Reply from Glen,
Still looking into this- the only problem is with the QD0/QF1 multipoles- mainly the sextupole component. If the multipoles are added with an otherwise perfect lattice- effect is to blow x and y IP sizes up (y size becomes ~90nm). I tried to re-match the sextupole strengths to reduce the usual 2nd order terms at IP, but this didn't make much difference in the tracking simulation (despite MAD claiming IP size was ok)- must be higher order terms, and therefore need full FFS design process...
As you can see, if I just run the tuning simulation anyway it seems to largely compensate for this, but not perfectly...