Thanks Eden, this is immensely helpful. The FITS that Astrometry.net gave me is here (900 Kb):
https://dl.dropboxusercontent.com/u/...lateHadar.fits
I checked the FITS header (using Fitswork) and you are correct that all of the fitting information is supplied there, so I suspect this is indeed giving AT a head start.
Now although I'm trying a blind plate solve (from a software perspective), I actually know that I was pointing my astrograph just near Hadar (which is just of the FOV), so I am certain that the original Astrometry.net solve was in the correct position. Here is the location of the solve page:
http://nova.astrometry.net/user_images/344663#annotated
I also tried it blind on an image of M20 I took with the same CCD (high quality stacked 25 x 60 sec etc.) and this also works fine on Astrometry but fails on AT (unless I use the Astrometry.net FITS):
http://nova.astrometry.net/user_images/345065#annotated
My subsequent AT solve also puts it in the right position, and solves in 13 seconds. By the way, here are the AT settings I used for the successful solve on the Astrometry FITS - it is tuned to my setup.