I don't think the blame belongs on using "theory-driven mode." The problem is not thinking like reality—for example, most people who think "the house could not possibly have gotten colder" are probably at least somewhat aware that their house's walls don't comprise a perfect Dewar flask, but if "walls" don't happen to be in the same mental filing bin as "thermostats" they are automatically assumed to be irrelevant.
Here's an interesting but very old paper - two theories of Heat Control.
It discusses mental models of home heating systems (thermostats) non-experts use.
These models tend to be extremely wrong from theoretical perspective, but surprisingly useful in practice.
The findings are applicable to a much wider range of subjects than just thermostats, and have certain epistemological significance, especially with regard to compartmentalization.