Hi Andrew I have not really used extend yet, but have it in mind for some future projects. I am hoping to use 'revise' to allow multiple extended elevation projections of the same piece of cave and to move the 'extend' out of the survey data and into the map definition where it more logically sits if one is keeping map structure independent of survey structure. I digress. I had a play with your hazelnut file with v5.3.8 to see what would happen. I got exactly the same outputs as you did. I read the Therion book on extend, and figured that although I agree, there is quite a bit of unexpected behaviour as you described, the extend normal or reverse before the first station did not make sense as there is no previous station. See description on pg 21. So that aberration at least could be excused. The examples near the end of the Therion Book are also quite helpful. Maybe Therion needs a bit more specification to help it along with the hazelnut survey? For example... extend start 2.0 extend normal ...before the first data line makes it do what you might expect, although possibly not for any logical reason. Similarly... extend right 2.0 2.1 ... before the first data line makes all the legs head right correctly. So there are some aberrations, but some lateral thinking seems to make it work out quite efficiently. And I agree, PocketTopo does seem to make a better stab at where the splays should be drawn. Maybe the algorithm could be borrowed from Beat? Bruce