With the example of the non-resonant inelastic x-ray scattering (NIXS) at the O45 edges (5d→5f) of the actinides, we develop the theory for shallow-core to valence excitations, where the multiplet spread is larger than the core-hole attraction, such as if the core and valence orbitals have the same principal quantum number. This involves very strong final state configuration interaction (CI), which manifests itself as huge reductions in the Slater-Condon integrals, needed to explain the spectral shapes within a simple renormalized atomic multiplet theory. But more importantly, this results in a cross-over from bound (excitonic) to virtual-bound excited states with increasing energy, within the same core-valance multiplet structure and in large differences between the dipole and high-order multipole transitions, as observed in NIXS. While the bound states (often higher multipole allowed) can still be modeled using local cluster-like models, the virtual-bound resonances (often dipole-allowed) cannot be interpreted within such local approaches. This is in stark contrast to the more familiar core-valence transitions between different principal quantum number shells, where all the final excited states almost invariably form bound core-hole excitons and can be modeled using local approaches. The possibility of observing giant multipole resonances for systems with high angular momentum ground states is also predicted. The theory is important to obtain ground state information from core-level x-ray spectroscopies of strongly correlated transition metal, rare-earth and actinide systems. {\textcopyright} 2011 American Physical Society.