A Width-Chain Estimate of the MOND Acceleration Scale from MIGHTEE-HI COSMOS, with its Flux-Scale, Line-Width and Kernel Systematics

Version 1.0: the first deposited version (internal revision 1.2; not peer reviewed). Revision history: an undeposited first draft read the catalogue line widths as observed-frame and divided them by 1+z; the catalogue paper's own worked examples show they are rest-frame, which a frozen follow-up test confirmed, so the rest-frame reading is primary; a second review added the frozen single-dish reading to the flux-scale range, both distance conventions and two statistical-error estimates to every footing statement, and the selection term to the abstract. The observed-frame numbers are kept as a disclosed variant. The programme ties MOND's acceleration scale to the vacuum, a0 = kappa c sqrt(G rho_Lambda), with kappa = 1/2 fitted, not derived; its footings are 9.3603e-11 (rho_Lambda) and 1.1312e-10 m s^-2 (rho_crit). We apply a width chain, frozen before any width was read, to the MeerKAT MIGHTEE-HI COSMOS catalogue: V comes from the rest-frame W50 and inclination, the baryons are 1.33 M_HI + M* at half the HI diameter, and a median-residual estimator with the exponential kernel returns a0. Forty-seven golden discs at 0.027 <= z <= 0.093 survive, all at g_bar/a0 <= 0.084, 37 gas-dominated. Result. On the catalogue's flux scale and H0 = 70 distances a0 = 1.31e-10 m s^-2 (statistical +-0.034 or +-0.048 dex, by bootstrap or per-galaxy spread; recipe +-0.128 dex), kappa = 0.70 or 0.58: consistent with SPARC and the alternative footing, with the canonical footing 0.019 dex outside the recipe width. On the H0 = 67.4 distances of rho_Lambda (a0 = 1.21e-10) the canonical footing is inside it (0.111 dex away). Arecibo fluxes exceed the catalogue's for brighter, nearer galaxies; applied to our discs' gas they give 0.90-1.13e-10 (0.90 for the frozen pairs, 1.07 for a post hoc extrapolation), inside the recipe width of both footings. Line-width corrections lower a0 by up to 0.10 dex; kernels without the exponential's next-order term raise it by 0.06-0.07 dex; dropping the S/N cut, by 0.08 dex. A SPARC closure tests only the baryon recipe. The width frame rests on the catalogue paper's worked examples; independent data tests could not decide it, and all three lean weakly towards the observed frame. What it can and cannot say. The result is not a footing discriminator and, with a lever of 0.009 dex, not an a0(z) test. The cold mass (Omega_c h^2 ~ 0.12) is still required; no dark-matter particle is added; no preference of the data for one reading over another is claimed. Every number carried by an audit row is re-read from committed files by PAPER40_audit.py (417 of 417; 19 literature values are checked only for being printed; 260 plain numbers are also found in the PDF text); the post hoc rows come from make_paper40_figures.py, which first reproduces the committed values of the frozen analyses exactly (33 checks). AI-assisted research programme (models from Anthropic, OpenAI, DeepSeek, Zhipu AI, Alibaba and Google); not peer reviewed.