Hi @Anonymous,
What @michallach81 says is consistent with the information I got back from our core geometry team. The bottom line is that Loft is a picky tool. You have to get it really accurate in order to get good results. I will try to summarize the concerns that the team found with this design that are contributing to the failure of the loft with tangent conditions. Michal's workflow avoids many of these problems.
First, sketch46, which is one sketch used for Loft15, does not have any constraints:

This causes many downstream problems, because, although the curves look like they are OK, they are not quite tangent in all places. These near-tangencies can cause lots of downstream problems for Loft, Fillet, and other Fusion features.
Loft16 has a few smaller problems. Sketch2 is just a bit off-center, and the rails used in this loft are a tiny bit off from the profiles, as well. These might be very small problems, but a lot of small problems can have big effects.
It's hard to see, but if you rotate the model a bit, you can see that Loft16 is not symmetric:

This is probably a result of the above errors.
Loft 18 also does not produce a good-quality surface, again because of the accumulation of errors from above:

So, the reason why Loft19 fails with a tangent condition is also a factor of the problems up to that point. You can see that, even without the tangent condition, the loft is not of good quality:

So, my recommendation is the same as Michal's: You will get much better results if you carefully reconstruct this model. Use projections, and project->intersect to make sure that you have precise sketch points, use edges directly in profiles when you can (don't create a sketch where you don't have to), etc.
Jeff
Jeff Strater
Engineering Director