There is an open-source browser model that renders the sky on a flat vault and on a globe, side by side, and reports that both fit what one observer sees. It is offered as proof that the Earth’s shape is a choice of coordinates rather than a fact.
The model is real, the code is public, and it works. This page takes it seriously enough to read its own files.
The pitch is a strong one, and worth stating at full strength. A browser simulation renders the positions of the Sun, Moon, planets and stars as a flat-Earth observer would see them, and as a globe observer would see them, and the two agree. Sunrise happens on time. Eclipses land where the catalogues say. Nothing is fudged, and the code is there to inspect. If a flat model reproduces the sky as faithfully as the round one, the argument goes, then the shape of the Earth is not something observation decides. It is a projection you pick.
That argument would be serious if the two models were being asked the same questions. They are not, and the model’s own documentation is the clearest place to see why.
The project’s notes describe a renderer built on a single normalized constant, with the flat-Earth radius set to one. In the author’s own summary the model carries no Earth radius, no astronomical units, no kilometers, and no great-circle trigonometry. Every distance in it is a bare ratio.
Read that as a specification rather than an apology and it is exact. The model computes directions for one observer. It does not compute distances, because it holds none. And a model that holds no distances cannot be wrong about any distance, which is not the same thing as being right about them.
A planetarium dome is a hemisphere a few tens of meters across, and it can put the entire night sky in the correct place for a seated audience, including the motions of the planets and the turning of the stars through the year. Nobody concludes from this that the universe is thirty meters wide. The dome reproduces the sky because the sky, for one observer looking outward, is a set of directions and nothing else. Any surface that can carry those directions will do.
This is the whole of the simulator’s achievement, rendered in a browser instead of plaster. It is a real achievement and it settles nothing about the world, for the same reason the dome does not.
A model that computes directions still has to know where the bodies are. This one does not derive that from flat-Earth premises, because there is no flat-Earth theory of planetary motion to derive it from. It reads standard ephemerides — the same numerical solutions for planetary and lunar positions that space agencies use, computed on a heliocentric, spherical-Earth model — and renders their output onto a disc.
That is the hinge. The sky the simulator draws is not predicted by the flat model. It is imported, from a body of work whose equations assume the thing the simulator is being used to dispute, and then repainted on a different surface. Remove the imported ephemerides and the flat model has nothing to say about where Mars will be next Tuesday.
Eclipses are the sharpest test anyone offers, because they demand a real prediction: a place, a minute, a duration, decades ahead. The model ships a large set of eclipse demonstrations, and they reproduce the catalogue correctly.
They reproduce it because they are replaying it. The demonstrations are driven by the published eclipse canon, which is computed from Sun–Earth–Moon geometry on a sphere. The repository is candid about the distinction: its own notes carry flat-Earth eclipse prediction as an unfinished placeholder rather than a working component.
The most interesting thing in the project is a file it ships for a different purpose. To demonstrate long-haul routes, it includes a real flight track as a mapping file, with per-waypoint data taken from the actual flight. That data includes airspeed, in knots. The model uses the track to show a path across its disc, and reports the aircraft’s progress as a central angle over time.
Carrying both quantities at once is more than the model can afford, because they are related by the one number it declines to hold. Speed multiplied by time is a distance. That distance divided by the angle swept is a radius:
Run it on the shipped figures. A true airspeed near 480 knots over about 12.5 hours is roughly 6,000 nautical miles of track. The corresponding central angle is about 102°, which is 1.78 radians. Divide:
The Earth’s actual radius is 3,443 nautical miles. The model’s own data file, read with one division, returns it to within about two percent — the sort of margin you would expect from rounded airspeeds and a great-circle track that is not perfectly flown.
Each of these is a single quantity, each has been measured for a long time, and each is outside what a single-observer angle renderer can represent.
A second observer. At one instant Polaris stands overhead at the north pole, sits on the horizon at the equator, and is invisible from the entire Southern Hemisphere, which sees a different sky turning the opposite way around a second pole. One vault over one disc cannot hold two opposite centers of rotation with half the sky hidden from half the observers. (The southern sky.)
The Sun’s width. The solar disc spans about half a degree at noon and the same at sunset. On a vault the Sun is a nearby lamp whose distance from the observer changes enormously through the day, so its apparent size should swing by roughly two to one. It does not move measurably. This is an angle, which is the one kind of quantity the model claims to own, and it is the wrong answer. (The Sun stays the same size.)
Any distance at all. Radar ranging to the planets, lunar laser ranging, stellar parallax, undersea cable lengths, flight times, network latency. Every one of them closes on a sphere of about 40,000 km circumference. The model carries none of them, and that is the design, not an oversight. (Long-path radio.)
Three things, and they should be said plainly.
It is honest in its documentation. The statement that it carries no radius, no AU and no kilometers is the author’s, not an accusation, and the eclipse placeholder is labeled as a placeholder rather than dressed up. A great deal of flat-Earth material is not this candid.
It is correct about projection. For one observer computing directions and nothing else, flat and round genuinely are interchangeable, and anyone who claims otherwise is overstating the globe’s case. The dome proves it.
And it is a real piece of work. The renderer functions, the code can be read, and the demonstrations run. The problem is not competence. It is that the question the model answers is narrower than the question it is being cited for.
Until then the honest description of the model is the one its own files support: a single-observer angle renderer, fed by spherical-Earth ephemerides, replaying a spherical-Earth eclipse catalogue, shipping a flight file from which the Earth’s radius can be recovered by division. It is a picture of the sky. It is not a theory of the world, and it does not claim to be.