Show HN: Starlink constellation health – 108 reentry anomalies in TLE data

Abstract Current approaches to satellite reentry prediction propagate orbital elements forward through atmospheric density models to estimate reentry windows. We propose a complementary approach: geometric trajectory similarity monitoring applied to the sequence of TLE updates for a decaying object. The State Topology and Trajectory Storage (STTS) framework[^stts] asks whether the current TLE trajectory resembles trajectories that preceded confirmed reentries — without invoking atmospheric models, propagation, or force model assumptions. Applied to 257 confirmed Starlink reentries (2020–2026) against 500 confirmed operational satellites, STTS achieves V1=250.6x basin separation (p≈0), Recall=1.000 [0.953–1.000], mean detection lead of 471 days before confirmed reentry. V2 does not apply to this domain: deliberate deorbit produces a cliff-structure trajectory, not a gradual monotonic approach, and this is a physically correct characterization of the signal geometry rather than a framework limitation. Applied to the full 15,170-satellite Starlink operational constellation, the framework identifies 108 satellites showing sustained reentry-like orbital signatures at nominal operational altitudes — not in the deliberate deorbit campaign, not noise, correlated with Solar Cycle 25. The largest cluster (38 satellites) operates 50–100 km below nominal altitude with sustained basin excursions spanning weeks to months. Excursion onset rate increased 9× from 2020 to 2025, tracking the solar activity cycle. Applied to the February 2022 geomagnetic storm holdout, the framework returns TERRA_INCOGNITA for all 6 objects with available TLE records (100%), at 570× training corpus distance. The remaining ~32 storm casualties — which reentered before generating individual TLE solutions — are TERRA_INCOGNITA by definition. Applied across five physically distinct domains — turbofan degradation, battery electrochemical fade, bearing vibration, asteroid close approaches, and orbital decay — the identical three-stage pipeline produces valid detection results in each, extending the corpus sufficiency gradient to a new domain and revealing a previously uncharacterized population of operationally stressed satellites in the world's largest LEO constellation.