I think between Simon's post and the link posted by Richard, you've pretty much got the main issues covered for a meaningful conversation with CycleStreets.
Is the study for pedestrians or for several travel modes? I ask because one issue to consider is to ensure the link weights used by the routing algorithm optimise by time and not distance, as minutes of exposure is likely to be the key metric here. And of course, the time taken to navigate different road segments is very much mode dependent (pedestrians being largely inelastic wrt vehicle density, car drivers much more elastic).
If your modelling is temporally dependent (e.g. acknowledging that rush hour pollution levels show a different pattern and concentration to inter-peak periods), then this would add another level of complexity to the routing algorithm. It may be that you would have to simplify into peak vs non-peak periods rather than try to model the transition of a longer journey though the peak-time bulge.
The other potential problematic issue is data quality, or more specifically, the distribution of your pollution sensors/modelling. For patchy pollution data, your routing can become very dependent on the (arbitrary) weights you attached to missing data. Allied to this is the comparative weights you attach to pollution concentration vs time of exposure. If time of journey is weighted too highly, you just get a standard routing results, too low and you get long tortuous routes that avoid any hint of pollution. I am not aware of any work that provides much guidance on these relative weights for typical journey decisions, but your local geography colleagues may possibly be able to help there.
Sounds like an interesting project. Good luck with it.
