In general, asymptotic matching is used to find an approximation to a solution to a (system of) differential equation(s). The uniqueness of the underlying, 'real' solution can be obtained from general theory on existence and uniqueness of solutions to differential equations. In particular, under certain general conditions, an initial condition gives rise to a unique solution.
Asymptotic matching can now give more explicit knowledge about that unique solution. So, we know it exists, we know it is unique, but we don't know what it looks like -- and that's the point where you start to use asymptotic methods.
A restriction (or, actually, a feature) of matched asymptotics is that the underlying solution is approximated up to a certain order in the small parameter of the problem. That means that, for 'nearby' initial conditions (i.e. initial conditions so close to the 'real' initial condition that they are all identified on the 'level of magnification' employed by the asymptotic method), the associated solutions are indistinguishable from the matched asymptotics point of view. So, every solution approximation you construct using matched asymptotics is actually a 'tube' of nearby solutions. You can make this 'tube' smaller by zooming in, i.e. carrying out the matched asymptotic method to higher orders in the small parameter, but this still does not solve the problem entirely. In particular, you can have solutions which are exponentially close, for which you would need so-called asymptotics 'beyond all orders'.
For more information on techniques dealing with these issues of approximation and uniqueness, I highly recommend you consult
C. Kuehn, Multiple Time Scale Dynamics, Springer (2015),
ISBN 978-3-319-12316-5 .