This arXiv preprint introduces a 260,000-molecule dataset of ground-state and conical-intersection geometries for small organics limited to ten heavy atoms (C, N, O, F). Geometries were optimized at the semi-empirical OM2 level and single-point energies plus conical intersections were computed with OM2/MRCI, a cost-effective but approximate multireference method. While the authors correctly highlight the scarcity of such data for training excited-state ML models, they understate key limitations: OM2/MRCI systematically underestimates barriers and over-stabilizes certain intersections compared with CASPT2 or MRCI+Q benchmarks, and the dataset lacks solvent effects or experimental validation. Synthesizing this with the QM9 ground-state benchmark (Nature Communications, 2018) and the 2023 photochemical dynamics review in Chemical Reviews reveals a missed opportunity—the new set could enable transfer learning from QM9 to excited states, yet the authors provide no baseline ML performance metrics. For drug discovery this matters because photodegradation pathways in pharmaceuticals often route through conical intersections; accurate ML surrogates could screen for photostable candidates far faster than direct quantum chemistry. Materials applications include screening organic photocatalysts where intersection topology dictates quantum yields. The preprint status means these claims await peer review, but the scale already positions the resource as a foundational training corpus for nonadiabatic dynamics models.