A uniform verdict on the K2-18 b dimethyl-sulfide claim and a population-scale biosignature census of the JWST exoplanet atmosphere archive
The claimed detection of dimethyl sulfide (DMS) in the atmosphere of K2-18 b is the highest-profile candidate biosignature in the James Webb Space Telescope era, but independent re-analyses under heterogeneous forward models and reduction pipelines have arrived at incompatible verdicts . Here we resolve the controversy with a single methodological instrument: a uniform, pre-registered, machine-learning-accelerated Bayesian retrieval of the public JWST exoplanet atmosphere archive that implements the full Green et al. 2021 Standards of Evidence protocol end-to-end. A conditional normalizing-flow neural posterior estimator trained on 10^6 physics-based forward simulations over a 31-dimensional atmospheric-parameter space (including 15 life-indicative species and a non-isothermal three-layer pressure-temperature profile) produces a calibrated posterior for any archive spectrum in seconds. Every flagged (target, species) tile is driven through 14 independent gates — Holm–Bonferroni and Benjamini–Hochberg multi-testing correction, asymmetric-tail significance, an abiotic-mechanism catalogue of 25+ published pathways, a stellar-contamination posterior (Rackham2018,LustigYaeger2023), injection–recovery, null-injection false-positive-rate measurement, line-list ablation, and five further robustness tests — before any biosignature-relevant interpretation is permitted. We apply the instrument to all four published K2-18 b reductions and report CONFIRM/REJECT the DMS claim at _10BF_bio:abio=X, and simultaneously deliver the first population-scale 95%-credible upper-limit table on 15 biosignature gases across the 15-planet sample. The posterior catalogue, trained surrogate weights, 10^6 simulation library, preregistration, and analysis code are released as a community resource under MIT licence. We confirm/refute the previously-claimed detection of dimethyl sulfide (DMS) in the atmosphere of K2-18 b, and score every atmosphere against thermochemical equilibrium to identify headline molecular-disequilibrium anomalies. The posterior catalog, trained surrogate weights, preregistration, and analysis code are publicly released as a community resource.