I'd bet the Rosetta has at least a 5K input impedance, if not much higher. I don't know. Therefore; ohms law for parallel impedances to come up with a load resistor to make a 600 ohm H (or T) pad attenuate correctly INTO a 600 ohm load.
I don't have one to play with; fully shorted would be max attenuation, but is 'by the book' 10 db, and less at any other setting unless I'm not understanding something. It's a textbook 10 db 600/600 H pad at that setting, meant to reduce 10db between two 600 ohm devices. Change the impedances and you get a different attenuation.
Not sure what you mean about the legs not being the same; four identical leg resistors is indeed the same on both sides. Maybe you have a mod or a different schematic I've never seen.
You might do best to scrap the built-in pad/output control and build NewYorkDave's variable H pad as an outboard box. I also agree with putting a Daven type 600/600 ladder on the output. It's not symetrical, but it works fine and you shouldn't be scared by the idea. You can shell out the big bucks for a Daven T or H, but it ain't gonna make a noticable difference. The ladder has an automatic 6 db loss, and has a more relatively fixed impedance load on it's input side, without having to be loaded like a T or H to get attenuation that matches the settings. Plus, people are scared of asymetrical ladders and haven't driven their cost up....yet. Most are 40db total in 2 db steps.
Even cheaper, more scary, and just as good: Daven 150/300 ladder with your output transformer re-connected for 150 ohms. Transformer re-tap will give a few db more level loss.
