URUANA DE MINAS — PROJECT 3087 CHANNEL / HEIGHT / BIN-PHASE SENSITIVITY AUDIT Date: 2026-08-18 Status: POST-DISCOVERY ROBUSTNESS / ADVERSARIAL SENSITIVITY. NOT A PREREGISTERED CONFIRMATORY TEST. PURPOSE ------- Test two reviewer-sensitive alternatives without changing the frozen primary claim: (1) whether the target-negative -> north-positive turnover is confined to the primary MAGBRU channel; and (2) whether a simple accompanying flight-height/topography trend duplicates the universal magnetic sign pattern. A third sensitivity shifts the origin of the 250-m bins to expose bin-boundary dependence rather than hiding it. UPSTREAM SOURCE AND FROZEN GEOMETRY ----------------------------------- Source: SGB Project 3087 original Mag-3087.XYZ. Frozen city center: X=365408.554 m, Y=8224103.341 m (WGS84 / UTM 23S). Frozen production lines: 12340, 12350, ..., 12470 (14 lines). Target window: dy=+6 to +8 km, r<=8 km, clipping before binning. Northern same-line control: dy=+16 to +20 km. The primary manuscript operator remains the declared 250-m binned OLS MAGBRU slope. This audit does not redefine it. IMPLEMENTATION NOTE ON THE PHASE-0 RECONSTRUCTION ------------------------------------------------- The upstream sensitivity script is an independent reconstruction rather than a byte-identical replay of the already packaged primary table. Its phase-0 MAGBRU target median is -1.539506 nT/km, whereas the paired same-line primary implementation used in the manuscript gives -1.554762 nT/km. The small difference arises from direct re-binning on fixed UTM phase boundaries and associated edge/bin-boundary implementation details; both implementations retain 14/14 negative target signs at their respective primary/phase-0 settings. The sensitivity reconstruction is therefore used to test sign stability across channels and bin origins, not to replace the exact primary summary. EXACT EXISTING SAME-LINE MAGMIC CHECK ------------------------------------- Using the already packaged primary line/control table and its exact manuscript binning: - MAGMIC target slopes: 14/14 negative; median -1.660952 nT/km. - MAGMIC northern-control slopes: 14/14 positive; median +1.164173 nT/km. - MAGMIC paired north-minus-target changes: 14/14 positive; median +2.542801 nT/km; minimum +0.675563 nT/km. Thus the turnover is not confined to MAGBRU. DIRECT RAW-XYZ CHANNEL CHECK ---------------------------- A separate script streams the original upstream XYZ and reads the supplied columns MAGBRU, MAGCOM, MAGCOR, MAGNIV, MAGMIC, IGRF, GPSALT, BARO, ALTURA, and MDT. No gridded derivative or IGRF subtraction is used to create the sign turnover. Raw-point OLS sensitivity (no along-track independence claim; used only as a channel check): - MAGBRU: 14/14 target-negative, 14/14 north-positive. - MAGCOM: 14/14 target-negative, 14/14 north-positive. - MAGCOR: 14/14 target-negative, 14/14 north-positive. - MAGNIV: 14/14 target-negative, 14/14 north-positive. - MAGMIC: 14/14 target-negative, 14/14 north-positive. - IGRF: 14/14 positive in target and 14/14 positive in northern control. 250-M BIN-PHASE SENSITIVITY --------------------------- The same clipped windows were aggregated using four 250-m UTM bin phases: 0, 62.5, 125, and 187.5 m. OLS outcomes: - MAGCOM: 14/14 target-negative and 14/14 north-positive at all four phases. - MAGNIV: 14/14 target-negative and 14/14 north-positive at all four phases. - MAGMIC: 14/14 target-negative and 14/14 north-positive at all four phases. - MAGBRU: 14/14 target-negative at three phases and 13/14 at the 125-m shifted phase; northern control is 14/14 positive at all four phases. - MAGCOR: 14/14 target-negative at three phases and 13/14 at the 125-m shifted phase; northern control is 14/14 positive at all four phases. At the exceptional 125-m phase, line 12360 is nearly flat in MAGBRU (+0.0290 nT/km) and MAGCOR (+0.0004 nT/km), while MAGCOM, MAGNIV, and MAGMIC remain negative. A rank-robust pairwise-median (Theil-Sen-type) sensitivity likewise shows that the exact 14/14 MAGBRU count is not invariant to every estimator/bin-origin combination: shifted phases can yield 13/14. This is retained as a sensitivity boundary, not suppressed. FLIGHT-HEIGHT / TOPOGRAPHY SIGN CHECK ------------------------------------- Across every one of the four bin phases, target positive-slope counts are: - GPSALT: 13/14 positive (1/14 negative). - BARO: 13/14 positive (1/14 negative). - ALTURA: 4/14 positive, 10/14 negative. - MDT: 10/14 positive, 4/14 negative. None reproduces a universal 14/14 negative target sign accompanied by a universal 14/14 positive northern magnetic control. The height/topography channels therefore do not provide a simple all-line sign mimic for the magnetic turnover. IMPORTANT LIMIT --------------- This sign comparison is not presented as a complete altitude correction. Some target lines contain only 5-8 binned support points, so multivariable line-wise regressions of field against northing plus height are ill-conditioned on the weakest lines and are not promoted as an "altitude-adjusted 14/14" result. The scientifically defensible conclusion is narrower: the supplied height/topography channels do not reproduce the universal magnetic sign architecture, while multiple supplied magnetic-stage channels independently retain it. INTERPRETATION -------------- The frozen primary MAGBRU result remains exactly as reported. The new audit strengthens artifact resistance by showing multi-channel persistence and by making bin-origin/robust-estimator sensitivity explicit. It also prevents overclaiming: 14/14 is a property of the frozen primary operator, not a universal statement across every possible slope estimator and bin phase. REPRODUCIBILITY --------------- scripts/verify_3087_channel_altitude_sensitivity.py can stream either Mag-3087.XYZ itself or the original ZIP archive containing it and regenerate the included derived sensitivity tables. The large third-party upstream archive is not redistributed.