How accurate are the positions? A check against the Swiss Ephemeris

Where the positions come from

Planet positions come from Astronomy Engine (MIT licence), which documents an accuracy of about ±1 arcminute and is tested against NASA JPL's Horizons system. Its ecliptic longitudes are true of date and corrected for light time and aberration, the same frame the Swiss Ephemeris uses by default.

The check

The first version of Jyotishya was written in Python on the Swiss Ephemeris, and the browser version is tested against it: planet positions agree to about 0.01°. The Lahiri ayanamsa formula (23.85709° at 1 January 2000, plus general precession) agrees with the Swiss Ephemeris mean Lahiri to about 0.01 arcseconds from 1900 to 2100. The ascendant formula gives the Swiss Ephemeris result to within 10⁻¹³ degrees when both use the same obliquity.

Worked example

The Moon moves about 13.2° a day, so 0.01° is about a minute of its motion. A nakshatra is 13°20′ wide. A difference of 0.01° can change the nakshatra at noon only when the Moon is within 0.01° of a boundary, which happens on about one day in a thousand.

Known differences, stated plainly

Two small ones remain. The ayanamsa is a mean value, while the planets are true of date; the official definition of Lahiri includes nutation, which differs from the mean value by up to about 17 arcseconds (0.005°) over an 18.6-year cycle. Including it would move dasha dates by up to about 2.6 days.

The ascendant uses the mean obliquity of the ecliptic, which can differ from the true value enough to move the ascendant by up to 0.011° at latitude 60°. That changes the ascendant's sign in roughly 4 charts in 10,000. Neither difference changes a day rating, except within seconds of a boundary.

Conventions, not errors

Rahu and Ketu use the mean node, the traditional choice; the true node differs by up to about 1.9°. Houses are whole signs. Dasha years are 365.25 days; a 360-day year, which some software offers, moves dates by 52.5 days for every 10 years elapsed.