How accurate is Umbriel?
This is what Umbriel's numbers are checked against, how close they come, and how every chart is put together. You are trusting it with something that matters to you, so all of it is out in the open.
Our positions are checked against NASA/JPL Horizons for solar-system bodies and SIMBAD for stars: the sources astronomers use. A position that looks right is not the same as one that has been checked, so every release is measured against the same data spacecraft navigate by. Since 3 November 2026 the charts themselves have also been checked against a second, independent reference ephemeris on 6,450 births from 1800 to 2100, at every latitude from the equator to the poles, including midnight births, 29 February, daylight-saving changes and the polar circle. Where that check found something less flattering than we had written, this page now says the less flattering thing.
None of this is meant as a last word. If it is useful to you, we are glad, and if it is useful to anyone else making these tools, please take it and go further. We would rather this kind of care were ordinary than rare.
What the numbers are worth
An arcminute is a sixtieth of a degree. For scale, a whole zodiac sign is 30 degrees: 1,800 arcminutes.
| What | Checked against | Worst error |
|---|---|---|
| The Sun | NASA/JPL Horizons, 1650–2050 | under 1″ |
| Ascendant & Midheaven | a second reference ephemeris, 6,450 births, 1800–2100, every latitude | Ascendant 0.75′, Midheaven 0.04′ (the Ascendant is within 0.1′ at every latitude below 60°; its worst case is within a degree of the polar circle, where the Ascendant races through whole signs in minutes) |
| The Moon | JPL Horizons, 312 dates, 1600–2050 | 0.27′ (most within 0.16′; after 2050 it grows with the Earth’s future spin, which is a forecast: 0.4′ by 2060, 1.3′ by 2100) |
| Jupiter, Saturn, Uranus, Neptune | JPL Horizons, 120 moments, 1910–2044 | 2.5″ |
| Mercury, Venus, Mars | a second reference ephemeris, 6,450 births, 1800–2100 | 0.17′ (0.01′ on average) |
| Pluto | a second reference ephemeris, 6,450 births, 1800–2100; JPL Horizons, 1600–2200 (1.65′ in the error budget) | 1.6′ (0.4′ on average; about one birth in sixteen is more than 1′ out) |
| True lunar node | a second reference ephemeris, 6,450 births, 1800–2100; JPL Horizons, 17 moments | 1.55′ (0.33′ on average; about one birth in fifty is more than 1′ out. The JPL spot check of 17 moments between 1920 and 2044 found 0.63′, which was too few moments to find the worst) |
| Mean Lilith (mean lunar apogee) | a second reference ephemeris, 6,450 births, 1800–2100 | 0.33′ (it was up to 9.5′ out until 3 November 2026, when we corrected how it is calculated) |
| True Lilith (lunar apogee) | a second reference ephemeris, 6,450 births, 1800–2100; JPL Horizons, 12 moments | 15′ (2′ on average. The true apogee is a small difference between large numbers, so two programs that define it slightly differently disagree by minutes of arc. The JPL spot check of 12 moments between 1920 and 2044 found 7.2′) |
| Chiron | a second reference ephemeris, 6,450 births, 1800–2100 | 1.4′ (all of it in births from 1842 to 1889; from 1900 on, under 1′) |
| Comets (21 of them) | JPL Horizons | 5.6′ |
| Main-belt asteroids (numbers 1 to 6,214) | JPL Horizons, 641 bodies, 1901–2088 | 1.2′ (most within 0.5′) |
| Main-belt asteroids past number 6,214 | JPL Horizons, 100 real bodies past the grid | 70′ (on a 1950 birth, 95% within 36′; on a 1985 birth 42′, 95% within 20′) |
| Ceres, Pallas, Juno, Vesta, as numbered asteroids 1 to 4 | a second reference ephemeris, 6,450 births, 1901–2088 | 2.3′ (Juno; Ceres 1.1′, Pallas 1.5′, Vesta 2.2′; about 99 births in 100 within 1′) |
| The same four, before 1901 | a second reference ephemeris, 1,983 births, 1800–1900 | approximate: up to 107′ (Pallas, about 1.8°; Ceres 56′, Juno 37′, Vesta 41′; on average 5′ to 16′). The detailed orbits Umbriel carries start in 1901, so a birth in the 1800s gets positions that are good for the sign and often the degree, not the arcminute |
| The same four, added with “Natal + Asteroids” | a second reference ephemeris, 2,454 births, 1950–2050 | 15′ (Pallas; Ceres 8.7′, Vesta 6.4′, Juno 5.1′; this option uses older orbit elements and reaches about 1° at the ends of 1800–2100) |
| Near-Earth asteroids (all 183) | JPL Horizons, 732 checks | 27.5′ (95% within 1.7′) |
| Primary directions (in mundo) | an independent implementation, 42 arcs | 1.02′ (six more, every sinister trine, differ by one fixed amount: the error budget says why) |
| Speculum (the traditional table) | an independent implementation, 73 values | 1.6′ |
| Rising times (oblique ascension) | an independent implementation, 372 cells | 0.30′ |
| Sidereal (Vedic) positions, Lahiri | a second reference ephemeris, 6,450 births, 1800–2100 | ayanamsa within 0.01′ (under a second of arc; every sidereal position is the tropical one minus it, so each is as accurate as its tropical row above) |
| Fixed stars (277) | SIMBAD | see below |
How to read the rows that name 6,450 births. We cast 6,450 charts (5,200 at random between 1800 and 2100 and 1,250 chosen to be awkward) through Umbriel and through a second reference ephemeris, and compared every position. “Worst” is the single worst of those births, not a typical one, and where we give a typical figure we say so. It measures Umbriel against that reference: where the two define a point differently (the true Lilith is the clearest case) part of the gap is definition, not error. The full tables, the worst cases and every birth that came out more than an arcminute off are in the repository (docs/engine-accuracy-2-nov.md).
Every figure in that table is checked against the suite that produced it before each release, and there is a companion page that goes one step further: the error budget is generated from the check runs themselves, so each row carries the number of positions behind it and the day the run happened, and a figure measured against engine code we have since changed comes down until it is measured again.
What Umbriel leaves out
- Nine fixed stars, left out rather than averagedUmbriel computed 286 star positions from SIMBAD and cross-checked every one against a second source. Nine disagreed, so they were left out rather than averaged into something new. The page carries 277.
- Every star shows its position; 37 have a written meaningThe rest show the star's real identity, its Bayer designation and constellation, with nothing invented to fill the space.
- Techniques that always find something say so on the pageSome methods cannot fail to produce a result. Our parans page prints how many you would expect from coincidence alone. The profection schedule says every contact recurs every twelve years. The coalescent chart marks which of its bodies to read with care, and why.
How it is calculated
Everything above is about how close the numbers are. This is about how they are arrived at. The choices any astrology program has to make, and which ones Umbriel made. None of it is secret, and where a choice is contested we say so rather than presenting one convention as the truth.
The ephemeris
Umbriel computes positions itself, in your browser, from published series. It does not call an ephemeris service and it does not ship a table of pre-computed positions. The planetary series are the public-domain ones from Paul Schlyter's How to compute planetary positions; the Moon runs on a truncated ELP-2000/82 series, which is a great deal more work than the simple version and is the reason the Moon lands within a third of an arcminute (to 2050) rather than a few arcminutes. Comets use a universal-variable orbit propagator over a five-year grid of JPL orbital elements, because a comet's elements change enough that a single set is wrong within months.
The practical consequence: your birth data never leaves your device to be calculated. There is no request to make, so there is nothing to log.
Time zones and historical daylight saving
This is where most birth-chart errors come from, and it is rarely mentioned. A chart cast an hour out is a chart with the wrong Ascendant and often the wrong houses.
Umbriel resolves your birth moment through the IANA time-zone database, the same database operating systems use, via your browser's own internationalisation engine, rather than a lookup table we maintain and would have to keep current. That matters for old charts: wartime summer time, the year a country moved its clocks on a different date, a zone that has since been split. Those are all in the IANA data with their historical dates, and a hand-rolled table is where they get lost.
One subtlety: converting a local birth time to UTC needs the offset that applied at that moment, and you cannot know the offset until you know the moment. Umbriel resolves it twice (once to get an approximate offset, then again from the corrected instant) which is what makes a birth in the hour a clock changed come out right.
Tropical, sidereal, and which ayanamsa
Charts are tropical by default, which is what Western astrology means by a sign. The sidereal zodiac is offered alongside it with thirty-one ayanamsas; the default is Lahiri (Chitrapaksha), the standard in Indian astrology and the one the Indian government adopted. Umbriel’s Lahiri matches a reference ephemeris to under a second of arc; until 3 November 2026 it ran about 0.4′ low.
These are not two accuracies, they are two conventions: the sky is the same and the coordinate frame differs. Umbriel does not treat either as the correct one.
House systems
Twenty-six house systems are available, and an option for none: Placidus (the default, and the most widely used in modern Western astrology), Koch, Whole Sign, Equal, Regiomontanus, Campanus, Porphyry, Topocentric, Alcabitius, Morinus, Meridian, Vehlow and the rest.
Not every system exists everywhere. Placidus and Koch have nothing to divide inside the polar circles, where some degrees of the zodiac never rise, and a few others (Campanus, Topocentric, Regiomontanus) can fold back on themselves at high latitudes. When a system cannot be drawn, Umbriel draws Porphyry houses instead, which keep your real Ascendant and Midheaven as cusps 1 and 10, and says so in a note on the chart. It never draws a set of houses that contradicts itself. The rule is checked for every system at 70° north and south, and across every latitude from pole to pole.
House systems disagree with each other, sometimes by a whole sign, and no amount of precision resolves that. It is a difference of definition, not of measurement. If your chart shows a planet in a different house here than somewhere else, the house system is the first thing to check, and it is shown on every chart for exactly that reason.
Every one of them needs an exact birth time. Without one, Umbriel calculates no houses at all rather than defaulting to noon and printing something that looks like an answer.
Aspects and orbs
The default orbs are conjunction and opposition 8°, trine and square 7°, sextile 5°, quincunx and semisextile 3°.
Aspects involving the Sun or Moon get 2° more, applied once, even when both bodies are luminaries. That is standard practice, and it matters: without it, a Sun trine Moon at 7°08′ or a Moon conjunct Saturn at 8°06′ would be missed, and a missing Moon–Saturn conjunction changes the reading.
The luminary bonus is not given to the minor aspects. A quincunx or a semisextile stays at 3° whatever it touches: widening those as well would count a 4° semisextile, which is not a contact so much as a coincidence with a name.
You can set orbs to tight (0.6×), standard, or wide (1.4×), and that scale does apply to every aspect including the minors, so "wide" really will show you a 4.2° semisextile. That is a deliberate choice left to you, not made for you, and the setting is shown alongside the aspects, because an aspect list means nothing without the orb that produced it.
True Node or Mean Node
Umbriel uses the True (osculating) Node by default. The Mean Node is available in settings.
The difference is real: the Mean Node regresses steadily and always reads retrograde, while the True Node wobbles and genuinely turns direct for short stretches several times a year. When Umbriel marks the node direct, that is a measurement, not a glitch, the direction is read from a symmetric window around your birth moment, and the speed shown and the ℞ mark are taken from the same window.
On 6,450 births from 1800 to 2100 the True Node is within 1.55 arcminutes of a second reference ephemeris at worst, 0.33 on average, and about one birth in fifty is more than an arcminute out. (An earlier JPL check of 17 moments found 0.63; that sample was too small to find the worst.) It takes more care than it sounds: the node amplifies the Moon's latitude about eleven times, so the lunar model underneath it has to be very close for the node to come out right. A node that sits within an arcminute or two of a sign boundary is inside the error bar.
Lilith, which one
"Lilith" names at least four different points, so it is worth knowing which one you are looking at. Umbriel offers seven and labels every one:
- Lilith (Mean)The mean lunar apogee: smoothed, and the most commonly published.
- Lilith (True)The osculating apogee: the real one at that instant, which wanders.
- Lilith (Natural)The interpolated apogee. A different quantity again, and the one most often confused with the true apogee.
- PriapusThe opposite point: the perigee.
- Selena (True)The "white moon", as the true apogee of a second convention.
- Selena (White Moon)The other convention of the same name: computed separately rather than assumed identical to the first, because they are not.
- WaldemathThe hypothetical "dark moon". Included because people ask for it, and labelled as hypothetical.
Mean Lilith is within 0.33′ of a second reference ephemeris across 6,450 births. True Lilith is the one to hold lightly: against that reference it is 2′ out on average and up to 15′, and a JPL spot check of 12 moments between 1920 and 2044 found 7.2′. The true apogee is a small difference between large numbers, which is why two careful programs can disagree about it. The seven were first cross-checked against published values on three independent charts, worst disagreement 2.2′; the figures above are the wider measurements.
Precision and rounding
Positions are carried at full precision internally and displayed to the arcminute, because an arcminute is already far finer than any astrological reading can use: a sign is 1,800 of them.
Where Umbriel rounds, it rounds only for display. Nothing downstream reads a rounded number: aspects, dominants, house placement and every derived point are computed from the full value. So a planet at 29°59.6′, displayed as 30°00′, is still in the earlier sign, and is listed there.
The latest improvement
In September 2026 Umbriel's asteroid positions took a large step forward. Every asteroid, Chiron, Pluto before 1885 and every comet now place the Earth in the same reference frame as the body itself, which matters most for near-Earth asteroids, the closer they pass. And the first 6,214 main-belt asteroids now carry NASA/JPL's own orbit at every four-year mark from 1900 to 2100, which brought a 1950 chart to within about an arcminute. Asteroids numbered past 6,214 use a single orbit, so on an older chart give them a little more room; the table above has the figures.
How it is checked
Every release is checked before it goes out. Positions are re-verified against JPL, not against stored expectations, so the check cannot drift along with a mistake.
Two watchers run on a schedule: one recomputes the sky and confirms nothing has moved that shouldn't have, and one loads this website and confirms the live engine still reproduces a JPL-verified chart to under an arcminute, because a half-applied deploy can leave every page looking perfect and every number wrong.
With thanks
Every number on this page rests on centuries of people watching the sky and then doing the harder, less glamorous part: writing it down. Recording it, correcting it, arguing about it, and publishing it in enough detail that somebody else could check the working. That is true of the astronomers whose data we test against, and just as true of the astrologers who documented these techniques precisely enough that they can be tested at all. A lot of this is only checkable because someone took the trouble to be exact.
We are a business, and we think the honest way to say thank you for that is to keep it going rather than to say it. That means holding ourselves to the same standard of writing things down, this page is part of it, and it means building toward something we care about a great deal: a place where astrologers can earn a living from their own work here, on their own terms. That is the thank-you we are working toward.
What this page does not claim
Accuracy of position is not accuracy of meaning. Everything above is about where the bodies are: a question with a right answer, checkable against spacecraft navigation data. What a placement means is not that kind of question, and no amount of arcminutes will make it one.
Umbriel is offered for reflection and self-understanding, not as prediction or advice. We take the astronomy seriously so that the reflection has something solid underneath it.
Free, no account, and the arithmetic above is the arithmetic behind it.
14160+ people cast their chart here in the last 30 days.
Beginner guides · Free calculators · The twelve signs · The planets
Common questions
How accurate are Umbriel's planet positions?
The Sun is within 1 arcsecond of NASA/JPL Horizons from 1650 to 2050 and the Moon within 0.3 arcminutes from 1600 to 2050. On 6,450 births from 1800 to 2100, checked against a second reference ephemeris, Mercury to Neptune are within 0.2 arcminutes at worst except Pluto (1.6 arcminutes), the Ascendant is within 0.75 arcminutes and the Midheaven within 0.05, the True Node within 1.55, Chiron within 1.4 and Mean Lilith within 0.33. True Lilith can be up to 15 arcminutes out.
How accurate are asteroid positions?
The first 6,214 main-belt asteroids use NASA/JPL's own orbit for the nearest four-year mark, and checked against JPL from 1901 to 2088 the worst was 1.2 arcminutes (2.3 for Ceres, Pallas, Juno and Vesta against a second reference). Before 1901 those orbits do not exist, so a birth in the 1800s gets approximate positions: Ceres up to about 1 degree out and Pallas up to about 1.8. Near-Earth asteroids, checked at 732 moments across all 183, land within 1.7 arcminutes 95% of the time. Asteroids numbered past 6,214 carry a single orbit, so on a chart from decades ago they can sit further out: the table above gives the figures.
What does Umbriel check its positions against?
NASA/JPL Horizons for solar-system bodies and SIMBAD for stars: the sources astronomers use. They are independent of any astrology software, which is what lets us check our own work against something outside it. Charts from 1800 to 2100 are also checked against a second, independent reference ephemeris.