Well I wasn't able to spend a lot of time with 3DGS yesterday but so far with the models vertical size down to 250 quants we still have the transparent polygon problem. Sac, last month I had a post at:
http://www.conitec.net/ubb/Forum1/HTML/001678.html where I ran into a problem with model size and the invisible wall. Take a look and there might be some correlation with the problem that you are having where the automatic assignment of say the Narrow Fat is not working. And be sure that you fully understand the difference between a bounding box and the fat and narrow hull and how they will relate in size and configuration compared to your models size and configuration.
And yes the list of questions is not short but it is not long either. I do hope Doug does take the time to answer each question in detail as I think each one is very important to know and to understand. And Sac, if what you said about the near and far camera plane is correct then I don't see how that would apply in this case. What I understood from what you said concerning the Near Camera Plane would be the closest point to the model that the camera would be able to render the model correctly. What Fink and I have found is that you do not have to be right up to the surface of the model for this problem to appear but some distance away.
When I noticed this problem in the beginning I needed to be very close to the model surface and also be at the bottom of the model and then look up to see the transparent polygons. I have found out that if instead of being at the bottom of the model you center yourself with the model in a vertical direction. Now you approach the model and you do not have to get very close to the surface for the problem to appear. And it is not always necessary to look up or down but the problem also appears when viewing in a horizontal direction.
I found that if the monitor screen is overrun at the top and the bottom by about 25 to 30 percent at the bottom and 25 to 30 percent at the top of the screen of the model in the vertical direction the problem begins to appear. To make sure that you are in the center of the model press the "D" key so that you can read your viewing angle. Now with the arrow key look down at the bottom of the model. Keep the arrow key pressed so the bottom edge of the model is going to the top of the screen. When the bottom edge of the model just goes off the top of the screen your angle should be about 56 to 58 degrees. You need this same angle for the top of the models edge, this way you know that you are centered on the model in the vertical direction.
Now at this time I have reduced the model to 250 quants in vertical size. The only position of the camera that will give me a transparent polygon is if I am near the top edge of the model and slightly look in a downward direction. I am going to continue to reduce the model size and see what I can come up with but this weekend has been pretty full.
Now maybe this is telling us that there is a minimum percentage of the polygon that must be in the camera view or otherwise clipping may result. But then again Finks cube does not seem to be large enough where this would hold true.
Now since Doug has mentioned the possibility of polygon clipping we have tried to zero in on this as the problem. What is still curious though is that when I went inside my model to observe the problem for the possibility of the Normals flipping I was able to see the outside texture of the model from the inside on the polygons that were not holding steady. In this case from the inside view of the model, all polygons should be transparent but when you can see your outside texture from the inside of the model on the unstable, flipping or clipping Polygons it gets you to really wonder what the actual problem is. Maybe it is a combination of things where the actual root problem lies with the polygons being clipped for whatever reason but one of the visible cues is an apparent Normal being flipped.
I guess it is really up to Doug with his A4 engine expertise to ultimately solve this problem where we can work and not be hindered by this.
Ron