From Excluded to Consistent: A Vacuum-Tied MOND Law With Per-Host, Condensed-Sourced Settling Against Raw KiDS-1000 Cosmic Shear (Working Note)

Working note, version 1.0 (2026-10-14). AI-assisted research programme; not peer reviewed. We test a gravity framework in which the acceleration scale is tied to dark energy, a0 = kappa c sqrt(G rho_DE) (kappa = 1/2, fitted), and a cold collisionless component (cold energy, Omega_c/Omega_b = 5.364) settles inside bound systems into the profile the law demands. The test is framework-native: the framework's own 1024^3 particle-mesh simulations (L = 100 Mpc/h) are scored, through the gravitating-field matter power spectrum, on the raw KiDS-1000 and DES Y3 shear correlation functions, against a same-pipeline control without settling; no LambdaCDM-calibrated nonlinear machinery enters the verdict. The original settling rule (all baryons source the law; supply pooled across overlapping catchments) puts 19-25% too much power at k = 0.3-1 h/Mpc and is excluded (KiDS delta chi^2 = +68.7). A condensed-baryon census cap gives +14.0 (narrowly excluded). Two physically motivated rules - each halo settles only its own catchment's supply, and only condensed baryons (stars + cold gas) source the settled profile - bring the framework's boost to 1.005/1.012/1.023 at k = 0.3/0.5/1 and make KiDS-1000 consistent: delta chi^2 = +2.03 (robustness maximum +3.28; sensitivity tooth passes); absolute chi^2 146.3 (p = 0.011) vs the control's 144.3 (p = 0.014). DES Y3 is not diagnostic at this box size. Limitations: the passing combination was selected after small-box previews (the pre-declared primary run, adding gas expulsion, is in progress); the classifier is relative; initial conditions are inherited (a native-input rerun is prepared, preview unchanged); 54%/77% of KiDS/DES points are lost to the box; the alternative footing is pending; the condensed-only source rule is not yet derived from an action; massive systems still carry unexplained core mass. kappa is fitted; nothing here closes the theory. Companion note: PAPER47, DOI 10.5281/zenodo.23354043. AI assistance: Anthropic's Claude.