Sean Egan
Ask the agent

Research / Orbits

Satellite orbits from the public catalog

The public satellite catalog publishes two-line element sets, short orbit descriptions of every tracked object, and it is the orbit record anyone can read. I ask what it can measure with a stated error: when a satellite maneuvers, whether old satellites ease off station-keeping before they retire, how dense the upper atmosphere is, and whether a full screen finds every close pass that drew an official warning. Two papers with Derek Conklin are under review at Acta Astronautica, OrbitWatch runs the methods every day, and the 16 supercomputer studies had met 36 of their 95 pass lines as of 8 October 2026.

Papers under reviewStudies runningUpdated Live status

2
papers under review at Acta Astronautica, both preprints on Zenodo2026-09
36 of 95
pass lines met across 16 supercomputer studies2026-10-08
490 of 490
officially warned close passes found by a full screen of low orbit2026-10-04
67.7
node-hours on the supercomputer in the last 7 days2026-10-08
Pass lines by studyOne bar per supercomputer study, split by how each pass line read.0246810Screen against official warningsScreen against official warnings: met, 6 of 6 pass lines6 of 6 metClose passes beyond low orbitClose passes beyond low orbit: met, 4 of 5 pass linesClose passes beyond low orbit: not met, 1 of 5 pass lines4 of 5 metCatalog as an air-density sensorCatalog as an air-density sensor: met, 4 of 6 pass linesCatalog as an air-density sensor: not met, 2 of 6 pass lines4 of 6 metRepeated-burn detectorRepeated-burn detector: met, 3 of 6 pass linesRepeated-burn detector: not met, 3 of 6 pass lines3 of 6 metStorm waves in the upper air, pooled over the stormsStorm waves in the upper air, pooled over the storms: met, 2 of 4 pass linesStorm waves in the upper air, pooled over the storms: not met, 2 of 4 pass lines2 of 4 metDensity forecasts and reentryDensity forecasts and reentry: met, 4 of 9 pass linesDensity forecasts and reentry: not met, 5 of 9 pass lines4 of 9 metThermospheric winds from the whole catalogThermospheric winds from the whole catalog: met, 2 of 5 pass linesThermospheric winds from the whole catalog: not met, 3 of 5 pass lines2 of 5 metThe public fuel gaugeThe public fuel gauge: met, 2 of 9 pass linesThe public fuel gauge: not met, 7 of 9 pass lines2 of 9 metWho moves for whomWho moves for whom: met, 1 of 10 pass linesWho moves for whom: not met, 9 of 10 pass lines1 of 10 metDebris kicks on dead objectsDebris kicks on dead objects: not met, 4 of 4 pass lines0 of 4 metCollision chance of close passesCollision chance of close passes: met, 4 of 5 pass linesCollision chance of close passes: not met, 1 of 5 pass lines4 of 5 metError atlas of the catalogError atlas of the catalog: met, 2 of 6 pass linesError atlas of the catalog: partly met, 2 of 6 pass linesError atlas of the catalog: not met, 2 of 6 pass lines2 of 6 metOrbit-change detector, rerunOrbit-change detector, rerun: met, 1 of 4 pass linesOrbit-change detector, rerun: not met, 2 of 4 pass linesOrbit-change detector, rerun: being read, 1 of 4 pass lines1 of 4 metResonant gravity from dead objectsResonant gravity from dead objects: met, 1 of 4 pass linesResonant gravity from dead objects: not met, 3 of 4 pass lines1 of 4 metGravity from dead belt satellitesGravity from dead belt satellites: partly met, 1 of 4 pass linesGravity from dead belt satellites: not met, 3 of 4 pass lines0 of 4 metManeuvers that lead to close passesManeuvers that lead to close passes: not yet read, 8 of 8 pass lines0 of 8 metPass lines per studyMetPartly metNot metBeing readNot yet readPass lines by studyOne bar per supercomputer study, split by how each pass line read.0246810Screen against official warningsScreen against official warnings: met, 6 of 6 pass lines6 of 6 metClose passes beyond low orbitClose passes beyond low orbit: met, 4 of 5 pass linesClose passes beyond low orbit: not met, 1 of 5 pass lines4 of 5 metCatalog as an air-density sensorCatalog as an air-density sensor: met, 4 of 6 pass linesCatalog as an air-density sensor: not met, 2 of 6 pass lines4 of 6 metRepeated-burn detectorRepeated-burn detector: met, 3 of 6 pass linesRepeated-burn detector: not met, 3 of 6 pass lines3 of 6 metStorm waves in the upper air, pooled over the stormsStorm waves in the upper air, pooled over the storms: met, 2 of 4 pass linesStorm waves in the upper air, pooled over the storms: not met, 2 of 4 pass lines2 of 4 metDensity forecasts and reentryDensity forecasts and reentry: met, 4 of 9 pass linesDensity forecasts and reentry: not met, 5 of 9 pass lines4 of 9 metThermospheric winds from the whole catalogThermospheric winds from the whole catalog: met, 2 of 5 pass linesThermospheric winds from the whole catalog: not met, 3 of 5 pass lines2 of 5 metThe public fuel gaugeThe public fuel gauge: met, 2 of 9 pass linesThe public fuel gauge: not met, 7 of 9 pass lines2 of 9 metWho moves for whomWho moves for whom: met, 1 of 10 pass linesWho moves for whom: not met, 9 of 10 pass lines1 of 10 metDebris kicks on dead objectsDebris kicks on dead objects: not met, 4 of 4 pass lines0 of 4 metCollision chance of close passesCollision chance of close passes: met, 4 of 5 pass linesCollision chance of close passes: not met, 1 of 5 pass lines4 of 5 metError atlas of the catalogError atlas of the catalog: met, 2 of 6 pass linesError atlas of the catalog: partly met, 2 of 6 pass linesError atlas of the catalog: not met, 2 of 6 pass lines2 of 6 metOrbit-change detector, rerunOrbit-change detector, rerun: met, 1 of 4 pass linesOrbit-change detector, rerun: not met, 2 of 4 pass linesOrbit-change detector, rerun: being read, 1 of 4 pass lines1 of 4 metResonant gravity from dead objectsResonant gravity from dead objects: met, 1 of 4 pass linesResonant gravity from dead objects: not met, 3 of 4 pass lines1 of 4 metGravity from dead belt satellitesGravity from dead belt satellites: partly met, 1 of 4 pass linesGravity from dead belt satellites: not met, 3 of 4 pass lines0 of 4 metManeuvers that lead to close passesManeuvers that lead to close passes: not yet read, 8 of 8 pass lines0 of 8 metPass lines per studyMetPartly metNot metBeing readNot yet read
Figure 1. Pass lines of each supercomputer study, by how each line read when last scored. A pass line is a result written down before the runs; "partly met" means some of its parts were met. As of 8 Oct 2026, 02:12 UTC. Source: the orbit studies' published status (live status).
Every study, what it asks, and how it stands
Study, and what it asks or foundStatePass lines met
Full-depth screen against official close-approach warningsFound all 490 close passes that drew an official warning between January 2024 and 2 October 2026, against a line of 97 percent.Finished and scored6 of 6
Close-pass rates beyond low orbitClose passes in stretched orbits come at 0.998 of the rate a shuffled catalog gives, so they behave as chance crossings.Finished and scored4 of 5
The catalog as an upper-atmosphere sensorAir density read from how fast 26,111 passive objects fall agrees with the standard model on quiet days in all 15 shells read; the storm-time rise above the model was not recovered in any of the 8 shells from 250 to 650 km.Finished and scored4 of 6
Repeated-burn detectorHeld to the false-alarm rate of objects that never burn, the detector found almost none of the 50 mm/s cross-track burns planted into precise orbits.Finished and scored3 of 6
Storm waves in the upper air, pooled over the stormsTimes each low dead object’s drag response around 43 geomagnetic storms and fits one shared delay with latitude, the speed of the density wave that travels from the poles, against the speeds one research satellite measured.Finished and scored2 of 4
Air density forecast and reentry times from the catalogLearned air-density forecasts had 0.79 of the standard model’s error; predicted reentry times were off by a median of about two days, and missed the accuracy of the official messages (6.2 hours against 3.3).Finished and scored4 of 9
Thermospheric winds from the whole catalogReads the sideways push of the upper-atmosphere wind from the inclination and node drift of thousands of low objects, by height, year and local time, against the standard wind model.Finished and scored2 of 5
The public fuel gaugePlanted burns were recovered 92 to 100 percent of the time, but the fuel left did not predict retirement beyond a satellite’s age (skill 0.004 against a line of 0.02).Finished and scored2 of 9
Who moves for whom: the avoidance mapBurns on a comparison day with no close pass looked almost as protective as real ones (0.82 of the real rate, against a line of one third), so avoidance could not be told apart from routine maneuvering.Finished and scored1 of 10
Debris kicks on dead objectsSearches the daily mean motion of dead debris and rocket bodies for sudden jumps, with the day-to-day air density taken out, and reads the untracked small-debris flux from their rate by height and year.Finished and scored0 of 4
Close-pass chances with error ellipsesGives every close pass in low orbit a chance of collision from each object’s own position error, over 1,005 days of the record.Scored; follow-ups owed4 of 5
Error atlas of the public catalogEstimates each object’s position error in three directions for every quarter from 1959 to 2026, checked against satellites with precise orbits.Scored; follow-ups owed2 of 6
Orbit-change detector, rerun on a better height readingReruns our detector of orbit changes over the whole catalog, reading each satellite’s height from the standard orbit model (SGP4) and counting changes per object.Scored; follow-ups owed1 of 4
Resonant gravity from dead objectsFits the lumped gravity term each repeat-ground-track shell of dead objects feels, with planted terms and a synthetic null, against Kaula’s rule.Scored; follow-ups owed1 of 4
The catalog as one antenna: gravity from the dead belt satellitesFits the two equatorial gravity terms from the drift of every dead geostationary satellite, plants known changes through the catalog’s own fitting, runs a synthetic null, and measures how precision grows with the number of objects.Scored; follow-ups owed0 of 4
Does this change bring it near something later?Flies each detected orbit change 90 days forward on the new orbit and the old one against the live catalog and counts the close passes the change created, with swapped changes as the chance rate.Scored; follow-ups owed0 of 8
The two papers

Does station-keeping relax before retirement? A population-scale test, on 216.9 million element sets, of whether geostationary satellites ease off their station-keeping as they near retirement. Stopping north-south station-keeping turned out not to mark retirement: of 16 satellites that stopped, 3 kept operating in an inclined orbit, 4 later raised their orbits, 1 was abandoned and 8 could not be resolved. The planned comparison could not be run, because only 3 usable histories cleared the detector against a minimum of 15 matched pairs; for 97 of 113 candidates no baseline rhythm of burns could be estimated at all. Preprint, version 2.

A false-alarm control for maneuver detection. Operators do not publish burn logs, so a detector working from public element sets cannot measure its own precision. The paper uses debris and spent rocket bodies, which cannot maneuver, as a running count of false alarms over about 89 million intervals, and withholds the word “maneuver” when the payload rate does not clear that floor by the required factor of 10. Matched to the payloads’ orbits the separation was 8.83, and on the full archive 5.36, both below the requirement. Preprint, version 2.

Both were deposited on Zenodo on 22 September 2026 and submitted to Acta Astronautica later that month. An outside technical review of both papers on 2 October 2026 listed 51 problems, among them a wrong orbit-conversion offset and conclusions the data do not support; the referee study uses that review as its answer key.

OrbitWatch

OrbitWatch applies these methods to the whole public catalog every day: each object’s orbit, how it changes, and its operator’s pattern of life, with a maneuver reported only when the measured false-alarm bound allows it. A language model I serve writes each report in plain language, and code checks every number before it is published.

The supercomputer studies

Each study asks one question of the full catalog record and writes its pass lines before the runs: the result that would count as a yes, with the margin. Of the 16, 10 are finished and scored, and 6 are scored with follow-ups still owed. Across all of them 36 pass lines are met, 3 partly met, 47 not met and 9 not yet read. Figure 1 has each study’s count and the table under it says what each one asks.

Selected results from the finished studies:

  • A full screen of every low-orbit pair, every minute from January 2024 to 2 October 2026, found all 490 close passes that drew an official warning, against a line of 97 percent.
  • Air density read from how fast 26,111 passive objects fall agrees with the standard model on quiet days in all 15 shells read; the storm-time rise above the model was not recovered in any of the 8 shells from 250 to 650 km.
  • Burns on a comparison day with no close pass looked almost as protective as real ones, 0.82 of the real rate against a line of one third, so collision avoidance could not be told apart from routine maneuvering.
  • Planted burns were recovered 92 to 100 percent of the time, but the fuel a satellite had left did not predict its retirement beyond its age.

The studies finished 97 jobs in the 7 days to 8 October 2026, almost all on CPU nodes of 128 cores.

Records