One Conductor and One Score to Rule Them All: A Backend Agnostic Optimizer Employing Nonredundant Symmetrized...

Reliable composite geometries require stationary points of one potential energy surface assembled from derivatives at the same nuclear configuration. We introduce an open-source, multiplatform framework combining an optimization driver, local, symmetrized, nonredundant internal or Cartesian coordinates, and interchangeable electronic-structure backends. The driver exchanges Cartesian geometries with the backends and assembles their energies and derivatives according to machine-readable frozen protocols specifying electron correlation, basis-set completion, and core-valence effects. A standard reference test (referred to as Baker-30) compares the internal-coordinate optimizer with the Gaussian optimizer. Larger systems illustrate symmetrized Cartesian coordinates; five composite methods are then assessed on a common 20-molecule panel. The resulting composite-gradient optimization is auditable and extensible while allowing each backend to contribute its strongest capabilities. The tested models give bond-length errors on the milliangstrom scale; reported core-valence shifts range from 0.8 to 4.2 mÅ.