I was looking through Rovelli's Relational QM paper

http://arxiv.org/abs/quant-ph/9609002" [Broken]

and Unfortunatly I didn't find the initial set-up of the story compelling. In this set up, there is a system, S, and an observer O. S can be in one of two states [itex]|\psi\rangle =\alpha |1\rangle + \beta |2\rangle [/itex]. Assume that for a specific measurement at time [itex]t=t_2 [/itex] O observers S in state [itex]|1\rangle [/itex]. Now let a new observer, P, describe the S-O system quantum mechanically. Then without interacting with the S-O system, it is described by P as [itex]\alpha |1\rangle\otimes |O1\rangle + \beta |2\rangle \otimes |O2\rangle [/itex] where O2 and O1 indicate that O has interacted with S and observed result 1 or 2 respectively. Then from this Rovelli states,

and then that

But to me this conclusion and argument aren't particularly compelling. To me this says that O has a value for S, say 1, but P doesn't have a particular value for the S-O system. That's okay with me, since P never made a measurement in the first place, I'd certainly not expect P to have a value at hand. This seems just like if Sally and Oscar have a box in front of them with two balls, red and green, and Oscar picks a ball at random and looks to see what it is, but Sally can't check either the box or Oscar. Then Oscar has a value for the ball color, but Sally doesn't know the ball color, or what Oscar knows as the ball color.

I don't see a problem with this... This seems completely normal and non-exclusive. Can someone help me see the profoundness in this example.