Erratum: The Moonshine Grade Does Not Measure Moonshine — Null Controls, Surrogate-Constant Falsification and a Random-Coil Control for DOIs 19652114, 19687681, 20120257, 20200607

Combined erratum reporting three independent findings on the moonshine grader used in four records. (1) The grader assigns Grade 3 to pure Gaussian noise, and the patch shipped in DOI 10.5281/zenodo.20342613 increases rather than decreases that rate (26.2% to 41.2% to 71.0% across original, shipped patch and current rule; direction reproduced under a second noise normalisation). (2) The grader does not depend on the Monster group at all: 0 of 120,000 published grades change when every Monster-derived constant is replaced with a fabricated surrogate, because the J-invariant coefficients and representation dimensions cancel algebraically under the final normalisation. Fabricated constants outperform the real ones in a best-fit contest (p = 0.974). (3) On protein data, a structureless random coil outscores folded ubiquitin in 5 of 5 independent realisations (15.9-30.0% Grade 3 vs 8.60%). Finding (2) explains finding (1), and (3) shows the consequence on real data; together they imply no recalibration can recover the intended measurement. All measured quantities in the affected records are unchanged; the grade is a terminal readout never fed back into the pipeline. Recommendation is withdrawal of the grade as a reported quantity rather than reissue of corrected tables. This erratum makes no claim that the published 71.36% of DOI 10.5281/zenodo.20120257 is wrong; that record's encoder is unpublished, so no like-for-like regrade is possible.