The VIP bound on the generalised Károlyházy collapse model does not follow from its stated analysis

Bortolotti et al. (New J. Phys. 28, 064511 (2026)) report a lower bound R_K > 4.64 m (95%) on the correlation length of the generalised Károlyházy model from germanium detector data, and conclude that the model is excluded because a theoretical argument requires R_K ≤ 1.98 m. Evaluating the paper's own expected signal formula, efficiency table and data in consistent SI units, with a setup normalisation reconstructed from the same group's earlier analyses of the same detector and data, gives R_K > 0.008 to 0.009 m. Even assuming zero background and zero observed events, the data cannot give more than about 0.024 m. The quoted bound requires about 2 to 3 × 10^5 times more expected signal counts than the stated formula yields. We find no physically justified reading of the analysis that reaches 4.64 m; evaluating the energy factor in keV while keeping the rate constant in SI units gives 3.5 to 4.0 m, which we note only as a possible origin. Without this bound, a single-frequency recast of the XENONnT limit by the method of the original theory paper with the theory paper's rate gives R_K ≳ 1.3 to 1.45 m (95%) and does not exclude the model, leaving a narrow window below 1.98 m; a full spectral fit could change this. We also note that the prior as stated in the paper does not reproduce the published posterior, and that the theoretical upper bound depends on a localisation threshold chosen as 10. Supplementary archive: scripts that reproduce every number in the note (setup normalisation from the published CSL and Diósi Penrose results, expected signal, likelihood, zero-background ceiling, alternative readings, CSL recast, theoretical upper bound, XENONnT recast), their outputs, and the bin contents read from Fig. 3 of the commented paper. Run scripts/run_all.sh (Python 3, numpy, scipy).