Thank you for your feedback. That was the whole program. I took 3 samples hits on manual mode, then went for DCC, realized there was a big offset on Y direction from CYL-C, so I added a sample hit on datum B which was over 8 mm. I found one issue, 6 hits might not be enough for a cylinder on manual mode.
I did read 101 again. Manual cylinder does not work, all hits should be in layers, but we can not control hits, no matter 3 or 4 hits in a layer at all.
When taking points insure you do not encroach on the bend area.
Change workplane to X and measure a 3 point circle left of datum 'A' and do the same on right side of datum 'A' Change workplane to Z axis create a line between the 2 circles. Set probe in Y axis at where point 'c' should be and take point on top of tube and do the same for point 'B'
Create offset plane using the 2 circles and point 'C' ( you will have to work out the nominal for the distance between center of circles and point 'C' in Z axis and use that nominal as the offset in the plane for point 'C' You can then use that plane to level and datum in Z axis.
best bet for manually learned cylinder is to use 3 levels with 4-3-5 hits per level, there is no way Pcdmis can muck up the vector that way as only 1 cylinder will ever fit those hits. As for 'control', no need for 'precise' control for the levels, 'just close' is close enough.