The cosmic microwave background temperature measured at z = 0.886 to 2 % precision, together with molecular absorption data towards PKS 1830−211 and Sunyaev-Zel'dovich cluster data up to z ≈ 3, provide a test of a question that torsion cosmologies leave open: if spacetime exchanges energy with several matter components, which one receives it? In Einstein-Cartan theory, the Noether identities of each sector fix the answer: a sector's energy-momentum fails to be conserved only in proportion to its own spin, so minimally coupled light, which has none, is conserved whatever the torsion does.

A metric-affine cosmology fitted to DESI DR2 data, by contrast, distributes the exchange "universally" among all fluids, radiation included, and photons would gain almost a third of their energy. If the photons remain a blackbody, as FIRAS requires of any source set by local conditions, the model predicts a background colder than T₀(1 + z) by 7 % at z ≈ 0,9 and by 8,5 % at z ≳ 3, with no free parameter.

The comparison with observations is decisive. PKS 1830−211 yields 5,08 ± 0,10 K against the predicted 4,78 K; together with the CO absorbers, the data favour the standard scaling by Δχ² = 13, and the 18 Planck cluster bins give χ² = 50,8 versus 15,5. Under the same reading, the primary anisotropies would change their spectral shape by amounts Planck excludes.

If dark matter carries the spin instead of radiation, the picture changes: H(z) moves by less than 10⁻⁴ below z = 3 at the fitted parameters, and T(z) is standard again. The result points to torsion, if it exists, talking only to spin and leaving light untouched.