Attractor-invariant formula under quiver mutation

Let $(Q,W)$ and $(Q',W')$ be quivers with potential related by mutation at a vertex $k$, and let $d,d'$ and stability weights $\theta,\theta'$ be related by the mutation formulas in the source. Let $\Omega_\theta(d,y)$ and $\Omega'_{\theta'}(d',y)$ be the corresponding DT invariants. Suppose that $d$ is not collinear to $e_k$, equivalently that $d'$ is not collinear to $e'_k$. Attractor-invariant mutation conjecture. If $d\in\mathbb{N}^{Q_0}$ and $d'\in\mathbb{N}^{Q'_0}$, then $\Omega_\theta(d,y)=\Omega'_{\theta'}(d',y)$; if $d\in\mathbb{N}^{Q_0}$ but $d'\notin\mathbb{N}^{Q'_0}$, then $\Omega_\theta(d,y)=0$; and if $d'\in\mathbb{N}^{Q'_0}$ but $d\notin\mathbb{N}^{Q_0}$, then $\Omega'_{\theta'}(d',y)=0$. In particular, the attractor invariants coincide or vanish under the same conditions. This expresses the expected mutation, or Seiberg-duality, invariance of DT invariants away from the mutated basis direction; the source gives no resolution status.