How we calculate your chart

1. The astronomical layer

Planetary positions come from the Swiss Ephemeris, via its built-in Moshier analytical model — the same engine used by professional astrological software. Our configuration is set once, at process start, and applies to every chart:

• Ayanamsa: Lahiri (Chitrapaksha) — the Government of India standard, and the assumption behind the classical material our rules are drawn from.
• House system: whole-sign — each sign is one house, counted from the Lagna. Every rule in our knowledge base was written against this convention. We don't offer Placidus or Sripati, because our rules would silently mean something different under them.
• Rahu and Ketu: true node, not mean node. This is a school choice: many almanacs and programs print the mean node, which can sit up to about 2° from the true position — enough, near a boundary, to put Rahu and Ketu in a different nakshatra or sign. Your dasha sequence is not affected; it starts from the Moon's nakshatra.
• Birth time is resolved in your actual timezone, including historical daylight-saving transitions.

Two schools that define the chart differently get their own settings, in their own labelled blocks and never mixed into the kundli above: KP, when a chart follows it, uses K.S. Krishnamurti's own ayanamsa — his original, often called "KP old" (Swiss Ephemeris mode 5), not the later "KP New" some KP software defaults to — with Placidus cusps, as KP requires; the Western astrologer reads the tropical zodiac with Placidus houses.

From those positions we compute deterministically: sign, degree, nakshatra and pada for all nine grahas, retrograde state, the Lagna, dignity and combustion, Ashtakavarga, Shadbala, birth panchang, the divisional charts of Brihat Parashara Hora Shastra (all sixteen are implemented; a reading uses the Navamsha and the chart of the question's matter), your Vimshottari dasha timeline (the antardashas of the next eight years and the pratyantardashas of the next eighteen months), and a calendar of the slow planets for the next three years — when Jupiter, Saturn, Rahu and Ketu change sign or turn retrograde, when Sade Sati begins and ends, and the real eclipses. You can see the first of these for yourself on our free kundli tool, and the day's equivalent on today's panchang.

A recorded birth time is rarely exact to the minute, so we recompute the chart five minutes either side. Anything that changes within that margin — often the Navamsha Lagna, sometimes the Lagna itself — is stated as conditional rather than as fact, and dasha dates carry their margin in days. Without a birth time, every dasha date is given as a range.

2. The deterministic layer — the rule engine

None of the above involves AI. Neither does the next step.

We maintain a knowledge base of classical rules stored as machine-readable schemas — not prose for a model to interpret, but conditions a program evaluates. The Parashari set — 44 yogas and 10 doshas — is encoded as executable rules (five of them — Pravrajya, Kalatra Sukha, Putra Sukha, Videsh Vasa and the curse of the father's line — are read only for the question they belong to), plus 3 transit-level conditions (a planetary war, which needs two planets less than 1° apart, and two slow conjunctions that count for you only when they touch your chart); Jaimini, Lal Kitab and Nadi charts add their own rule sets. A rule engine walks every one against your chart facts and returns exactly the set that fires, with the classical cancellations applied — and grades each yoga by the planets that form it: their dignity, combustion and house, and whether the dasha now running belongs to one of them.

Timing is computed the same way. Each transit is judged from your Moon sign by the rules of Phaladeepika (the favourable houses, obstruction by other planets, dignity and Ashtakavarga points), and for a question about marriage, career, money, home, children, study, travel, family or spiritual life, every antardasha of the next eight years is ranked by how its two lords connect to the house in question, with the classical reason for each link and any obstacle shown, never hidden. For health, the dasha and transit periods may name a time that asks for extra care, and the area of concern, as a forewarning; lifespan and death are never timed. The reasons are classical; the order in which we weigh them is our own and has not yet been calibrated against known event dates.

This inverts the usual order. The yogas in your reading are not the ones a language model recalled or thought sounded plausible — they are the ones a deterministic evaluation of your chart actually produced. Run the same birth data twice and you get the same set, every time. You can read how individual combinations work in our Raj Yoga and dasha explainers.

Every one of those rules names its source — most often Brihat Parashara Hora Shastra and Phaladeepika, by chapter and verse, and for horary questions Prashna Marga. Seven of the doshas — Kaal Sarp, Pitra, Grahan, Guru-Chandal, Shrapit, Vish and Angarak — appear in no classical text; they are computed because people ask about them, and the reading labels them modern.

3. The AI layer — and its limits

Only now does a model get involved. Four core specialist agents (astrology, numerology, horary/Prashna, transits) — plus a tropical Western astrologer and a BaZi master when a chart follows those traditions, and a palmist and a face reader when the person has added palm and face photos with their consent — each receive the facts computed for their own system and a filtered slice of that system's knowledge base: the astrology voices the chart, the numerologist only the numbers of the birth date and name (each labelled Pythagorean, Chaldean or Vedic Ank Jyotish) with their computed cycles, never the birth chart. The Indian schools a chart follows (Jaimini, Lal Kitab, Nadi, KP) arrive as labelled fact blocks inside the astrology specialist's prompt; the calendar systems (Nine Star Ki, Mahabote, Mayan, Tibetan) inside the numerologist's. A planner decides which of these voices a question actually needs (all of them on a deep reading, or exactly the ones the person picks on a custom one), so a simple question is not billed for eight specialists. On a deep reading the traditions a chart does not follow are shown as labelled extra perspectives and never vote on the answer. The two photo-based voices read only a fixed feature set extracted once from each image — the hand's type and proportions and the face's three zones measured from the landmarks in the photo's own pixels, the lines, the signs on the mounts, the structure of the face — with rules cited to Cheiro, Benham, Samudrika Shastra and Mian Xiang; the palm is read from the hand the person writes with, the hand type first. They describe character and, only where a cited classical scale fired on a readable feature (Benham's and Cheiro's fate-line scale, Benham's sun, life and head lines, Cheiro's marriage-line heights beside an influence line, the Mian Xiang age positions at region level for the fortune of a region's years), give timing as an approximate age window ("roughly between 30 and 49", the whole span — a window ten years or wider is never narrowed to a single age); a hand's window also needs another line or the other hand to repeat the sign. Never a date, month, year or festival, never a window narrower than the scale's band, never lifespan, death or a diagnosis, never an illness or an accident predicted to happen (an area that asks for extra care may be named as a forewarning; only the palm gives it an approximate age window, from an island on the life or head line that the other hand or the other line repeats — the face never times health), and never a judgement of health, age or appearance from the photo itself. These two voices are told the person's age bracket in words, never the date of birth; whether a window is behind, open now or ahead is computed in code. A supervisor agent synthesises every output into one reading; where two specialists disagree, the reconciliation weighting is fixed in code, not decided by the model. Methods that read the same data — the birth chart and its transits — count once, a method that cannot judge the question asked (palm, face and numbers on a yes-or-no event) does not vote on it, and the answer is given in words — leans yes, uncertain or leans no — with a star rating for how strong and how independent the agreeing evidence is. A "when" question is answered with its most likely window and the dasha period behind it (or a plain "no clear window" when the periods ahead show none), and the stars then rate the confidence in that window, never a yes or no. We do not print a percentage: no method here has been calibrated against outcomes well enough to earn one.

The models' job is interpretation and language. What they are explicitly instructed not to do, in the prompt itself: contradict the computed chart facts; invent yogas, doshas, dignities or dasha lords beyond those supplied; or cite any classical entry not present in the reference slice they were given. On horary questions the Prashna specialist's yes/no verdict comes from the rule engine, and the model is forbidden from reversing it — its role there is phrasing and remedial guidance only; the final answer weighs that verdict with the other specialists'.

4. The validation layer

Instruction alone isn't enough, so the specialists' notes are checked before the answer is written.

Claims in the notes are validated against the chart facts computed in step 1. In the astrology agent's notes, a named yoga that doesn't exist in our knowledge base is flagged as fabricated, and a yoga that exists but did not fire for your chart is flagged too. Dasha lords are checked against your computed dasha; sign and dignity against the computed position; numerology values against the deterministic calculation; a timing window from the astrology or transit agent against the dasha periods and transit spans computed for your chart — one that matches none of them is removed, and one narrower than its period is widened to the period. The default response to a failed check is removal from the notes, and the answer is written from the checked notes.

We'd rather state this precisely than impressively: coverage differs by specialist. The numerology agent runs a structured claim contract with field-level checking, and its notes are checked sentence by sentence: a number that differs from the calculation, a percentage, a birth-chart term or a timing window outside its computed cycles is removed. The astrology agent runs a prose-level check that catches fabricated yoga and dignity statements. The Western and BaZi agents' notes are checked against their own computed charts: a transit date the ephemeris did not produce, a wrong natal sign or house ruler, an impossible pillar, a year paired with the wrong pillar or a wrong ten god is removed. Every agent's notes are also checked against what was computed for it, and these are taken out: a timing line or a dated dasha statement that names a period your dasha timeline does not contain, or dates it outside that period; a natal placement your chart contradicts; a planet put in the wrong sign "now"; a Jaimini karaka, Lal Kitab house or Nadi conjunction your chart does not show; another system's vocabulary; and any percentage, probability or odds. The Prashna specialist's verdict is deterministic by construction, and it joins the other specialists' verdicts in the weighted consensus. The transit agent's verdict cannot reverse the transit judgement computed in code, and its timing window must stand on a computed period. When two agents' windows do not overlap, the reading says so and its confidence rating drops by a star. Finally, the answer itself is checked line by line as it is written: its verdict line is the computed one and no percentage, probability or odds reaches you — nothing is regenerated, so what you see is exactly what is saved. The answer is not re-checked yoga by yoga: a yoga it names that did not form is logged rather than removed, because the notes it was written from were already checked. That is the honest current state, and the part we're still extending.

What we don't claim

We do not publish an accuracy percentage. You'll see competitors advertise figures like "90% accurate" — there is no methodology under which such a number could be honestly derived for astrological prediction, and quoting one would contradict everything above.

What we stand behind is narrower and real: the astronomy is computed to a published standard, the classical rules are evaluated deterministically and cite their sources, and the specialists' notes are checked against those calculations before the answer is written. Astrology remains interpretive. Our calculations are not, and that is the part we hold ourselves to.