BIW-Bench-Public: a de-identified body-in-white structural-CAE benchmark

A de-identified, reproducible subset of an industrial body-in-white (BIW) structural-CAE benchmark, released to support a peer-reviewed publication (reference added upon acceptance). Contents: five real automotive sheet-metal assemblies as neutral-named geometry — left/right door and side-frame full/sub assemblies (native CATIA V5, preserving spot-weld / seam-weld / adhesive connector semantics in geometrical-group names) and a bumper-beam + crash-box set (ISO-10303 STEP); seven structural-CAE tasks across three physics regimes (free-free modal, nonlinear dent resistance, explicit axial crush) — six with corroborated reference baselines and one marked preliminary — with task specs (config.yaml), reference acceptance targets (golden.yaml, corroborated on up to four commercial solvers), and a self-contained evaluator (evaluator/evaluate.py) that scores a result file against the acceptance rules and enforces the crush task's solution-quality physics gate. A SHA-256 manifest of every file is included. De-identification: identifiers tying the geometry to a vehicle, program, customer, or supplier — part numbers (filenames, CATIA PartNumber/instance names, assembly link tables), embedded PLM/BOM identifying metadata and descriptive values (organization id, part names, lifecycle states, design dates, workflow ids), cross-reference and standard-part numbers, and internal paths — were removed with a length-preserving pipeline and audited by multiple independent multi-encoding scans (including even/odd byte-offset UTF-16 decoding). Geometry, assembly structure, and connector semantics are preserved bit-for-bit. See data_card.md inside the archive for the methodology, what remains by design, re-identification risks and limitations, contamination analysis, and licensing (data CC BY-NC 4.0, evaluator code MIT).