Overhead · the method
How this sky is computed
Every number on the screen comes from somewhere, and every approximation has a size. Both are written down here, including the ones that are not flattering.
The stars
The Yale Bright Star Catalogue, 5th revised edition — 9,096 stars, which is essentially every star the naked eye can see from a dark place. Positions are stored to the milliarcsecond and drawn down to magnitude 6.5. Colour comes from each star's published B–V index; size and brightness come from its magnitude, on a scale where a star five magnitudes brighter is a hundred times the light, because a linear scale throws away the only hierarchy a star field has.
Precession, which is not optional
The catalogue is measured against the sky as it stood in the year 2000. The sky has moved since: the Earth's axis swings slowly round a circle, and in 2026 the two reference frames are 0.36° apart — most of the width of the Moon — growing by 50 arcseconds a year. Draw a catalogue star and a planet of today on one sphere without correcting for it and every planet sits visibly outside the constellation it is actually in, while the picture still looks entirely plausible.
So every star is rotated to the equinox of date before anything is drawn, by the rigorous method rather than the linear approximation — the linear one goes quietly wrong near the pole, and the pole is where Polaris is.
The planets, and where they are approximate
Positions come from a periodic-terms series and are accurate to well under a tenth of a degree in longitude. Latitude is the honest gap: the Moon carries its real latitude, which matters because it strays up to 5° from the Sun's path, and the Sun and the two nodes are exactly on that path by definition. The five planets are drawn on the ecliptic and can be up to about 2° off it — Mercury and Venus at the extreme, under a degree for the rest. That is enough to put a planet in the right constellation beside the right star, and not enough to measure with. This map only ever claims the first.
The nakshatras
Twenty-seven equal divisions of 13°20′, measured sidereally on the Lahiri ayanamsa, which is the Indian government's own standard. Each is anchored to a junction star — its yogatara — and the figure drawn on screen is the asterism that star belongs to, so Krittika draws the Pleiades and Hasta draws the quadrilateral of Corvus.
Those figures are built from catalogue positions, not from a diagram: every star in every figure is a named entry in BSC5, and the drawing is checked automatically for stars in the wrong constellation, for lines nobody would trace, and for shapes that cross themselves. A hand-drawn figure looks right and cannot be checked; this one can, and the check has already found and fixed one wrong star.
A nakshatra is a division of the ecliptic, so a patch of sky far from the zodiac band has a longitude but not really a home. When you point somewhere off the band the readout says toward rather than in, and it means it.
Which way you are facing
On iPhone and iPad, Safari reports a heading that iOS has already corrected to true north, and it is used as given. On Android, the browser reports magnetic north — and the angle between the two reaches 20° in parts of the world, which is two hand-widths of sky. That correction is computed here from WMM2025, the World Magnetic Model, at your own latitude and longitude. The model is published for five years at a time; if it has expired the readout says so rather than answering anyway.
Some devices have a gyroscope and no compass at all. There the sky will be the right shape pointing the wrong way, the readout says exactly that, and the two arrows let you turn it until it matches what you can see. A phone's magnetometer is also easily confused — by a car, a desk, a magnetic case — which is a thing no model can know about and you can. That is what the arrows are for.
Where you are
Before you give it anything, the map guesses from your browser's time zone, which needs no permission and puts you within a few hundred kilometres — close enough that nothing overhead looks wrong and the horizon might. It says on the front page that it is guessing. Tap Use my location and it asks the browser once, coarsely: a sky map needs a few hundred metres, not GPS, and asking for GPS-grade accuracy means ten seconds outdoors and never indoors.
The answer is kept on your device and is not sent anywhere. There is no server in this product.
What is deliberately left out
Refraction — the atmosphere lifts everything near the horizon by about half a degree, all together, which is invisible in a picture with no ruler in it. Proper motion — stars drift, and over the 26 years since the catalogue's epoch the fastest movers have moved less than a drawn dot. Nutation and aberration — about 20 arcseconds each. Lunar parallax is the one real omission: the Moon is close enough that where you stand on Earth shifts it by up to a degree, and that is not corrected here.
How it draws
Two renderers behind one interface. The first uses your GPU with a bloom pass; the second is plain 2-D canvas and uses nothing a phone has not had since 2013. Which one you get is decided by measuring the first one for a second and demoting it if it cannot hold the frame — never by guessing from your browser's user-agent string, which cannot tell an iPad from a Mac and cannot tell two phones with the same name apart at all. Both draw the identical sky.
Method · BSC5 positions, precessed J2000 → equinox of date · planets tropical, converted with the obliquity of date · sidereal longitudes on the Lahiri ayanamsa · azimuth from true north · magnetic declination from WMM2025 · no refraction, no proper motion, no nutation, no lunar parallax.