I am trying to show IND-CCA1 does not imply NM-CPA. From what I have read the "classical" proof goes by taking an IND-CCA1 scheme $E$ and modifying it so that the encryption of the inverted plaintext is appended to its ciphertext: $$E': m \mapsto (E(m), E(\bar{m}))$$
It is clear $E'$ is not NM-CPA since just swapping the two ciphertext parts leads to a non-trivial relation between the two plaintexts (i.e., inverting it).
But how do we show $E'$ remains IND-CCA1?
My work so far:
Assume $E'$ is not IND-CCA1 and can be broken by a PPT $\mathcal{B}$. Let us then construct a PPT $\mathcal{A}$ which breaks $E$.
$\mathcal{B}$'s decryption oracle for $E'$ can be simulated by $\mathcal{A}$ by using its own decryption oracle for $E$: it receives $(c_1, c_2)$ as query and decrypts $c_1$ while ignoring the redundant part $c_2$.
$\mathcal{B}$ chooses two plaintexts $m_1$ and $m_2$, which $\mathcal{A}$ forwards to its challenger. $\mathcal{A}$ receives $E(m_i)$ as challenge. This is where I am stuck.
It suggests itself to append, e.g., $E(\bar{m_1})$ to the challenge and forward it to $\mathcal{B}$, but it does not seem to work. If $i = 1$ everything is good, but as far as I can tell $\mathcal{B}$'s behavior is undefined if it receives $(E(m_2), E(\bar{m_1}))$ at this point.
What am I missing?