Spectral Overlap and Structural Degeneracy in Atmospheric Retrieval Architectures: A Controlled Synthetic Experiment

This technical note presents a controlled synthetic experiment designed to illustrate how spectral overlap between molecular templates can generate structural degeneracy in atmospheric retrieval architectures. Starting from a synthetic transmission spectrum composed of H₂O, CH₄, and a simple cloud component, we construct alternative molecular bases with tunable spectral similarity to the true CH₄ template. By continuously varying the overlap between the true and alternative molecular bases, we map how the statistical distinguishability between retrieval architectures changes as a function of spectral similarity. The results reveal a continuous degeneracy regime: as spectral overlap increases, the statistical penalty associated with incorrect molecular bases decreases while stable but spurious molecular attributions persist. A representative configuration occurs at spectral overlap ≈ 0.98, where the alternative architecture produces visually plausible spectral fits but remains statistically distinguishable from the correct model. The experiment is intentionally simplified and is intended as a methodological illustration rather than a realistic atmospheric retrieval analysis. Its purpose is to highlight how inference outcomes in atmospheric retrieval can depend on the architectural structure of the molecular basis used in the model. All figures and results are generated from fully reproducible synthetic experiments. The Python code used to generate the synthetic experiments and figures is included in this record for reproducibility.