary_search

Well I am a software architect and I can tell you that using a human-driven binary search with iterations of ??PRODUCTION RELEASES?? for this is the most ridiculous thing I have heard in a while.
Wait...I am in software development. that last bit was a total lie.

Regardless. I agree with other posts on running automated tests of some sort. Hell, I think there is a program/website that does just that already?! (albeit without ALL the numbers - blizzard has the numbers)

@eden... yes... but you don't SHIP like that. You make the change and test. That's what I don't understand about this - testing should have shown that BM dps was too low with these changes... so why weren't more changes made to bring it back up?

No one's saying they should be perfect the first time, but they should iterate until they're close to the goal. If they're just going to make changes and not catch this in test or listen to PTR feedback, then... what's the point?