I am having some trouble understanding the distillation of EPR pairs using a stabilizer code. This idea goes back to the paper by Bennet, DiVincenzo, Smolin, and Wooters.
The idea (I think) is that $k$-EPR pairs can be distilled via an $[n,k]$ stabilizer code $S$. In the protocol two parties share (noisy) $n$-EPR pairs, the code can distill (perfect) $k$-EPR pairs. That is, the decoding and encoding unitaries of the code imply that there is a way in which the parties can teleport a $k$-qubit state meaning they must have (perfect) $k$-EPR pairs. It also seems to be the case that if a correctable error occurs when the $n$-EPR pairs are initially shared it can be corrected.
In particular, the brief Wikipedia article on this subject (which appears to be verbatim from section D. of arxiv.org/pdf/0708.3699) claims that after Alice applies her codespace projection (via measuring the code generators), which via properties of the maximally entangled state applies a similar projection on Bob's side
Alice restores her qubits to the simultaneous +1-eigenspace of the generators in $S$. She sends her measurement results to Bob. Bob measures the generators in $S$. Bob combines his measurements with Alice's to determine a syndrome for the error.
My confusion is understanding the details here. How does Alice "restores her qubits" and how does "Bob combine his measurement with Alice's to determine the error"? Perhaps I am just misunderstanding what's going on in the protocol, but maybe someone here can help me out?