Positions and angles
vs Swiss EphemerisGolden fixtures regenerated and diffed in continuous integration. Ascendant, midheaven, the nine grahas, nutation and the obliquity all pinned to stated tolerances.
The accuracy programme
A wrong degree is the highest-severity bug there is.
Every technique in Jade is checked against an implementation that was written by someone else — because software that only agrees with itself proves nothing. Here is what is verified, what is not, and where the check found a mistake.
Verified, and against what
Aṣṭakavarga
vs Jagannātha HoraFour of the sixty-four benefic rows were wrong on the first pass. Three of them cancelled out — the totals still summed correctly — and only the independent check found them.
Aṣṭakūṭa
vs All 11,664 pairingsThe technique reads only two nakṣatras and two pādas, so the entire input space is enumerable. Four of the eight tables were wrong before that check, including seven zero cells in the yoni matrix.
Yogas
vs Jagannātha HoraTwelve of sixteen agree exactly across the fixture set. The other four are documented divergences with a named option, not silent disagreements.
Transit dates
vs Swiss EphemerisTwo five-year windows a century apart, with an independent bisection on the reference side. Every event count matched exactly.
Workspace isolation
vs A live PostgresRow-level security is enabled and forced on every table holding your data, and a probe tries to read across the boundary on every deploy rather than trusting the configuration.
Open
What is not verified, and therefore not shipped
Absence is a decision. Each of these could have been implemented from a text and left to look right — which is how software ends up confidently wrong.
There is no ṣaḍbala total
Six sources assembled from about twenty sub-components, and authorities differ on at least half. Six components verify exactly against the reference and are exposed individually; ten are not implemented. A total is only as trustworthy as its weakest part, so Jade does not print one.
The East Indian chart is not shipped
It renders correctly as geometry, but the traditional Bengali sign arrangement could not be confirmed against a reference implementation. A plausible guess at a regional convention is exactly what a Bengali astrologer spots in one glance.
Interactive transit times are approximate
The in-browser ephemeris is fast enough to scrub a timeline at sixty frames a second, and its timing error runs from a fraction of a minute for the Moon to about an hour for Saturn near a station. It drives the interface; it does not drive a printed date.
The other direction
Sometimes the reference was the one that was wrong
Verification is not deference. Where the reference implementation disagrees with the classical rule, Jade keeps the classical reading and records why.
- Dig bala’s default method returns values above the 60-virūpa ceiling on a third of the sample — up to 99.95.
- Pakṣa bala returns −41.45 on one chart and 101.45 on another, both outside the classical range.
- Tārā kūṭa never awards full marks, where the classical rule does.
- Varṇa kūṭa inverts Śūdra and Vaiśya.
Two astrologers disagreeing about ayanāṁśa is normal. Software that hides which one it used is unusable. Jade stores the frame, the node type, the house system and the position basis with every chart, and shows all four in the interface.
Check it against your own software
Cast a chart you already know well and compare it degree for degree. That is the only test that matters, and it is free.