When is a rigid pelvis–trunk approximation sufficient? An output-specific model-reduction protocol applied to lumbar mobility in multibody biomechanics

This final manuscript revision presents an output-specific computational model-reduction protocol for lumbar mobility in multibody biomechanics. The primary study compares mass-equivalent exact-lock, axial-yaw-only and three-coordinate architectures using ModelicaHumanBodyPArts v0.10.0, with six mirrored nominal cases, 45 operating-sweep cases, 20 lumbar-impedance sensitivity cases, 18 numerical-refinement runs and an analytical two-inertia benchmark. At nominal effective transverse stabilization of 1000 N·m/rad, pelvis and thorax yaw meet the stated 10% locked-normalized criterion in all 15 primary conditions, while horizontal ground force fails in all 15. At 100 N·m/rad transverse stiffness, all eight audited outputs fail, including pelvis yaw (30.37%) and thorax yaw (17.23%). The supra-passive transverse coefficients are effective stabilization surrogates. Numerical refinement supports the primary operating-range classification; the transverse-stiffness and upper-body campaigns are supplementary sensitivity analyses after their stated verification gates. Restoring rigid head and upper-limb inertia gives pelvis yaw 13/15 and thorax yaw 15/15 at 10%; the two pelvis exceedances are 11.19% and 11.95%, and both meet 20%. RMS and driven-window post-hoc audits distinguish transient-sensitive maxima from persistent losses of support-output transferability. Figure 4 has been redrawn and collaborator review markers have been resolved. This revision changes manuscript wording, presentation and interpretation without changing the frozen Modelica source, campaign definitions, raw trajectories or reported numerical findings. The manuscript is a preprint and has not undergone peer review. The results do not establish human or clinical validation. The frozen reproduction dataset remains https://doi.org/10.5281/zenodo.23175126 and the software release remains https://doi.org/10.5281/zenodo.22941517. The preceding preprint version, https://doi.org/10.5281/zenodo.23175187, is preserved. The earlier reproduction update reran 7/7 harness checks, a fresh 45/45 supplementary campaign and one primary-runner smoke case; the full archived primary campaign was not rerun in that update. Layout correction, 6 October 2026: the public reading PDF has been reformatted to 27 pages with reviewer line numbers removed. All nine numbered figures (10 image panels), 11 tables and the manuscript text are retained. No scientific results, model source or reproduction data were changed.