Regional specificity of the two-survey Uruana signature
This directory extends the Uruana supplementary package. Run the following from the root of Supplementary Material S1 with Python, NumPy, SciPy, pandas and the bundled ppigrf copy. The raw-flight bin medians, grid values, coefficients, independent implementation and prior tables are already in S1. Computations use geographic cell centers at the original dates and survey heights specified in the study. No network access is needed.

  export OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1
  python regional_specificity/scan.py --step 2 --workers 4
  python regional_specificity/scan.py --step 1 --workers 4
  python regional_specificity/relax.py --workers 4
  python regional_specificity/refine.py --workers 4
  python regional_specificity/check_results.py

Inputs and design. The 23 eligible origins, northing windows, widths, channels and slope-sign rules are those in the previous nested_gradients and background_transplant analyses. Each control now supplies its OWN residuals, tangent-plane geometry and aircraft elevations. The regional main field is removed at the original coordinates and epoch, and a main field for the same epoch is added at each transported position. Both years must pass at the same destination center. 2011 uses paired slopes on 14 flight lines. 2008 uses four bands in both channels at five corridor widths. This analysis uses the raw two-survey sign rule and does not additionally require dipole-negative/full-positive model slopes. The first/last bins and native grid nodes are processed as specified in S1.

Outputs. sectors_1deg.csv and sectors_2deg.csv contain all eligible sectors, native gradient-shift intervals, separate yearly and joint grid counts, maximum signed margins (nT/km), and the locations of the best joint margins. The raw count uses MAGBRU in 2011. The compensated sensitivity substitutes MAGCOM on the same 2011 bins. Both 2008 channels are required in both analyses. Inside/outside percentages weight latitude strips by spherical area within their respective class, classified at both epochs using surface intensity. Counts of equal-angle grid centers are not proportions of Earth's surface area. transition denotes cells crossing the threshold. A positive joint margin means the least favorable of the 68 signed slopes exceeds zero.
relax_1deg.csv drops the most adverse entire 2011 line or the most adverse entire 2008 width, or both, at each sector/grid center, retaining every other line/width and both 2008 channels. This is an intentionally permissive sensitivity on the raw channel.
refined.csv records a local search around each best 1-degree grid center and agreement with the independently implemented ppigrf synthesis. The independent calculation evaluates both the native and the destination fields. The search is local, with no mathematical guarantee that it finds the continuous global maximum.
PROTOCOL.txt distinguishes the expanded primary comparison from sensitivities chosen after the first results. Reuse of the Brazilian surveys and known Uruana signature makes this a retrospective specificity analysis. Survey data are observed in Brazil only. A positive result elsewhere is a model calculation for a transported Brazilian residual, not the discovery of another anomaly. Geographic cells and eligible sectors are spatially dependent.
