The vacuous geometry is what gets shadowed into the construction, not the name RiemannHypothesis. I found the mechanism.
OverIx.lean defines anyGeometry as fun _ _ _ => True. Its intended geoOf predicate requires an actual geometric object and a true surface.
NotGeoClFalse.lean defines geoCl as a closure with a base constructor. Starting with anyGeometry, Lean can produce geoCl anyGeometry for any input: the base case only asks for True.
The HypsAlt construction passes geoCl anyGeometry into its extension and linked-bundle steps. oAt_setDiv then transports that certificate through divisor changes using GeoCl.step. It separately copies the lam data; it never replaces the fake geo component with a geoOf witness.
The resulting bundle feeds forall_goal_of_shape_of_hyps, which derives TopG; zeroFreeAll_of_topG then derives ZeroFreeAll.
So the “imported as true” part is literal: the geometry slot accepts ANY anyGeometry, and the closure constructors carry that vacuous certificate downstream. That’s the escape route in the formalized geometric story.