Adding in temperature to the noise budget.
From this entry (this sinc^2 phase matching curve), I get a temperature calibration of
 ~(43  37.5 C)=pi radians
 5.5 Kelvin / radian or 0.18 rad / K
From this entry we see the level of temperature noise expected.
Sadly, I only trust this curve for one of my temperature controllers, because the resistors (for temperature sensing) are at very different places on the oven, and (probably related) the loop gains which make it not oscillate seemed to be at different settings.
The lowest upper bound which I can get from my data is made of the following:
 The 300 Fast temperature curve (which reflects the settings on one of my temperature controllers  presumably the one which was used in the measurement)
 The calibration above
 A single pole at 0.1 Hz (details on its estimation below)
Pole Freq Estimation:
 From looking at the step response of the oven to a current, I found a pole at 1/60 Hz.
 This is the response from the bottom of the oven to the top (where the temp. sensor is)
 I assume the pole frequency scales as 1 / thermal mass (Volume)
 The bottom piece of the oven (between the TEC and the PPKTP) is 35x as thick as the top (between the PPKTP and the resistor)
 This implies the transfer function between the temp sensor and the PPKTP is a pole @ ~1/60*(4 or 6) which is 1/15 to 1/10 Hz
Note that at 0.1 Hz, my temperature noise is ~4 mK/rtHz ~ 0.5 mrad/rtHz (sqrt(2) added to the obvious calibration). The temperature noise of the other oven could easily differ by a factor of 23 and account for this, or the excess noise could still be some air current left over after the HEPA is off.
