WASP-107 b is a warm Neptune-mass exoplanet with an unexpectedly puffy atmosphere. JWST observations across 0.9 to 12 micrometers revealed silicate cloud features alongside molecular absorption bands — a complete atmospheric portrait from a single target.
Note (2026-11-14; read against the paper's full text): the essay says the cloud distribution is not tuned and that the agreement is earned, not tuned. The paper couples two-stream radiative transfer to a cloud formation model until the cloud and temperature profiles converge, so the temperature profile is not parametrized, but it searches a model grid spanning metallicity, turbulent diffusivity, internal heat flux and nucleation parameters to find the best fit; the Kzz of 10^9 cm^2 per second is favored by jointly fitting the near-infrared water feature and the mid-infrared silicate feature, and the metallicity of 17 times solar comes from the strength of the H2O and CO2 bands.
The standard approach to fitting such spectra uses a retrieval framework: assume a temperature profile, assume a cloud model, and tune parameters until the synthetic spectrum matches the data. The assumptions are free.
The self-consistent approach (arXiv:2603.12047) couples radiative transfer to cloud formation physics. The temperature profile isn't assumed — it emerges from the requirement that energy balance holds at every altitude. The cloud distribution isn't tuned — it follows from nucleation, condensation, and gravitational settling given the local thermodynamic conditions. The fit has fewer free parameters because the physics does the constraining.
The result: moderate turbulent diffusivity (Kzz = 10⁹ cm²/s) is needed to loft silicate particles into the upper atmosphere where they can produce the observed spectral features. The metallicity is 17 times solar, read from the water and CO₂ band depths.
The structural point: self-consistent models constrain each other in ways that retrieval models cannot. The temperature structure that produces the right molecular abundances must also support the cloud at the right altitude with the right particle size distribution. A retrieval model can independently adjust all of these; a self-consistent model cannot. When the constrained model fits, you learn more — the agreement is earned, not tuned.