We analyze nonequilibrium fluctuations of the averaging process on $\mathbb T_\varepsilon^d$, a continuous degenerate Gibbs sampler running over the edges of the discrete $d$-dimensional torus. We show that, if we start from a smooth deterministic non-flat interface, recenter, blow-up by a non-standard CLT-scaling factor $θ_\varepsilon=\varepsilon^{-(d/2+1)}$, and rescale diffusively, Gaussian fluctuations emerge in the limit $\varepsilon\to 0$. These fluctuations are purely dynamical, zero at times $t=0$ and $t=\infty$, and non-trivial for $t\in (0,\infty)$. We fully determine the correlation matrix of the limiting noise, non-diagonal as soon as $d\ge 2$. The main technical challenge in this stochastic homogenization procedure lies in a LLN for a weighted space-time average of squared discrete gradients. We accomplish this through a Poincaré inequality with respect to the underlying randomness of the edge updates, a tool from Malliavin calculus in Poisson space. This inequality, combined with sharp gradients' second moment estimates, yields quantitative variance bounds without prior knowledge of the limiting mean. Our method avoids higher (e.g., fourth) moment bounds, which seem inaccessible with the present techniques.