Directional resolution of historical Brazilian aeromagnetic profiles
This is a retrospective extension. The v5 offshore comparison, the previous 15-degree rotation test on nearby legacy coordinates and the near/far error differences were known before this extension. No new acquisition is introduced.

Test the entire 360-degree cone on datum-corrected coordinates, both near and distant, separately for segment shape and for detrended segment curvature. For each hypothetical field, share only its tangent bearing (in the v5 local geographic frame) across sites. Retain each segment's own 2008 direction, 1971-to-2008 angle, intensity ratio, elevations and source fit. The measured IGRF benchmark has a slightly different bearing at each tile and height. Describe the minimum, the share of 1-degree grid bearings as good as the local IGRF, and the offsets within 5% of the best RMSE. The 5% threshold is descriptive, not a confidence interval or a significance test. Retain the fixed-field comparison, the offshore direction, and the distant results regardless of their ranking.

Tune the common bearing from historical profiles in other blocks by minimizing pooled squared error separately for each score and region. Exclude the held-out block, all its production-line IDs, and training tile centers less than 24 km from any test tile center. The 2008 source fitting is preserved. Adjacent 24 km training disks may overlap. As a geographic transfer, fit the bearing in one region and score the other using only source-region line IDs absent from the target and with at least 48 km separation of tile centers. Scoring is always point weighted. Present selection and prediction separately, and compare the tuned bearing to the historical variable local field. No independent magnetic history is inferred from these checks.

After examining the main results we added a stricter within-region sensitivity. Require more than 48 km between all calibration and test tile centers, so the 24 km source-fit disks cannot touch or share training locations. A fold without eligible calibration segments is unscorable. Summarize the historical reference on the same surviving folds.

After evaluating the 5% criterion, compare relative (1, 2.5, 5 and 10 percent) and absolute (0.05, 0.10 and 0.25 nT) tolerances to check how much the much larger distant residual baseline drives the apparent width difference.
