I created a very rough model of the LM336 (emailed to Frank and Jan to try to improve) to see if the noise of the LM336 might explain the noise at lower frequencies. I think from this rough sketch that might be the case, but again, it was a very rough model. This shows the new model with the instrumentation noise and measured noise. The instrumentation noise probably explains the higher frequency discrepancy - I will test that later today - and there's a chance that the LM336 explains the lower frequency discrepancy.