Joint falsification of gravity, magnetotelluric and surface-wave data resolves a fertile-corridor discriminator: the Al Amar suture, Arabian Shield

Jef Caers

A prospective mineral province is defined by deep structure that is known only from sparse
geophysical data, and any one dataset permits several geological interpretations. This paper
describes a prototype that converts a geologist's written tectonic history into a falsifiable
prior on deep petrophysical structure and asks which measurements separate geologically distinct
histories. A language model restricted to constraint extraction compiles the history into typed,
provenance-tagged priors; a coarse lithospheric generator produces density, shear-velocity, and
resistivity fields; benchmarked forward models predict gravity, magnetic, surface-wave, and
magnetotelluric data; and candidate scenarios are tested by calibrated robust-Mahalanobis
falsification rather than inversion. Applied to the Al Amar-Halaban suture and Najd fault system
(Ar Rayn terrane, Arabian Shield), no single dataset separates a fertile deep corridor from
engineered barren rivals; only the four datasets read together reject every rival, and the
discriminator survives out-of-model truth. All four datasets are checked against real
observations: the falsification loop closes on EMAG2v3 magnetics and on a published shear-velocity
model, WGM2012 gravity admits the model out-of-sample, and 122 measured magnetotelluric soundings
validate the background compilation chain. A clause-level sensitivity analysis reports which
sentences of the history carry the discrimination and which the data informed, returning an
updated narrative and a survey design to the geologist.