I ran a simulation with quite a narrow beam, scattering switched off; (and in fact too high energy, not that it should matter). I used the 7417 geometry with all of the fields (including quads) set to 0. I plotted the difference between truth and recon. I can see a clear misalignment at the few millirad level in dx/dz and dy/dz residuals. I presume that the correction to rotations etc is not applied correctly.

I attach residual plots (virtual hit - sf hit).

I am using:

 $> bzr revision-info 
1105 phuccj@gmail.com-20160607075611-09902p7yrnu6gsqj
 $> bzr info
Checkout (format: 2a)
Location:
       checkout root: .
  checkout of branch: bzr+ssh://bazaar.launchpad.net/+branch/maus/merge/

I made a few changes:

• Did a bzr update; scons
• Made the input beam distribution a bit bigger, in case I was getting bias from "unlucky" combination of fibres
• Made the beam energy 200 MeV, not that it should matter
• downloaded the latest 7417 geometry, as of this morning, leaving quads on full current (i.e. not touched the geometry)

I still see a 1.5 mrad bias in dx/dz; this makes 4.5 mm over 3000 mm distance between the trackers, which is obviously significant. Plot attached as residual_xp-v2.png