When the AI Got the Maths Wrong: The Complete Audit and Correction of an AI-Assisted Cosmological Framework The Lomas Framework Paper 8

In early 2026 an AI-assisted formalisation layer attached a fabricated closed-form exponent to an independent researcher's cosmological framework, and the resulting zero-free-parameter claim for the CMB spectral index reached government addressees before it was caught. This paper is the complete public audit that followed. It withdraws the headline claim; identifies the fabricated formula, the sign error, and a phantom reference the same layer invented; recomputes every load-bearing number with a converging-shock solver validated against Lazarus (1981); reclassifies the framework's constant C_rel as calibrated rather than derived; and executes the framework's first data confrontations — including a supernova test against 1,371 Pantheon+ objects whose prediction is excluded and published as such. A two-column scoreboard separates survival (consistency everywhere tested) from discrimination (one hint-level lean at DESI; several exclusions; sharp tests pending). Appendices L–N add further test executions against the published literature, a complete re-mine of Papers I–VII, four open derivations posed as formal problems under blind protocols, and the author's timestamped visual inputs. Nothing herein is asserted beyond what has been verified; nothing verified has been withheld.