Accuracy & Methodology

The core principle: astronomy is computed, never guessed

Every planetary position, house, dasha date, and divisional chart in your reading is computed deterministically from real astronomical data — before the AI ever sees it. The AI's job is strictly to interpret those verified facts using classical Parashari principles; it never calculates or guesses a placement. This is a hard rule enforced in the system prompt itself: the AI is explicitly instructed to use only the exact figures we give it, never to re-derive or invent one.

The engine

Chart calculations run on astronomy-engine, an open, MIT-licensed astronomy library built on VSOP87 (planetary theory) and ELP2000-82 (lunar theory) — internationally recognized, peer-reviewed analytical models developed at the Bureau des Longitudes, the same class of theory used in multiple national astronomical almanacs. We compute Lahiri ayanamsa, the Ascendant, and the lunar nodes (Rahu/Ketu) independently on top of it using standard published formulas, then derive every divisional chart (D2 through D60) as a deterministic mathematical subdivision of those same verified positions — no separate, less-precise calculation ever enters the pipeline.

We're precise about this because "NASA-based" gets thrown around loosely in this space, and we don't want to make a claim we can't stand behind: our engine is not a copy of NASA JPL's numerically-integrated ephemeris. It's an independently-derived theory validated to observatory-grade precision — which is a claim we can actually prove, below.

How we validated it

Rather than assert precision, we measured it — directly against Swiss Ephemeris (the JPL-derived engine widely used across professional Vedic software) across 13 test charts spanning both hemispheres, near-polar to equatorial latitudes, dates from 1950 to 2020, and three confirmed planetary retrograde windows (Mercury, Mars, and Saturn, October 2020).

On that last row: the single D60 disagreement we found wasn't an engine error — it was a planet whose true position happened to sit within 0.06 arcminutes of an exact division boundary, so which side it falls on is genuinely a coin flip at that resolution. That same ambiguity would appear comparing any two legitimate ephemeris sources at that precision, including two different runs of the same software. We're telling you about it because a validation report that only shows wins isn't a validation report.

What this means in practice

A sign spans 30° (1,800 arcminutes). A nakshatra pada — the finest classification used in your dasha, yoga, and interpretation logic — spans 3°20' (200 arcminutes). Our largest measured deviation from Swiss Ephemeris, across every category above, was 0.35 arcminutes: roughly 500 times tighter than the smallest boundary that could change what your chart says about you. In practical terms, this precision has no chance of moving a planet into the wrong sign, house, nakshatra, or pada for any real birth chart.

What the AI does — and doesn't do

Once your chart is computed, planetary aspects (graha drishti) and combustion are also calculated deterministically — not left for the AI to infer — following classical BPHS rules (e.g. Saturn aspects the 3rd, 7th, and 10th houses from itself; Mars aspects the 4th, 7th, and 8th). The AI is handed these as verified facts alongside your placements, with explicit instructions to reason from them, never around them. If you ever see the AI state a placement, aspect, or dasha date, it's reading it directly from this pipeline — not recalling it from general astrological knowledge.