..so how does the camera know if it's looking at a real source or a reflected one?
... it doesn't?
(...)
Let's say you measure the temperature of a polished gold sphere with an emissivity of 0.03. Then 97% of the reading would be of the temperature of the environment, and not the gold sphere.
So if the gold sphere was 100 'C and the environment reflecting on the sphere was 0'C, the temperature reading would be way too low, maybe between 3 and 4 'C.
If you set the emissivity in the TIC to 0.03 and the Background Temperature to 0'C, I believe the idea is that the reading of the TIC would be correct (100'C). When looking at a circuit board with dozens of different materials, it is not possible to set the emissivity correctly for more than one of the materials. So, I guess the usefulness of the Emissivity setting is limited, and should be left at something like 0.95 normally. Then you evaluate simply put into the equation that shiny metal readings are not likely to be correct, and put a piece of tape onto these objects or paint which has high emissivity.
However the shiny metals we may want to measure the temperature of in a circuit (copper trace or - cable) often conducts (by contact) it's heat to nearby materials with higher emissivity, i.e. the PCB or a cable insulation. Then the near by PCB or cable insulation would be close to the real temperature of the copper trace or -cable.
The only thing that has an emissivity of 1.000000000 and would give a 100% correct reading on a TIC is a theoretical abstraction called a Black Body.
Here is a link to an emissivity table I found after a quick search:
http://www.engineeringtoolbox.com/emissivity-coefficients-d_447.html I have not checked how close it is to other tables by Fluke/Flir etc.