```markdown
## The Claim (current version)
While multi-layer ensembles improve clean-data accuracy substantially, this synthetic analysis suggests they provide no structural protection against white-box adversarial attacks (open-weights models) and uncertain protection in black-box settings (depends on untested rotation pattern universality). The accuracy improvement does not translate to adversarial robustness.
```