Henry the optimal thrust direction is *not* tangential.

Ah hah!

Can you suggest, or give a reference that would suggest, a better
parameterized model for pointing during launch? So far I've thought
of:

- point along velocity vector (what I do now).
- point into the wind (air is rotating around earth center, so not
the same as pointing along velocity).
- point parallel to earth surface (orthogonal to altitude vector)

I suspect there are better pointing profiles, but it's tough to
experiment without a model to constrain the optimization space. At
the same time, I'm at a loss to explain why pointing not along the
velocity vector would be anything but less efficient, since
E = F*d
Any force orthogonal to motion would appear to add no energy to the
vehicle. Is there some sort of rotating-frame thing going on here?