The tautological ring of ${\bar{\mathcal{M}}}_{g,n}$ is rarely Gorenstein

We prove that the tautological rings ${\mathsf{R}}^{\ast }({\bar{\mathcal{M}}}_{g,n})$ and ${\mathsf{R}\mathsf{H}}^{\ast }({\bar{\mathcal{M}}}_{g,n})$ are not Gorenstein when $g\ge 2$ and $2g+n\ge 24$, extending results of Petersen and Tommasi in genus $2$. The proof uses the intersection of tautological classes with non-tautological bielliptic cycles. We conjecture the converse: the tautological rings should be Gorenstein when $g=0,1$ or $g\ge 2$ and $2g+n<24$. The conjecture is known for $g=0,1$ by work of Keel and Petersen, and we prove several new cases of this conjecture for ${\mathsf{R}\mathsf{H}}^{\ast }({\bar{\mathcal{M}}}_{g,n})$ when $g\ge 2$.