It appears to me that grid positioning is a major problem. This seems to support the idea that we build this system without a grid.
On page 2 you state that Bands 1 and 2 will use a different calibration system.
Are you refering to a "traditional" ambient/sky chopper?
Is the limitation for using the ML device with these Bands the size of the feed aperture?
I had some trouble visualizing how this ML device will work above the receiver from the figures provided in the report. Could you provide some better figures in the final draft?
Do the uncertainties you quote for your ambient and hot loads include variations over the loads? If so, how did you make these measurements (I thought that you had only one temperature sensor on each load)?
Your description of the receiver mixer used to make the 300 and 351 GHz measurements seems inconsistent. On page 9 you describe it as a UVA 6-junction mixer, while on page 11 you say that it is a 4-junction mixer.
Given the good correspondence between the load temperature measurements at a variety of temperatures between 50 and 87 C, should we consider operating this device with a variety of Thot values (i.e. Thot1 at 90 C, Thot2 at 50 C)? This might also allow for the determination of a correction for the receiver drift (similar to what you did with Thot, Tamb, and Tcold on page 16) over shorter timescales (driven by the temperature settling time of the load).
What is the temperature settling time of the hot load when its drive temperature is changed?