Is there a clear-cut analogue of the strong form of Serre's Conjecture for residually reducible Galois Representations? Let $p$ be a prime and $\mathbb{F}$ a finite field of characteristic $p$. The theorem of Khare and Wintenberger roughly states that an irreducible, odd Galois representation $\bar{\rho}:G_{\mathbb{Q}}\rightarrow GL_2(\mathbb{F})$ which is ramified at finitely many primes lifts to a modular Galois representation associated to an eigenform of optimal weight and level. The optimal level of this form is the prime to $p$ part of the Artin conductor of $\bar{\rho}$. This was the strong form of Serre's conjecture.
Hamblen and Ramakrishna showed that the hypothesis of irreducibility may be relaxed by showing that under some conditions, a reducible and indecomposable $\bar{\rho}$ lifts to a Galois representation associated to an eigenform. However, they are not able to optimize the level of the eigenform.
My question is the following: is it expected that the optimal level is the prime to $p$ part of the Artin conductor of $\bar{\rho}$? It may be the case that it is not exactly this, and if it is not, is there an explicit counterexample to this?
 A: Billerey and Menares have studied this question  for the reducible representations $\bar{\rho} = 1 \oplus \chi_p^{k-1} $ in https://arxiv.org/abs/1309.3717
In this case the prime-to-$p$ part of the Artin conductor is 1, and it is not always the case that $\bar{\rho} $ arises from a cuspidal eigenform of weight $k $ and level $1$.
A: In addition to François' answer, here is what can be said if you only seek for an isomorphism between the semi-simplification of your residual representation $\overline{\rho}$ and the semi-simplification of the reduction of a $p$-adic representation attached a Hecke eigenform.
Consider an odd mod $p$ Galois representation $\overline{\rho}=\nu_1\oplus\nu_2$ of Serre weight $k$ and level $N$ (coprime to $p$) and assume $p>k+1$. Then, there exist $\epsilon_1,\epsilon_2$ two Galois characters unramified at $p$ such that $\overline{\rho}\simeq\epsilon_1\oplus\epsilon_2\chi_p^{k-1}$. Set $\eta=\epsilon_1^{-1}\epsilon_2$. Then, there is a newform $f$ of (optimal) weight $k$ and level $N$ and a prime ideal $\mathfrak{p}$ over $p$ in $\overline{\mathbf{Q}}$ such that we have $\overline{\rho}\simeq\overline{\rho}_{f,\mathfrak{p}}^{ss}$ if and only if we have $B_{k,\eta}=0$ or $\eta(\ell)\ell^k=1$ for some prime $\ell$ dividing $N$.
Here, $\eta(\ell)=\eta(\mathrm{Frob}_\ell)$ if $\eta$ is unramified at $\ell$ and $\eta(\ell)=0$ otherwise. (Roughly speaking $B_{k,\eta}$ is the mod $p$ reduction of the $k$-th Bernoulli number associated with a lift of $\eta$.)
