Retrospective fourth-paper extension. Fix the nearby angle interval at the min and max of the 23 prior Brazilian tile centers at the surface, 1971.5 and 2008.5. Define exterior by surface field >=28 microtesla at both dates. Lead map uses WGS84 geographic coordinates and geodetic zero height at 0.125 degree cell centers. Include all areas. Verify with 0.25 and 0.5 degree grids, alternate cutoffs, angular interval padding and one-year changes. Numerical support denotes cell centers, not independent observations, nor a global probability of the crustal anomaly. Land mask is Natural Earth at 1:110m (major land masses; omissions of islets possible).
Post-map extension: additionally retain nodes whose historical/2008 intensity ratio lies between the nearby tile extremes, using the same 23 Brazilian tiles and the archived segment fields. The 15 resulting offshore nodes are geographic alternatives. The map and their declination sign were inspected before testing their directions on the 1971 profile reconstruction. Choose their circular-median-nearest bearing as a representative, without using any magnetic profile error. Test all 15. At each Brazilian segment's 2008 direction use the tangent basis toward increasing declination and downward inclination. Reproduce the magnitude and historical/2008 intensity ratio of the true local update exactly. Substitute only the offshore tangent bearing of the backward (2008 to 1971) rotation, expressed relative to that local basis, for its Brazilian bearing. Keep fits, geometry, observation bins, damping and weights identical to v4. Include projection and induction in the model. The nine nearby 24-km blocks are the primary comparison and the geographically held-away Brazilian profiles are a comparison of portability. Score centered shape and also the nonlinear remainder after removing each segment's eastward slope. Reproduce the archived fixed and historical point predictions before interpreting any change. Tabulate leave-one-block and block-level outcomes for the representative. All transferred predictions are counterfactuals on Brazilian data.
