For example, a well-detailed hand probably needs more edge loops than does a pressed shirtsleeve. So is there any reason to make sure these kinds of things actually connect, vertex-to-vertex? In this example, it seems better to me to just extend the edge of the wrist into the sleeve, right through some polys that you’d use to cap off the edge of the sleeve. That’s a little messier where the pieces meet, perhaps, but the overall poly count could be kept lower. Any advice?
The amount of polys you’re going to add by doing stuff like that is really insignificant unless you’re going for the absolute rock bottom of technology. That said, people use intersecting geometry all the time if it makes sense/makes things easier. Go for it.
Edge flow is more about having a clean mesh for animators, the more f***ed your edge flow is the more problems can arise during the rigging animating process and also animations may have strange deformities.
Personally I just do it because sometimes you just gotta add loops although I tend to optimize as much as possible here and there through re-topology or other means.
Alec nailed it.
Edge flow is mostly an animation term.
While it can be related to gaming, it’s mostly related to film and video animation where close up detail is important.
For gaming, you’re going to be working with mostly low-poly / triangles so it’s a bit harder to get cleanly bending geometry with such low resolution meshes.
If you can hide the fact that the hand is seperate from the body by using the shirt sleeve then sure, no reason not to have them seperate or intersecting as long as it doesn’t make the animations or control of the character more complex than if it was a unibody model.
-Will
Thanks for the advice, people. I’m going to be doing some animating soon, and see how it goes when using some models that I create with extreme polygon efficiency in mind.
and dont be afraid of tri’s. We all should know by now that game engines convert all models to tris.
While thats true, its usually better to avoid triangles until the very end when everything is converted at once. Modeling with quads usually ensures that things deform better when animating and if you ever use subdivision modeling for high poly models, tris dont subdivide very well.
What I ment in my previous post was in the context of ultra-low and low polygon models, sometimes it’s even better to have tris such as in knee and ankle meshes.
I have found that triangles should only be used to “fork” polygon flow.
Again thats another use of it for when things such as hands and heads don’t match up but can result in some nasty deformation.
Edge flow is a synonym of Poly extrusion, isn’t it ?
If it is, poly extrusion is what I use for human low-poly modeling, and I came to a stunningly fast workflow with this method 
It leads me to 500 polys chars easily.
Starting from the Left feet, then going up to the hips, torso, and arms, and then mirroring.
At the beginning, it used to take me 8 hours to come to a correct quality model, but after producing 3 characters, this work length is cut by more than a half.
intersecting geometry is not a big deal if your character is well skinned and there is enough geometry around the wrist area of the shirt and the wrist itself so when it bends the underlying mesh does not pop through itself.
with regards to triangles I think people really started to get hung up on quads. Tris are perfectly fine as long as your model is efficient, clean and the silhouette is good and there is enough geometry around animating area to hold the shape. The only area I try to avoid tris is if I have a face which will need to animate…
I find tris can also useful when animating hinge joints on models for example a finger or an elbow. A good example would be on an elbow I see a lot of people do the 3 loops around the joint, when bending this will cause the geometry to collapse on itself where if you was to use 2 loops have an extra edge over the elbow which ties into these 2 loops with tris the bend will not collapse and the elbow will retain its shape.
did that make sense?
with regards to high res and low res I tend to use to separate models so I don’t have to worry about changes on one affecting the other. This allows me to work without restriction. I used to always try and keep them using the same mesh but I found it very restrictive and found keeping them separate gave me a lot of freedom.
I am sure there may be a lot who disagree or have other thoughts on this I guess at the end of the day it is what is going to work best for your personal pipeline and what methods of working you are most comfortable with.
Good discussion all around!
I can highly recommend this book for those interested:
-Will
Or head over to the gnomon workshop for they’re amazing stuff! 
not to be pluging…