Any idea on how to prevent it?
If a Baylis-Hillman-like reaction is operating, switching to a less nucleophilic base (e.g. DIPEA) should stop/reduce it.
if I add water to the mixture I get my starting conduritol and its epimer
Why are you adding water? Is this in the workup? MsCl and Ms
2O are both water sensitive, nobody claimed they weren't.
Another explanation is S
N1 and/or S
N2' possibly involving neighbouring group participation by a benzyl ether. It might be useful to know the relative stereochemistry of the four centres if you can provide it.
In that case it might be better to switch to a tosylate as it's not as good a leaving group, so S
N1 pathways will be reduced. Not sure if that's really the cause though.
If you purify and take the mesylate on to the next reaction (which I assume is a displacement, can you tell us?) do you get complete inversion of configuration?