Cross-Campaign Magnetometer Non-Equivalence in the Public MagNav Challenge Flight Dataset: A Practitioner's Note

The flight dataset of the MagNav Challenge Problem, released through the DAF-MIT AI Accelerator program, is a standard public benchmark for airborne magnetic anomaly navigation research. Its sixteen public training flights come from two independent flight campaigns (2020 and 2021/22). The dataset's per-campaign datasheets present identical magnetometer-position tables and report no erratum, which invites the assumption that a given sensor channel is statistically comparable across campaigns. It is not. The campaigns used different aircraft of the same type with independently controlled internal configurations, and the mid-cabin magnetometer (Mag 3) is severely corrupted in the second campaign, facts documented in the dataset's lead author's doctoral thesis (Gnadt 2022, Section 5.7) but not discoverable from the dataset distribution itself. This note (i) reproduces the cross-campaign non-equivalence entirely from the public training files, whereas the thesis quantification relies on held-out flight lines; (ii) shows the Mag 3 corruption is strongly intermittent within a single flight, spanning 23-444 nT in high-frequency amplitude across same-day segments; (iii) presents correlational evidence linking cabin magnetometer noise to inertial-navigation-system (INS) electrical activity (Spearman r up to 0.72, INS-specific, robust to attitude control in four of six flights); and (iv) shows that sensor-corruption rankings are metric-dependent: Mag 5 is corrupted in map-referenced total error yet quiet in high-frequency amplitude. The practical recommendation is to treat each campaign's sensor channels as distinct instruments: models, thresholds, and noise assumptions tuned on one campaign should not be transferred to the other without per-campaign validation.