I've been working more on this shroud (http://communities.ptc.com/thread/56128) and now I'm trying to figure out how to control the boundary conditions of a style surface.
The boundary conditions at my edges where I have control curves are good (i.e. they are curvature continuous and normal to the plane, but in between control curves the surface does not keep that condition.
This becomes obvious when I thicken the surface with a normal offset. If the surface were normal, the part of the edge would be coincident with the horizontal red line (surface edge).
All of my edged except for those on the inner circle have edge control arrows. Why don't the arcs?
ACCEPTED SOLUTION
Accepted Solutions
Ok, making progress.
First thing I found (and my first mistake) was thinking that the wireframe curves would be sufficient for defining the curvature was wrong.
Second, I found that the surfaces (even if the part won't actually have those physical surfaces) adjacent to the smooth (G2) surfaces must be defined. In my case, these were the two cylindrical surfaces that represent the final smooth airflow direction and the 'rim' type surfaces perpendicular to the airflow.
With those surfaces, I could then go and define the end points of my styles curves to be Continous.
Then I could go and make connections between the style surfaces and my reference continuos surfaces by selecting the broken line indicators.
If more than one surface was possible, then you would get a prompt like this (asking if the red surface was the correct surface to connect to).
Once those connections were made, then I could set those connections to be surface continous (they defaulted to tangent).
Still have some investigation to do and cleanup, but this looks like a solid solution.
Ok, making progress.
First thing I found (and my first mistake) was thinking that the wireframe curves would be sufficient for defining the curvature was wrong.
Second, I found that the surfaces (even if the part won't actually have those physical surfaces) adjacent to the smooth (G2) surfaces must be defined. In my case, these were the two cylindrical surfaces that represent the final smooth airflow direction and the 'rim' type surfaces perpendicular to the airflow.
With those surfaces, I could then go and define the end points of my styles curves to be Continous.
Then I could go and make connections between the style surfaces and my reference continuos surfaces by selecting the broken line indicators.
If more than one surface was possible, then you would get a prompt like this (asking if the red surface was the correct surface to connect to).
Once those connections were made, then I could set those connections to be surface continous (they defaulted to tangent).
Still have some investigation to do and cleanup, but this looks like a solid solution.
Be careful with mirrors. I am seeing some interpretation going on when you mirror with some assumptions. not every feature truly mirrors! I noticed this when I used curvature analysis but it was clear with the knot analysis...
the left is a mirror of the right where a C2 curvature was chanded to a simple arc.
This is actually a fairly disturbing development. the left side is the mirrored feature.
Both sides are treated equally. there is no reason for this to have developed differently. Each mirrored feature was investigated for commonality.
Wow! THAT is very disturbing! Hey, PTC, I EXPECT things to mirror perfectly. I EXPECT pattern geometry to work perfectly. I do NOT expect this kind of garbage from what is priced as premium CAD software.
I've seen funny things, and geometry not working right, but never expected it to be this bad. Thanks for the in depth analysis Antonius! PTC should be paying you for developing their software for them...... 
This is scary indeed. I remember arguing the very opposite problem to this one with helical sweeps...
Thank you for the tip, Tom. I never knew you had the option to select only the quilt without its definition. Of course, smart is the default selection filter.
However, the right edge is a C2 blend giving the course rooster tail. Notice how it still softens this edge closer to radius. PTC is known for its surfacing prowess but my recent experiences say that it does a lot of fudging. Not as bad as the mirror of the geometry, but still something to note for those relying on curvature blends.
This is the same mirror with a quilt. Mirror is highlighted in green.
Hmmm, I have a boundary blend I'm using on a handle for our new handpiece, and I simply mirrored that and merged them....and am having problems with the shell. Now, seeing this garbage, I think I'll make 2 separate boundary blends and merge that. Thanks for the analysis Antonius, It'll make me look closer at mine!
It is odd that the mirror looks smoother. Accuracy seems to be affecting this as well. (Mirror is shown in green.)
Relative Accuracy - 0.001
Relative Accuracy - 0.00001
Absolute Accuracy - 0.000001
You would think that the programmers would make this analysis instead of having to have the users do this.
Antonius - How much do you get paid from PTC for all the bugging, error fixing work arounds and miscellaneous things that you do? 
...why would PTC pay for their programmers to do then when they can get customers to PAY THEM for the software and do it for them for free???
I usually notice that surfaces are not exact copies or exact mirrored copies by looking at their isocurves.
The right side is the mirrored side, and as you can see on the picture it's topology is slightly different.
It'd propably be better to avoid mirror feature and use copy and paste special transformation instead, or maybe FMX mirror would do the right job. Not sure.
One question, how does the Flip Normal feature work in Creo Parametric? I can't seem to be able to edit it's definition.
If an industrial design guy saw this, he might have a fit
Did you mirror the quilt instead of the feature, Jakub?
I used one of your models, Tom. It propably doesn't matter, the mirror feature is just all wrong.
If the designer poked one of his eyes before looking at the picture then yeah he might have a fit
