Assuming this one is the question we should all be answering, here goes one attempt to do so:

Thanks for supplying the impact criteria table, Yongqing. This is interesting, I've not seen anything like this before, but I must say this is a real challenge to implement! Reason being, reflected glare such as we are trying to quantify here is view dependent. So, while this table clearly places luminance values into several classifications, one must consider the viewing location and direction of the evaluation. Further, the evaluation must be conducted at several points. The questions are, how many of these points must be sampled, and which viewing directions/angles? Without strolling into that minefield, let's just look at how one would calculate these values using Radiance. This is actually somewhat easy:

1. Make a model of the proposed building and surrounding structures, taking care to specify the curtain wall materials accurately (Radiance's 'glass' primitive will do an OK job of modeling the reflectance, but there are better materials to use if you have the angle-dependent data and time to model them)
2. Create view points you wish to calculate
3. Create sky model(s) representing the solar incidence angles you wish to evaluate
4. Create Radiance renderings of (2), using (1) and (3).
5. Evaluate the images for glare potential.

You could use Ecotect for (1), and I'm pretty sure Ecotect can also export views for (2), but you may need to modify them on your own; see the rpict manpage for view specification details. Finally, there are a couple of image analysis utilities in Radiance that could be used to find luminance values in the image. pextrem will tell you the highest and lowest pixel values in any Radiance image. Since a standard Radiance image is the radiance of each pixel, you could simply convert the radiance to luminance by applying v-lambda, and compare that value to the "impact table" you have supplied, above. Findglare may also be useful; it is a script that evaluates a Radiance image for glare potential, just using different metrics.

I am still unclear on how the glare problem is defined in terms of the number and location of evaluation points and view vectors, but assuming those are known, this is how I'd use Radiance to evaluate them for glare per the criteria above. I could envision a script that takes in all the rendered images, runs them through pextrem, and reports the glare metrics.

HTH!