Abstract
Atmospheric retrievals of M dwarfs, brown dwarfs, and exoplanets depend on molecular opacities whose quality varies widely across species, and that variation is not yet routinely quantified. This work introduces QORE (Quantum-validated Opacity Retrieval Engine), an opacity-maturity layer that scores each opacity product on five components (provenance, temperature coverage, pressure broadening, completeness, experimental refinement), combines them by a geometric mean, and folds the scored opacities into one forward model. The two least-mature members of a six-species demonstration set are treated with quantum chemistry: for the transition-metal oxides FeO and CrO, absent from standard line-list databases, double-shell CASSCF/NEVPT2 supplies provisional ground-state vibrational anchors (CrO harmonic frequency to 1.8%); for dimethyl sulphide and disulphide (DMS/DMDS) a VPT2 temperature–pressure grid replaces the single-condition cross section in current use. The upgrades raise the maturity of this six-species demonstration set from 0.25 to 0.58, a conclusion insensitive to the component weighting and to one-tier score reassignments. These values grade the selected opacity products only and are not a general quality score for QORE or for any atmospheric retrieval. All production opacities are computed classically; the electronic-structure workflow behind the new products is separately exercised end to end on quantum-computing hardware as a supporting certification, and no quantum advantage is claimed. The scored opacities fit the M8.5–Y4 emission sequence, matching the regime-appropriate Sonora grids in raw fit quality at the coldest objects while trailing them, on raw fit and on information criteria, at the warmer ones, and fit the JWST transmission spectra of K2-18 b and WASP-39 b; in a controlled K2-18 b test within QORE’s simplified transmission branch, the temperature- and pressure-resolved opacity lowers the DMS Bayes factor from lnBF =4.08 to near zero. The score is a heuristic input-quality diagnostic, not a detection statistic or an uncertainty estimate, and the retrieval results are model-conditional.