The same shell, twice: idealised, and where the catalogue says it is

The population

A measured figure whose population is unstated is not a measurement. The catalogue is a live artefact: it holds objects raising orbit, objects lowering it, spares parked below the shell and satellites whose element sets were fitted days ago. What follows counts 647 objects as this shell, out of 651 in the group.

What is held equal

A comparison that varies two things at once measures neither, so the idealised side is not the published shape. It is the same fleet size, on the same orbit, seen from the same longitudes, on the same Earth:

• Same fleet size. 648 satellites against the 647 in the catalogue, which is the nearest whole number of 12 plane loads. The published design figure is 648.
• Same orbit. The idealised shell flies at the measured mean altitude, 1207.4 km, and the measured mean inclination, 87.903°, rather than the published round numbers of 1200 km and 87.9°. A published altitude is quoted against the equatorial radius while this site models a mean radius, and that datum alone is seven kilometres, which would otherwise sit inside a difference meant to be about arrangement.
• Same longitudes. Both sides are worst-cased across 8 longitudes on the same latitude circle. The meanings differ even so: for the ideal shell those are eight phases of one repeating pattern, for the real one they are eight genuinely different geometries.
• Same Earth. The spherical Earth of radius 6371 km that the rest of the site uses. What the ellipsoid would change is measured further down, on its own.
• Several windows. 3 start times spread across the snapshot's validity, each 48 hours at 20 second steps. A single window is one sample of a time varying geometry, so the range across them is reported rather than one of them.

The comparison

At the mask this shape is published with, 25° from latitude 55°, the two engines agree completely: both say the shell closes and never opens. That agreement is a result, and the more useful one is what happens when the ask gets harder. At 55°, which is what a terminal that will not look near the horizon needs, they part company.

Most of the difference is a convention, not the ephemeris

The middle column is the point of this page. A Walker-delta constellation spreads its planes across 360 degrees of right ascension, and that is what this site models, for every shell, everywhere. It is right for an inclined shell. It is wrong for a near-polar one, and quietly: at 87.9° inclination a plane and the plane opposite it trace almost the same ground track, so twelve planes spread over the full circle are six distinct tracks with a near-duplicate of each. Real near-polar shells are flown as star patterns over 180 degrees, which gives twelve.

Splitting the difference that way accounts for most of it. At the 55° mask the site's model reports 11.0 min. Changing nothing except the plane span brings that to 60 s, a factor of 11. Everything else about a real fleet, the dispersion in altitude, the phasing that is nobody's design any more, the satellites that are not where a pattern would put them, moves it a further factor of 3, to 20 s.

The direction is worth stating plainly, because it runs against this site's usual warning. Every omission listed on the method page makes real service worse than the idealised figure. This one does not: for a near-polar shell the idealised figure is the pessimistic one, and the real fleet does better than the model predicts. That is a defect in the model rather than a margin of safety, and it is now on the record.

Where they disagree, and by how much

Both engines say this shell closes at every latitude it reaches, up to a 50° mask. Past that the two curves separate.

What the spherical Earth costs

Both columns above sit on the same sphere of radius 6371 km, so that the difference between them is about arrangement rather than about geodesy. The ellipsoid is real and worth measuring, so it is measured here instead: the same real elements, the same windows, with the site placed on WGS-84 with a true geodetic vertical. At latitude 50 that moves a site about 22 km and tilts its local vertical by about 0.19 degrees.

What this does not settle

• One snapshot, one shell. Element sets retrieved on 2026-09-08T19:46:53Z, propagated from windows starting 2026-09-07T11:00:00Z, 2026-09-08T11:00:00Z, 2026-09-09T11:00:00Z. This says what this shell was doing then.
• SGP4 is a model too. Its accuracy decays by roughly one to three kilometres a day away from an element set's epoch. The population here has a median epoch age of 0.11 days and a worst of 1.14, and the windows run 48 hours, so the far end of a window carries a few kilometres of position error. That is small against a footprint hundreds of kilometres across, and it is not nothing.
• Visibility is not service. No beams, no link budget, no capacity, no licensing, no terminals. The caveat below is the whole of it.
• The plane-span finding is about near-polar shells. For an inclined shell, planes 180 degrees apart in right ascension are genuinely different planes and the Walker delta convention is the right one. Nothing here says the calculator is pessimistic at 53 degrees.

Provenance

Element sets: 2026-09-08 snapshot of the oneweb group of the United States Space Force public catalogue, redistributed by CelesTrak, retrieved 2026-09-08T19:46:53Z and vendored into this project with a content hash. The build never fetches anything and never reads the clock, so this page is a pure function of the repository and reproducible from a commit. Propagation is SGP4, checked on every build against the 158 states published with the 2006 AIAA paper that defines the algorithm: the largest disagreement anywhere in that set is 5 micrometres of position.

The comparison is stored as a committed artefact keyed by the data, the pinned instants, the configuration and the code that computes it, so a published figure here cannot outlive any of them. Current key a991864c1a2a.

The idealised counterpart is A 648-satellite polar shell, the OneWeb pattern, where the same shell is flown as a perfect Walker-delta and the calculator is interactive. The validation page measures what the altitude dispersion visible in this population costs the idealised model, and the method page states what the model leaves out.