Dear Rubi,
In your case — as Francesco mentioned — there is no real geometry error in your model, the issue is related to numerical precision, as a particle moving tangentially with the beampipe wall.
Since FLUKA can recover from this kind of error, it should not slow down your simulation though.
Enabling single scattering for hadrons at the boundary (via MULSOPT card) may prevent it, but it’s not a guarantee, the stepsize was already quite small.
However, there is a potential mistake regarding your BEAM card, which is completely unrelated to the Geofar error. Your beam is not along the Z axis, but the beam shape is defined in the XY plane. This leads to the problem described in slides 15-17 from
Cheers,
David