A language model can write a lovely paragraph about your Moon in Scorpio. It cannot tell you that the Moon was in Scorpio.
Those are two separate jobs. One is describing a chart: choosing the words and the tradition to lean on. The other is placing the planets, and that one is plain astronomy. Where was the Moon at this second, and which degree was rising at this place? There is a right answer, and anyone can look it up. (The sign itself is a convention: in the tropical zodiac it is a 30-degree slice of the ecliptic counted from the equinox, not the constellation behind it. What can be checked is the longitude underneath.)
We build software around charts, so we keep the two jobs apart. Positions come from a calculation on our servers. No language model calculates positions, houses, aspects or card draws. The rest of this post is about how we test the first part.
Why positions can be checked at all
Interpretation is a matter of taste and school. Ask two astrologers what a Saturn square means and you will get two good answers.
Positions are different. They follow from orbital mechanics, and NASA JPL runs a public service, Horizons, that returns the position of a body for a given time and observer. We ask it for apparent geocentric positions: seen from the center of the Earth, corrected for light travel time and aberration, measured along the ecliptic of that date. You can treat it as an outside reference. If our number and the reference disagree by more than we allow, the test fails. Nobody has to argue about it.
We do not claim anything from NASA beyond that. Horizons is a public tool we compare against. The tests and the tolerances are ours.
The tolerances
We compare on a set of 57 reference charts. Each one is a fixed birth case, a date, a time and a place, with reference positions taken from Horizons. The tolerances we accept:
| Body | Tolerance |
|---|---|
| Sun | 3 arcseconds |
| Moon | 15 arcseconds |
| Planets | 30 arcseconds |
These apply to dates from 1800 to 2199. Outside that range the allowance is 60 arcseconds.
An arcsecond is 1/3600 of a degree. A full circle of the zodiac is 360 degrees, so these are very small slivers. For scale, the Moon is about half a degree wide in the sky. Our Moon tolerance is a small fraction of that.
Why does the Moon get a wider allowance than the Sun? Some of it is motion. The Moon covers roughly 13 degrees in a day, the Sun about one. A fast body is less forgiving: a tiny difference in the moment you assume turns into a larger difference in position, about half an arcsecond for every second of time. The Moon's orbit is also pulled hard by the Sun, so describing its path takes far more correction terms than the Sun's. This is textbook astronomy. Read it as the reason the table has this shape, not as a description of our engine.
Time is where charts usually go wrong
Most wrong charts are not wrong astronomy. They are wrong input.
Take someone born at 00:40 on a spring night in a town near a time zone border. You need to know which zone applied that night, and whether daylight saving was already in force in that country that year or began a week later. If the clock reading is converted to universal time with the wrong offset, even by an hour, the Moon lands about half a degree away. That is roughly 2,000 arcseconds, more than a hundred times our Moon tolerance. The rising degree moves even more, about 15 degrees on average. The engine can be perfect and the chart is still off.
Historical time makes this worse. Before standard zones, towns kept local mean time set by their own longitude. Zones were redrawn, daylight saving rules changed from year to year, and some places shifted offset for political reasons. So our reference set is not only modern dates. It runs from 1582 to 2100 and includes local mean time, wartime double summer time, clock readings that never happened or happened twice, and a calendar day one country skipped. A good test chart is one where the clock is awkward.
Two practical points follow for anyone building something similar:
- Test the conversion from local time to universal time on its own, apart from the positions. A failure there looks like an astronomy bug and is not.
- Keep birth cases that sit close to date changes and zone borders.
What we do not test
We do not test interpretation, and we do not think software can.
When a chart says "Mars in Aries", that is a longitude plus a naming convention, and the longitude we can check. What Mars in Aries means for a person is a reading, and a reading is the practitioner's work. Tradition gives a vocabulary. The practitioner decides what to say to the person across the table. We have no test for whether that is wise, and we would be wary of anyone who says they do.
The same goes for outcomes. A correct chart tells you where the planets were. It makes no promise about anyone's life.
Check any calculator yourself
You do not need our code to run this kind of test. Take a birth case you know well, put it through whatever chart software you use, then ask Horizons for the same moment and compare the longitudes. Convert the local time to UT by hand first, so you know which of the two layers you are testing. If the gap is larger than the numbers above, one of the two layers has a bug.
Disclosure: the authors work on Esotier, the chart software whose tests are described above.
This post was written with AI assistance and fact-checked by our team before publishing.
Top comments (0)