Thanks for the input everybody.
Unfortunately the magnet can't be magnetised before assembly. The strength of the magnet just makes it impossible to handle.
Neither can we demagnetise after a "trial run". For reasons I don't quite understand, our demagnetising machine (pulses of alternating polarity and diminishing magnitude) won't touch them. Annealing is also off the table because, as someone said, it spoils the plating.
The remark about a preferred magnetising direction is interesting though - I'll look into that.
After a discussion with the engineer who's field of expertise this is, it occurred to us that this may not be a problem with the neodymium at all - it could also be a variation in the steel that also forms the magnetic circuit. We're investigating that, too.
To TomKatts comment about QC: Absolutely! This is a QC issue somewhere up stream from us,
and in the long term someone will have to take responsibility for having supplied substandard materials. But, as usual, we have a huge stock and we need to keep production moving or everyone in the factory might as well go home. So it's just a temporary work-around until we can prove what's wrong and get the responsible party to fix it.