There are many answers to your question, a trivial one is to do the Analysis >Deviation>Checking Face to Face between the first Through Curve Mesh and the x-y plane trimmed second Through Curve Mesh. You will see that they are very different. But I think you do not want to know this.
The difference between the two surfaces (and the third one, a swept) consists in the use of these surfaces, the purpose-goal you have to reach.
I attach a file as an example. There are three surfaces, the first one is made of two Through Curve because it must meet the boundary conditions (Proiected Curve), the second one is made of two Swept because there are no boudary conditions.
The third is a Swept in which the Guide Curve are used to define a scale.
Thanks for the illustrative examples! They make me understand Swept much better.
In my original post, my confusion is that, the almost same two sets of meshes produce significantly different surface shapes by Thru Curve Mesh. I think I don't have good understanding of the mechanism Thru Curve Mesh uses to construction surface, especially of the different roles of the primary curves and the cross curves. I remember somebody told me that NX handles mesh surface and sweep surface in different mechanism.
What you're seeing isn't much of a surprise to me. You keep changing the UV directions on surfaces and your posts to me read like you're surprised that the results are different. They should be different because you're specificying completely different sections and guides (more or less). The same thing can happen with Sweeps when you change the section(s) and guide(s).
Depending upon the settings applied in the TCM dialog, primary curves can have more influence on the resulting shapes - think of those as the "sections". Cross strings are more like guides. Also, the resulting shapes are pretty indicative as to why these commands are considered "freeform" - the output is more difficult to control and can be less predictable in some instances - if you wish to control shapes, then you have to add strings (curves) or give the command additional information to help it know how you wish to drive the shape.
Take a closer look at the TCM dialog and expand all of the options on the dialog. There should be a setting under Output Surface Options. Take note of the Emphasis setting and edit your 2 TCM features and see which setting results in the outputs being closer to one another. It is conceivable that changing this setting may have tremendous or very little effect on the resulting output surface in certain situations...it will all depend on the input curves and their quality in some cases.
Also, assigning tangency at the plane of symmetry can also cause shape changes versus creating the entire surface on each side of the plane of symmetry. That's probably why good designers will not mirror a surface through the plane of symmetry when going above G1 (tangency).
This also demonstrates why designers who create high quality surfaces such as Class A tend to make more individual surfaces rather than try to throw a huge surface covering a very large area (think aircraft wings and automotive bodies) - the control they require for the quality of surface they expect cannot be achieved by creating very large slab surfaces in some cases.
What you're observing isn't much of a surprise to me. You keep changing your selection order of the primary and cross strings and seem surprised that the results are different. The results, more often than not, should be different since you're more than likely changing which curves have more influence on the resulting shape.
Unless you've changed settings on the TCM dialog, the Primary strings should have more influence than the cross strings, which results in the differences you're seeing. Also, assigning G1 (tangency) to an extrusion at the plane of symmetry can also change the surface shape a bit (closer to the plane of symmetry).
Take a look at the TCM dialog & make sure ALL options are expanded. Take note of the Output Surface Options. Here you can change the influence of the Primary & Cross strings and you may or may not see a visible change in the results (it will depend on the curve shapes and maybe the number of primaries and crosses being used).
What you're noting in the surfaces is normal and is just part of the whole process of freeform modeling. The results are more difficult to control, can be less predictable and take more time to achieve in many instances. If you were to simplify things down to a 2 curve Swept feature, you might notice the same change in shape if you were to switch the section and guide strings - that's was you're effectively doing with this TCM example.
The results you're noticing are due to you changing which curves are primary vs. which ones are cross which is also effectively changing the UV direction of the resulting surface. Primary strings typically have more influence over the cross unless you change some of the inputs on the TCM dialog.
Let's first look at the TCM dialog. Make sure all options are expanded fully and take note of the option directly below the Continuity settings called Output Surface Options. The first setting can directly affect the output surface because this is where you can choose if the Primary or Cross or Both have more emphasis. If I recall correctly, it's set to Primary by default but I change mine to both. I would urge you to edit both of your TCM features and see how changing this setting to the 2 other options affects both resulting surfaces. The Rebuild area can also result in a shape change if you choose to utilize those options/settings.
Another area that can influence the resulting surface shape is how you choose to model across planes of symmetry. Modeling the whole shape can at times result in a completely different output than modeling up to the plane of symmetry, assigning G1 (tangency) to an extruded surface then mirroring. You will notice this more often when dealing with higher orders of continuity (above G1 which would be G2, G3).
You've probably come to realize a bit by now that freeform/surface modeling can be very difficult to have complete control as well as the output being a little unpredictable at times - that can just be the way things are. You can, however, get used to it and learn a few tricks to help minimize the lack of full control that comes with this type of modeling. There's quite a bit of give and take and you have to be careful to select the best approach. There may be times where you have to start with something not very clean and trim and refine it 2-3 times before you finally end up with a usable mesh of quality curves that you can plug into the TCM dialog and end up with a predictable output that also meets any surface quality or continuity requirements you have. It's also important to realize that since this is all math based, there may be situations that arise where it's just mathematically impossible to achieve a certain level of continuity or quality given certain adjacent geometry or lead-ins.