THE QUANTUM INTELLIGENCER

A study newly posted to the arXiv service reports that a compiler, permitted to choose among equal-cost orderings of commuting phase terms, may reduce routed controlled-NOT gates by 9.14 per cent against an unoptimized ordering and by 11.10 per cent against a prior stochastic search. The same selection, applied for direct-depth advantage, reduced circuit depth by 12.83 per cent across twenty aggregates, while a matched control altered depth by only -0.41 per cent; that control result the paper leaves unresolved. The paper does not promise hardware gain. On a prospective IBM Heron panel, raw generator error shifted by -0.0025, or -0.59 per cent; fixed-panel shot uncertainty excluded zero, yet term-seed inference remained unresolved. The depth benefit transferred to a second generator and to forty-eight terms, but reversed under BasicSwap. The authors conclude that equal-primary-cost tours are a useful router-conditioned compiler freedom, not a guaranteed hardware benefit. The scaffolding of these figures is worth stating plainly. The experiments were run on synthetic sixteen-qubit assignment-Ising instances, and the existence of equal optima was established by counting rather than assumed: exact enumeration through twenty terms showed the multiplicity to be instance-dependent, and where the support lower bound was attained, the reversal-reduced width equaled the number of undirected Hamiltonian paths of the support line graph. Selection was then tested at thirty-six terms, on twenty-four sampled minimum-support-cost orders, set against a matched control of twenty-four restarts. Candidate rankings persisted across SABRE routing seeds, which the authors offer as the reason selection survived routing re-randomization. The reported percentages rest on a scaffold that is modest, but counted. —Elias Hartwell Dispatch from The Prepared E0