Retrieval degeneracies in JWST transmission spectra of temperate sub-Neptunes and biosignature claims

Madhusudhan et al. reported 3σ evidence for DMS and/or DMDS in the MIRI spectrum of K2-18 b (arXiv:2504.12267, April 2025). Luque et al. (arXiv:2505.13407; A&A 700, A284, 2025) found insufficient evidence for DMS and DMDS in a joint 0.6–12 µm analysis of NIRISS, NIRSpec and MIRI data, showed that ethane and other methyl-bearing molecules fit equally well, and estimated that about 25 more MIRI transits would be needed for a 3σ test; Stevenson et al. (arXiv:2508.05961, accepted to AJ) find that the MIRI spectrum is dominated by binning-dependent red noise, that 87.5% of retrievals with their preferred binning do not favour DMS/DMDS, and that K2-18 b does not meet astrobiology standards of evidence; and Schmidt et al. (arXiv:2501.18477, accepted to AJ), with 60 data treatments and over 250 retrievals, confirm CH₄ at about 4σ but find no statistically reliable evidence for CO₂ or DMS. Hu et al. (arXiv:2507.12622) added four NIRSpec transits and find DMS below 3σ (below about 2σ without haze), and 2026 modelling papers reproduce the spectrum without DMS, for example with a hazy 200–400× solar atmosphere (Lavvas et al., arXiv:2603.26474). The WASP-39 b early-release programme established CO₂ detection and the capabilities of the JWST instruments (arXiv:2208.11692). The community conclusion is that biosignature claims on current JWST data are not robust; no accepted detection of a biosignature exists. No new JWST observations of K2-18 b were published in 2026 as of October.

Robust retrieval protocols determine which JWST and Ariel/ELT observations can support claims about habitability, and the model-comparison methods transfer to all atmospheric characterisation. No other application.

Resolved when a retrieval protocol validated on blind injection–retrieval tests reproduces the true composition of simulated temperate-planet spectra to within stated confidence intervals and, applied to K2-18 b data (and a second planet), yields identical detection conclusions across at least three independent codes.