I checked the resistors and they were all in the right place. (I was being super meticulous, so it was a bit worrying to think that I'd maybe messed that up). I measured a bunch of point to point resistances around the opamp and they all summed to expected values. In the end I cut out the 2835 and replaced with the 2836. It's now showing +130mV offset on both channels - not ideal, but better than the whole volts I was seeing before. On my ****ty scope with no audio playing I'm seeing a 50Hz at around 3mV, plus what looks like a ringing pulse up in the MHz range at a similar amplitude. Audio looks good on the scope when I play something, so I'm going to be brave and hook it up to my (AC-coupled) amp.
Edit: audio sounds good to my untrained ear. I'm a much happier lad compared to last night - didn't really want to have killed 2 out of 3 cobaltmute/tomb kits :)
0.13V offset is perfectly acceptable for the pupDAC ... it may be that the OPA2835 got damaged somehow. Sorry you had to go this route, but you'll be happier with the OPA2836 anyway - it's a far superior opamp. JMHO, but the 2835 was just a silicon prototype for the 2836. When TI/Burr-Brown felt the design was vetted, then they really turned up the performance on it for the 2836.
The ringing pulse could be caused by anything, but if it's in the MHz range, it's not audible. The only other thing I would suggest checking is the lead length on the mica capacitors. Once again, it's the parasitic effects of the lead lengths, but it takes some effort to really get those micas down low to the PCB (like pliers pulling the leads through - carefully). As for the 50Hz, that's probably just a USB-power artifact.
Edited by tomb - 10/7/13 at 5:43am