From Fit to Attribution: What the Bullet Cluster and DESI DR2 Support

Strong cosmological evidence can support a model package without yet fixing what physical claim inside it is licensed. We ask a narrower question: where, in a published inferential pipeline, does the force of a component attribution enter, and where does it stop? We develop a contrast-indexed support-provenance audit that tracks minimally sufficient carrier-sets, their access status (direct or reconstructed), and the attributional ceiling of a claim ladder. In the Bullet Cluster, the lensing, X-ray, and galaxy configuration licenses a coarse claim: relative to independently constrained baryonic-source alternatives, an additional gravitationally effective contribution spatially separated from the dominant hot plasma is required. The focal pipeline does not by itself carry the stronger ontological claim that this contribution is dark matter in the entity/detection sense. In DESI Data Release 2, BAO–CMB–supernova combinations can license a rule- and contrast-relative preference for the w₀wₐ phenomenological model class while leaving the physical mechanism and microphysics open. A 2026 Lyα full-shape update provides an additional calibration: the Alcock–Paczynski output is validated at z_eff = 2.33, whereas the growth output was excluded from the final cosmological analysis after mock bias, showing that outputs within one pipeline can differ in evidential status. We distinguish contingent and unresolved attributional ceilings, state a source-backed criterion for potentially permanent ceilings, and identify bounded routes by which a ceiling can move. The result is not a new confirmation measure. It is a method for locating what a successful analysis supports, which inferential elements carry that support, and where stronger attribution requires new evidence.