Hi RB,
We are 'winging it'. There are about 80 of us world wide with various models of magnetometers. Some people bury their sensors in the ground, some put them in insulated boxes in a basement and the others - well they are rather silent on what they have done. The buried in the ground versions are the most 'stable' but given these instruments sensitivities to other things (like a person walking by them) we are 'guessing'. It is all still very experimental.
I am choosing +/- 0.5c because I cannot see that I could maintain a stable temperature better than that without going to extraordinary lengths. I should point out that we are interested in the relative changes in the magnetic field, not absolute values (we are trying to predict Aurora activity) so shielding the LM335 or associated circuit is likely not necessary since it will form part of the magnetic sensors environment and as long as the environment doesn't change - all is good.
The other point is that I want to log temperature with the changing field values so I can see, over time, the relationship between temperature and field value changes and make adjustments in software. The other advantage I have is that I have 2 professional magnetometers located nearby to compare by results with (one 6km SW and another 7km SE).