Global Navigation Satellite System (GNSS) phase data are commonly analyzed under assumptions of smooth, continuous variability driven by atmospheric, ionospheric, and instrumental effects. In this work, we adopt a complementary geometric approach by analyzing the wrapped phase variable within sliding temporal windows and characterizing its phase-space structure using circular statistics. We introduce three window-level observables: a geometric phase-order parameter (H), an effective oscillatory activity proxy (ω), and a nucleation indicator (λₚᵢ) quantifying the fraction of phase samples near the topological boundary at φ ≈ π. Applying this framework to Antarctic GNSS data, we identify a small subset of rare events (≈0.4% of windows) exhibiting a sharp collapse of geometric order (H ≈ 0.845 versus H ≈ 0.99 background), a strong reduction in oscillatory activity (ω ≈ 6.5 versus ω ≈ 114 background), and a distinct reorganization of the phase distribution into disjoint domains dominated by boundary clustering near 0/2π with nonzero occupancy near π. These events are spatially localized, occurring exclusively at a single station (DAV1), and persist across multiple window sizes and step configurations, demonstrating robustness against methodological choices. The results rule out purely statistical fluctuations or global instrumental effects and are consistent with a nucleation-like transition in GNSS phase space. This study provides empirical evidence that GNSS phase dynamics can exhibit rare, localized geometric transitions, motivating the use of phase-space and extreme-event methodologies in the analysis of geophysical and space-environment signals.

This work reports the detection of rare, localized nucleation-like events in GNSS wrapped phase data using a windowed circular phase-space analysis. The events are characterized by an abrupt loss of geometric phase order, a collapse of effective oscillatory activity, and a topological reorganization of phase distributions. The phenomenon is spatially localized and robust across multiple analysis configurations, suggesting a genuine phase-space transition rather than statistical noise or instrumental artifacts.