Accuracy and calculation method

Reading the figures

• One arc-second is 1/3600 of a degree. A nakshatra pada spans 3°20′, which is 12,000″; the Moon moves about half an arc-second every second of time.
• With the Swiss Ephemeris files and with JPL's own DE441 file the residual is the same for every body (at most 0.24″ and 0.24″), and it is largest at the start of the range, in 1900. Because the DE441 run reads the same planetary data Horizons uses, this common residual comes from the reduction to apparent place rather than from the positions: Horizons states IAU 1976/1980 precession and nutation for this quantity, and the Swiss Ephemeris uses newer models.
• The Moshier fallback stays within 2.16″. The Swiss Ephemeris switches to it silently when its data files are missing, which is why we check which one the deployed API uses.

Which ephemeris the live API uses

On 27 October 2026 we asked the live API (capability planet-positions, tropical) for the Moon and Neptune at 2 moments where the Swiss Ephemeris files and the Moshier fallback disagree by more than the API's four-decimal output. Every answer matched the Swiss Ephemeris data files. The check is tools/accuracy/probe_deployed.py.

Delta T

Dates you send are civil time, so every result also depends on Delta T, the difference between Terrestrial Time and Universal Time. For the 127 past moments in the set, the Swiss Ephemeris value differs from the one Horizons uses by 0.13 s on average and 0.68 s at most, which moves the Moon by at most about 0.37″. For future dates Delta T is a prediction in every ephemeris (Horizons holds its last prediction constant), so we do not compare it there.

What this does not measure

These figures check planetary positions. They do not check the astrological rules built on them: dashas, yogas, matching scores and the rest. Those are covered by the calculation engine's own test suite and by an audit we ran while mapping every engine operation to this API in October 2026. It logged 39 findings, from misread fields and silently substituted defaults to tables with wrong entries. One example: a packed degrees-and-minutes field was read as decimal degrees in many derived calculations (upagrahas, special lagnas, KP sub-lords, sahams), an error of up to 0.4°. Every finding was either fixed in the engine with a regression test or kept out of the public API until it was; planet positions, divisional charts and dashas did not change in the fix of that field. Changes that alter results are listed in the changelog.