Updated 18 October 2026
How accurate it is.
Every check of our numbers against the references astronomers use: what was compared, against what, how many cases, when, and the result.
The checks are scripts and tests kept with the code, and their results are written down with the date they were run. Where a test only has a pass mark, it is listed under its own heading with its limit, so a limit is never mistaken for a measurement.
The planetarium: where things are in the sky
The planetarium draws the sky from where you stand. We compared its positions with JPL Horizons, NASA's reference service for where things are in the solar system, and with the US Naval Observatory, at four moments: London at local midnight and before dawn on 14 October 2026, Sydney on 14 October and Denver on 18 October. An arcsecond is a 3,600th of a degree; the full Moon is about 1,800 arcseconds across.
-
The Sun and the planets
Worst error 2.2 arcseconds (Uranus, from Denver)
-
The Moon
Worst error 10.2 arcseconds, with the Moon 3 degrees above Denver's horizon
-
Bright stars
Worst error 0.6 arcseconds (Pollux)
-
Refraction near the horizon
Within 8.3 arcseconds anywhere above 1 degree below the horizon
-
Sunrise, sunset, moonrise, moonset and transits
Within half a minute (the USNO prints whole minutes)
-
Jupiter's four large moons
Within 0.123 of Jupiter's radius (Callisto, 1 December 2026)
The same comparison runs as a test in the code, with wider pass marks: 5 arcseconds for the Sun and planets, 30 for the Moon and 2 minutes for rises and sets.
Moon phases and full Moons
The Moon phase pages work out every phase, full Moon and supermoon, and draw the Moon as it looks at that moment.
-
New Moon, first quarter, full Moon and last quarter
Our times run 0.30 minutes late on average, 0.52 minutes root mean square, all within 1.44 minutes; 140 of 149 within a minute. The minute a page shows is the USNO's in 97 of 149 and one minute off in the rest.
-
Supermoons
The same nine full Moons, his relative distance to three decimals and his distance within 1 km, and no others
-
Blue moons
The same three: 31 May 2026, 20 May 2027 (a seasonal blue moon) and 31 December 2028, and no others
-
The Moon's face: its tilt, libration and lit share
Libration within 0.006 degrees, the pole's angle within 0.002, the point under the Sun within 0.01, the lit share within 0.01 percentage points
The total solar eclipse of 2 August 2027, town by town
The 2027 eclipse pages work out each town's times from the Besselian elements, the numbers NASA publishes for every eclipse. We checked them against four outside sources.
-
Contact times and totality
Median difference 0.1 seconds, largest 0.3 seconds; totality within 0.3 seconds at all 30 places in the path
-
The centre line and the edges of the path
Centre line within 0.28 km, the edges within 0.13 km, totality on the centre line within 0.19 seconds
-
Town times in a printed guide
Median difference 1.9 seconds, 151 of 180 contacts within 5 seconds; totality median 1.7 seconds. The larger gaps are towns where the guide used its own coordinates.
-
Town times on timeanddate.com
Median difference 3.0 seconds at first contact, 5.7 at greatest eclipse and 8.2 at last contact, ours a little earlier almost everywhere; totality median 1.8 seconds at 33 places
-
The longest totality
6 minutes 23.2 seconds on the centre line at greatest eclipse, as EclipseWise gives it (NASA: 6 minutes 23 seconds)
Tonight's sky, town by town
The tonight pages give sunset, twilight, the Moon and the planets for each town.
-
Sunrise, sunset, civil twilight, moonrise and moonset
57 of 58 exact to the minute; the other is one minute out (Reykjavik's sunset)
-
Planet rise and set
8 of 10 exact to the minute, 2 one minute out
Times are for a level horizon at sea level: hills, buildings and haze make a low Sun, Moon or planet rise later and set earlier.
The 3D viewer
The 3D viewer runs at any date from 1800 to 2200 and draws everything at once, so it uses simpler orbits than the planetarium. This is how far that goes.
-
Planet positions
JPL's own figures for the orbits the viewer uses, from 1800 to 2050: within 10 to 50 arcseconds for Mercury, Venus, the Earth, Mars, Uranus and Neptune, 400 for Jupiter and 600 for Saturn. Spot checks: the Earth to Mars at the opposition of 19 February 2027 within 0.01% of Horizons, and every Mars opposition from 2020 to 2035 on Horizons' date.
-
Noon on the globe
The Sun crosses the Greenwich meridian within 15 seconds of Horizons' time
-
Equinoxes, solstices and Moon phases in the time machine
Within 1 to 5 minutes. The Moon phase pages use the more precise method above.
-
Eclipse shadows
How centrally the shadow crosses the Earth (gamma): 0.0701 against NASA's 0.0698 for 2009, 0.1427 against 0.1421 for 2027. Lunar eclipse depth: 1.834 against 1.8436, 1.252 against 1.246, 1.262 against 1.256.
-
The Earth's colour from space
Clear ocean near the middle of the disc 0.94 of EPIC's brightness, the Sahara 1.05, the Amazon 0.95. The colours were fitted to EPIC, so this measures the fit rather than being an independent test. The whole disc reads 1.27 times as bright as EPIC, and clear ocean near the edge (75 to 85 degrees from the centre) 1.48.
-
The famous photos, live
Spacecraft positions within 0.1 km of a fresh run for seven of the eight photos (Apollo 17's Blue Marble is not in Horizons); in In Saturn's Shadow the Earth falls at pixel (868, 593), where Horizons predicts (870, 592)
Eclipse mode's local times run about a minute late on every contact, which is why the eclipse town pages use the Besselian elements instead.
Facts on the pages
Numbers worked out by the code are half of it; the other half is what the pages say. These were checked claim by claim against their sources.
-
The lessons for teachers
Every factual claim in the 25 lessons, each level audited twice. In the first six lessons no answer key was wrong; 17 passages were corrected, sharpened or given a source they lacked, and every curriculum statement is quoted word for word.
-
Planet facts in the viewer
23 values corrected to the current fact sheets, and more on a second pass (Mercury's mean orbital speed 47.36 km/s, the Sun's surface 5,772 K)
-
The famous photos
79 claims in the captions and notes: 54 confirmed, 25 corrected or taken out
-
Topic pages and guides
216 claims, checked in two separate audits: 21 corrected, rewritten, reworded, added or removed, and one source that could not be reached
Also tested, with a pass mark
These run as tests that pass or fail, so what we can give is the limit a result has to stay inside.
- Star map prints: Every body within 3 arcseconds of JPL Horizons (the Moon 12), so on a 50 by 70 cm print the worst is under 0.03 mm out.
- Planet posters: The pole and Saturn's ring opening within 0.06 degrees of Horizons, the face turned to us within 0.1.
- Planet brightness: Jupiter at its brightest opposition within 0.05 magnitudes of -2.94, Mars at its 2003 opposition within 0.06 of Horizons' -2.88.
- The stars' frame: Spica and Regulus within 0.05 degrees of their ecliptic places, the galaxy's pole and centre within 0.0001 degrees.
- The Earth's seasons: The point where the Sun is overhead within 0.3 degrees of NOAA's.
- The sky events calendar: Eclipses within 3 minutes of NASA's times, meteor shower peaks and oppositions within 1.5 days of the published dates.
- Sunspot regions: Within 1.5 degrees of the longitudes NOAA's Space Weather Prediction Center gives.
- Where the aurora is seen: Geomagnetic latitudes within 0.5 degrees of the AACGM-v2 model's at six places, from Hobart to Tromsø.
- Pointing a phone at the sky: The compass correction matches all 100 of NOAA's test values for NOAA's World Magnetic Model 2025 to 0.01 degrees.
What the models leave out
- The planetarium leaves out the Sun's bending of starlight (0.004 arcseconds 90 degrees from the Sun, 1.75 at its edge), the wobble of the Earth's pole (under 0.5 arcseconds), the aberration from the Earth's spin (0.3 arcseconds) and the small drift of the Earth's clock against atomic time (up to 13 arcseconds of hour angle, about 1.5 in October 2026).
- Rise and set times assume a level horizon at sea level.
- The 3D viewer's orbits are JPL's fit for 1800 to 2050; the viewer runs to 2200, where JPL makes no promise.
- Eclipse mode in the viewer gives local times about a minute late; the eclipse pages do not use it.
Questions
- How accurate is the planetarium?
- Against JPL Horizons the Sun and planets were within 2.2 arcseconds and the Moon within 10.2 arcseconds; the bright stars were within 0.6 arcseconds of the US Naval Observatory, and rise, transit and set times within half a minute (checked 14 October 2026). An arcsecond is a 3,600th of a degree.
- How accurate are the Moon phase times?
- Every new Moon, first quarter, full Moon and last quarter of 2026 to 2028 was within 1.44 minutes of the US Naval Observatory, 140 of 149 within a minute (checked 14 October 2026).
- How accurate are the 2027 eclipse times for each town?
- Within 0.3 seconds of Xavier Jubier's eclipse map at all 63 town pages (median 0.1 seconds), and with the centre line within 0.28 km of EclipseWise's path table (checked 14 October 2026).
- Are the planets in the 3D viewer where they really are?
- Yes, from JPL's approximate orbits: within under an arcminute for most planets and about 7 and 10 arcminutes for Jupiter and Saturn, by JPL's own figures for 1800 to 2050. The planetarium uses a more precise theory, within 2.2 arcseconds of JPL Horizons.
Found a mistake?
Tell us at hello@polyxmedia.com or through support, with the page, the value and where the right one comes from. Who makes the site and where its data comes from is on the About page, and every source is on the credits page.