As part of an example, Abstract and Concrete Categories, section 3.35, claims:
For every two minimal [by number of states] $\Sigma$-acceptors $A$ and $A'$, there exists at most one simulation $A \to A'$, and such a simulation exists if and only if $A'$ accepts each word accepted by $A$.
(A $\Sigma$-acceptor is defined the same way as an FSM is here, with $\Sigma$ automatically filling the language slot. A simulation $A \to A'$ is a function between the states of $A$ and $A'$ that preserves transitions, initial state, and finishing states.)
Suppose $\Sigma$ is $\{a,b,c\}$ and $A$ and $A'$ are languages denoted by the regular expressions a* and [ab]* respectively. The minimal acceptors seem to be as follows:
A:
Q = {S,E}
delta(A, S) = S
delta(x, y) = E
initial = S
accepting = S
A' same, except:
delta'(x, E) = E
delta'(C, y) = E
delta'(x, y) = S
However, none of the four functions $\{S,E\} \to \{S,E\}$ give a valid simulation. Am I missing some sort of condition, or misunderstanding the guarantee?