What is calculated, what an AI wrote - and the line between them.

The page above states the rules. This long read shows the working behind them: the classical statements the arithmetic exists to serve, one birth carried through every step from clock time to dated periods, the same engine run against a chart printed in one of this site's source books, how to read each kind of result the product returns, and how far a few minutes of birth time move it. Every number below was produced by running the engine's own functions; none was typed from memory.

What the texts ask the arithmetic to supply

Brihat Parashara Hora Shastra gives the whole specification in a few lines (Ch. 3 Sl. 2-6). The lights that do not move are the nakshatras; the ones that move through them are the grahas. The zodiac holds 27 nakshatras counted from Ashvini, and the same circle is cut into 12 rashis counted from Mesha. The rashi rising is the Lagna, and a person's results are deduced from the Lagna and the grahas. So a chart needs three things computed and nothing more: where each graha stands on a circle measured against the stars, which twelfth and which twenty-seventh of the circle that is, and which rashi was rising at the birth place.

The translator's addition to that passage supplies the two figures behind most of this page's caveats. A rashi takes roughly two hours to rise, so a degree takes about four minutes, and the true duration depends on the latitude. The Moon moves 13 to 15 degrees a day, while Saturn moves about 2 arc-minutes. The Lagna is the fastest thing in a chart and the Moon the second fastest. Whatever is sensitive to birth time is sensitive through one of those two.

Timing comes from Ch. 46. Sl. 2-5 call Vimshottari the most appropriate dasha for people in general. Sl. 12-14 give its lords in order from Krittika - Sun, Moon, Mars, Rahu, Jupiter, Saturn, Mercury, Ketu, Venus - across a 120-year span, Sl. 15 gives their years as 6, 10, 7, 18, 16, 19, 17, 7 and 20, and Sl. 16 says how to find the balance left at birth from how far the Moon has gone through its nakshatra. The sources are not unanimous here: Saravali's dasha chapter (Ch. 41) teaches the Moola Dasa, a different scheme. This engine's main timeline is Parashara's, and the conditional systems of the same chapter sit under alternate dashas.

What these chapters do not supply is a way to find a longitude. Parashara's account starts with the positions already in hand. So the positions come from modern astronomy - the formulas named on the page above - and the rules applied to them come from the texts. One convention joins the two: the ayanamsa, the angle between the equinox-based zodiac astronomers compute in and the star-based one the texts mean. Advanced Predictive Astrology says in its introductory pages that its author follows the Lahiri ayanamsa only and that its example horoscopes were cast with computer software. This engine uses Lahiri too, which is what makes the comparison further down possible. The ayanamsa calculator prints the value for any date: 23.8531° at noon on 1 January 2000 and 24.2266° on 5 October 2026.

One birth, carried through every step

Take a birth at 06:40 on 14 March 1992 in Jaipur. It is chosen because it falls a few minutes after sunrise, which gives a check that needs no ephemeris.

1. 1.Place. Jaipur is in the bundled city list, so no network lookup happens: 26.9124° N, 75.7873° E, zone Asia/Kolkata. The zone's own rules give the offset in force at that minute, +5:30.
2. 2.Instant. 06:40 less 5:30 is 01:10 UTC. That instant is Julian Day 2448695.54861, which is -0.07801373 Julian centuries from J2000 - the one number every formula below takes as input.
3. 3.Tropical positions. The solar formula returns 353.7136° for the Sun and the lunar series 109.0420° for the Moon, both measured from the equinox of that date.
4. 4.Ayanamsa. For that Julian Day the Lahiri value is 23.74415°, or 23°44'39″. Subtracting it gives the sidereal Sun at 329.9694° and the sidereal Moon at 85.2979°.
5. 5.Sign, nakshatra, pada. The sign is the longitude divided by 30 with the fraction dropped; the nakshatra is the longitude divided by 13°20'; the pada is the quarter of the nakshatra the point has reached. The Moon's 85.2979° is the third sign, Gemini, at 25°18', in the seventh nakshatra, Punarvasu, second pada.
6. 6.Lagna. Local sidereal time is 265.1937°. With the obliquity of the ecliptic and the latitude, the rising degree comes to 353.2914° tropical, 329.5472° sidereal: Aquarius 29°33'. Aquarius becomes the first house, Pisces the second, and so on round the circle; each graha's house is counted from there.

Now the check. The engine's own sunrise routine puts sunrise in Jaipur that morning at 06:37 local time, two and a half minutes before the birth. At sunrise the Sun is on the eastern horizon, so the rising degree and the Sun's degree should nearly coincide, with the Lagna slightly behind. They do: Lagna 29°33', Sun 29°58', 25 arc-minutes apart. The rules document lists this relationship among the anchors any future ephemeris must keep, and here it is reached by two separate pieces of code - the horizon geometry and the solar formula - agreeing.

Two flags are then added to each graha. Retrograde is decided by motion, not by a table: the planet's longitude is computed half a day before and half a day after, and if it moved backwards the flag is set. Jupiter is retrograde in this chart. The Sun and Moon are stated as never retrograde, Rahu and Ketu as always so by convention. Combust means within a fixed distance of the Sun, from 8° for a retrograde Venus to 17° for Mars - limits the engine cites to the Hindu astrology textbook behind its dignity tables. No graha in this chart is that close.

From the Moon to dated periods

The engine counts nakshatras from Ashvini with Ketu first, where Parashara counts from Krittika with the Sun first. It is the same wheel entered at a different spoke: Ashvini belongs to Ketu, Bharani to Venus, Krittika to the Sun, and the nine lords repeat three times round the 27. The Moon's nakshatra here is the seventh, Punarvasu, whose lord is Jupiter, so the timeline opens in Jupiter's 16-year period.

The Moon had covered 39.734 per cent of Punarvasu at birth, so 60.266 per cent of Jupiter's sixteen years remained: 9.64 years, ending on 3 November 2001. Parashara's Sl. 16 words the proportion as time - how long the Moon has stayed in the nakshatra against its whole stay - while the engine measures arc covered against the nakshatra's 13°20'. The two agree whenever the Moon moves evenly across the nakshatra, and the comparison in the next section shows the arc method reproducing a printed balance to within a day. After the partial first period the rest run at full length: Saturn to November 2020, Mercury to November 2037, then Ketu, Venus, Sun, Moon, Mars and Rahu, the last ending in 2105. A year is taken as 365.2425 days. That is this engine's choice; these slokas fix no length of year.

Sub-periods follow BPHS Ch. 51 Sl. 1-2. An antardasha's span is the product of the two lords' years over the 120-year total, the first antardasha belongs to the mahadasha lord, and the pratyantar level repeats the rule one step down. Mercury's own antardasha inside Mercury's mahadasha is 17 × 17 / 120 = 2.41 years, and the engine dates it 3 November 2020 to 2 April 2023. On 5 October 2026 this chart is in Mercury mahadasha, Venus antardasha (29 March 2024 to 28 January 2027) and Mercury pratyantar (5 July to 28 November 2026). One function divides every level, and it places each boundary from the cumulative years instead of adding spans end to end, so the nine parts of a period finish exactly where the period does. What is a Mahadasha and Antardasha and Pratyantar explain how the levels are read, and the Vimshottari calculator prints the whole table for any birth.

The engine against a chart printed in a source book

The page above says no test compares a computed position with a published ephemeris. That remains true. What a page can offer is one open comparison a reader can repeat. Advanced Predictive Astrology prints full computer-cast charts for its case studies, with the birth data, the ayanamsa used and every position to the arc-second. Its Example-11 (Vol. II, p. 53) gives 27 August 1910, 17:35, latitude 44°50' N, longitude 20°37' E, zone meridian 15° E. Entered into this engine with a one-hour offset, the result is this.

Three different things are visible. First, the Sun, the Moon, the Lagna and the ayanamsa agree with print to 15 arc-seconds or better. In two further charts from the same book - Example-3 (9 January 1922) and Example-10 (12 December 1905) - those four differ by 25 arc-seconds at most. Every sign and every nakshatra lord in Example-11 matches what the book prints, and the book's own commentary reads the Moon in the 5th house, Jupiter in the 9th and Saturn in the 4th, which is what counting whole signs from a Capricorn Lagna gives.

Second, four of the five planets agree with print to 42 arc-seconds or better, and Saturn is 12'46″ ahead. Until October 2026 all five sat about a degree and a quarter ahead. The Keplerian elements return a longitude measured from the equinox of 2000, while the ayanamsa subtracted from it is the one for the birth date, and the engine did not carry the planets from the one equinox to the other. The equinox moves at about 50.26 arc-seconds a year - the rate at which the engine's own ayanamsa grows - and over the 89.35 years between that birth and 2000 that comes to 1°14'50″, almost exactly the old error. The engine now adds the general precession (Meeus, Astronomical Algorithms, eq. 21.5) before subtracting the ayanamsa, and that degree and a quarter is gone. Saturn's remaining 13 arc-minutes the frame never explained; the elements are approximate, and that is what the approximation looks like for Saturn in 1910.

For a birth in 1990 the old offset was 8 arc-minutes and changed nothing read by sign or nakshatra unless a planet stood that close to a boundary; for a birth in 1950 it was 42 arc-minutes. What remains is the error of the elements themselves, which the project's check against the VSOP87 theory for 1950 to 2030 puts within 0.02 degree for Mercury, Venus and Mars and 0.16 degree for Jupiter and Saturn - Saturn's 13 arc-minutes above is that kind of error. The finer the division, the sooner it shows. A navamsa is 3°20' wide, but a shashtiamsa is half a degree (BPHS Ch. 6 Sl. 33-41), so for the slower planets the highest divisional charts are only as firm as this allows. The project's rules document says as much in its own words: precise divisional work is where a full ephemeris must become the source of truth.

Third, Rahu differs by 28 arc-minutes, and this one is a difference of definition. The engine's Rahu is the Moon's mean node. The engine also carries a diagnostic that computes how far the true node stood from the mean node at any moment, and for this instant it returns -0.4735°, or 28'25″. The printed value is the true node, to within 3 arc-seconds. No text in this project's library says which node a rule was written for, so the engine keeps one Rahu everywhere and uses the diagnostic in one place: the Kaal Sarp calculator reports whether its verdict would change on the true node.

How to read each kind of result

Positions

A position is given as sign, degree and arc-minute, nakshatra and pada. Read the degree before the label. A graha at 14° of a sign stays in that sign under every difference described above. A graha at 29°58', like the Sun in the Jaipur chart, is two arc-minutes from the next sign, and a different ayanamsa or a different program can name the other sign without either being broken. The rules document puts the Moon's own accuracy at 0.1 to 0.3 of a degree, so a Moon inside that distance of a nakshatra boundary should be read as undecided. Why apps disagree on your nakshatra works through that case.

Houses

Houses here are whole signs counted from the rising sign. The Lagna's degree is shown, but it is not a cusp. The book used above prints a second table beside each sign chart, a cuspal Bhava Chalit with twelve unequal cusps. A graha late in a sign can sit in one house by whole signs and in the next by cusps. Those are two conventions, not a right and a wrong answer, and this product uses the first throughout - in the birth chart, in every divisional chart such as the navamsa, and in each rule that mentions a house.

Dasha dates

Dasha dates are printed to the day and the arithmetic is exact, but the input is the Moon's position at the birth minute. In the Jaipur chart, where the opening period is Jupiter's sixteen years, a birth four minutes later moves the end of that period from 3 November 2001 to 17 October 2001, seventeen days. An hour later moves it to 14 February 2001. Every later boundary shifts by the same amount. A birth time known only to the nearest quarter-hour therefore leaves about a month of play either way at each boundary.

Strength

The strength score is a six-part total out of 20: position up to 10, aspect up to 3, direction up to 4, and natural, motional and temporal strength up to 1 each. Thirteen or more is labelled Strong, seven or less Weak, the rest Medium, and each label lists the reasons behind it. BPHS Ch. 27 names the same six components and works them in virupas, with a required minimum for each graha (Sl. 32-33). The score here is a compact scale built on that six-fold scheme, not Parashara's virupa computation, so compare it only with itself - one graha against another in the same chart - and not with a Shad Bala figure from elsewhere. The Shad Bala calculator shows all six parts.

A decision

A yes, wait or no is two numbers added together. The running mahadasha lord is classed good, neutral or challenging by its placement in the birth chart, worth 60, 30 or 0 points. The rule is close to BPHS Ch. 47 Sl. 5-6, which calls a dasha favourable when its lord is exalted or in its own rashi, and unfavourable when the lord is in the 6th, 8th or 12th house, debilitated or in an enemy's rashi. The second number is the strength score of the lord of the house the question belongs to - the 10th for career, 2nd for money, 7th for partnership, 6th for health, 11th for business, 1st otherwise - scaled to 40. A total of 65 or more reads YES, 35 to 64 WAIT, below 35 NO.

In the Jaipur chart the mahadasha lord is Mercury, which stands in Pisces, its debilitation rashi (BPHS Ch. 3 Sl. 49-50), so the first number is 0. The 10th house is Scorpio; its lord Mars scores 14.92 of 20, and 14.92 / 20 × 40 rounds to 30. The engine returns NO at 30 for a career question and prints both parts in its reasoning. Two things follow from the arithmetic. A good mahadasha contributes 60 on its own, so a YES then needs only 5 more from the house lord; a challenging one caps the total at 40, so it can never return YES. The score states the weight of those two factors. It is not a probability. Without a birth time there are no houses, and the engine returns WAIT at 0 with a request for the time instead of a guess. Decisions is the feature built on it.

Convergence

The public Convergence page asks a different question with less data. It takes only a birth date, so it cannot use houses or dignity. It classes the mahadasha lord by nature alone - Jupiter, Venus, Mercury and the Moon as favourable, Saturn, Rahu and Ketu as testing, the Sun and Mars as mixed - and the Personal Year by number: 1, 3, 5 and 8 favourable, 9 closing, the rest mixed. Each tone maps to one of three words, expand, build or hold, and when the two systems land on different words the more cautious one is reported and the split is named. For 14 March 1992, read on 5 October 2026, it classes Mercury mahadasha as expand and a twelve-month window dominated by Personal Year 1 as expand, and reports agreement.

So the same person can see expand on one page and NO at 30 on another. Neither is at fault. It is the distinction BPHS Ch. 47 Sl. 2 draws between a dasha's general effects, which come from the graha's nature, and its distinctive effects, which come from its placement. Convergence reads the first because a date is all it has. Decisions reads the second because it has the chart. Where both are available, the one that used the chart has used more of what is known about the birth. When astrology and numerology disagree covers the other kind of split.

Numerology numbers

Every numerology number is digit reduction on the calendar date or on letter values, and can be checked on paper. For 14 March 1992 the Mulank is the day alone: 1 + 4 = 5. The Bhagyank is every digit of the date, 1+4+3+1+9+9+2 = 29, then 2 + 9 = 11, kept as a master number and not reduced to 2. The Personal Year for 2026 is the day and month with the current year, 14 + 3 + 2026 = 2043, which reduces to 9; cycle numbers run 1 to 9, so masters are not kept there. Advanced Vedic Numerology works its Moolank and Bhagyank the same way. The books do differ elsewhere - which letter table a name uses, whether a personal year turns on 1 January or on the birthday - and this product keeps such methods under separate names instead of blending them. The numerology profile shows them side by side, and Mulank, Bhagyank and Life Path explains which is which.

Birth-time sensitivity, measured

The translator's four minutes a degree is an average. In the Jaipur chart the Lagna covers 1°24' in the four minutes before 06:40 and almost 41° in the two hours after it, because the signs rising at that hour and latitude rise quickly. The birth also happens to fall at the very end of a sign, which makes it a clean demonstration: the engine gives Aquarius 29°54' at 06:41 and Pisces 0°15' at 06:42.

Two minutes are enough here to move every graha one house back: the Moon from the 5th to the 4th, the Sun from the 1st to the 12th, Saturn from the 12th to the 11th, Jupiter from the 7th to the 6th. Each of those is read differently, and so is everything built on houses - yogas, house lords, the decision score. Most births are not this close to a boundary. The practical reading is to look at the Lagna's degree: within about a degree of either end of a sign, treat the house numbers as provisional until the time is confirmed. The Lagna calculator shows the degree, and the sub-lord and birth time takes the same idea down to finer divisions.

With no birth time at all the engine casts the chart for noon and leaves out the Lagna and houses. What remains still depends on the hour through the Moon, which covered 14.4° on that date.

A midnight birth and a noon birth share a Moon sign and nakshatra on this date. A late-evening birth has the Moon in the next sign and the next nakshatra, with a different mahadasha running today. The engine does not hide this behind the noon figure. The date-only Moon sign and nakshatra calculators compute both ends of the day and say whether they differ, and the Convergence page returned this note for the same date: 'a birth early on 1992-03-14 gives a Mercury Mahadasha and a birth late in the day gives Ketu'. With the Moon covering 13 to 15 degrees a day, nearly every date has a nakshatra boundary somewhere in its 24 hours, and a sign boundary falls inside a date a little under half the time. This is the ordinary case, not a rare one. Birth time unknown sets out what can and cannot be read without it.

What a language model is handed

When a model writes, its input is one block of computed values serialised as text, and the function that builds that block is the only route by which chart data reaches it. Beyond the question and messages the page above mentions, the block holds:

• the Sun, Moon and Lagna as the engine computed them - sign, degree, nakshatra, pada - and the names of the running mahadasha and antardasha lords;
• the Mulank, Bhagyank, Personal Year and Personal Day, with the signals from the other numerology systems;
• the names of the yogas and doshas found, each with a one-line effect, the strength label of every graha, and which is strongest and weakest;
• today's energy score and the house and sign of the transiting Moon;
• in chat, the patterns found in what you have logged;
• the app's own verdicts where they apply - the decision call with its score and reasoning, and which timing layers agree - so that the model words a conclusion the engine reached and does not form one of its own.

The instruction sent with it tells the model to ground every claim in that block and to name no placement, yoga, dosha or number that is not in it. In chat the model path is off unless the deployment switches it on; when it is on, a Groq-hosted model is tried first if its key is present, otherwise Anthropic's Claude. The Full Reading checks for the same two keys, allows the call 25 seconds by default, and on any failure composes the reading from the engine's own text. Whichever path ran is recorded with the reply as engine, groq or claude, and the visible note is written from that record. When the note names a model, take any number from the tile beside the prose, not from the sentence. AI Chat describes the feature, and the privacy page sets out how birth data is handled.

Limits that follow from the method

The page above lists what the engine does not do. These are consequences of what it does.

• Dates. The planetary elements are stated for 1800 to 2050, so an account refuses a birth date before 1 January 1800 with a message saying why, instead of returning positions from outside that range. The public calculators start at 1900, which is a product choice and not an engine limit.
• Clock changes. The zone offset is taken at the birth instant itself. Two clock times have no single answer: one repeated when clocks go back is read as the earlier of its two instants, and one skipped when clocks go forward is read with the offset in force before the skip. A birth in either hour is worth checking by hand.
• Interpretation is not measurement. The engine can show that a rule from a named chapter was applied to a correctly computed position. Whether the rule describes a life is not something arithmetic can test, and nothing here claims it does. The only check on that side is your own log: Your track record counts how often readings matched what you later recorded, from your entries alone.

Where to go next

Sources lists which calculation rests on which book. The calculators run each step on this page on its own, without an account. Kundali and Timing show the computed layer in use, and Navamsa, the second chart is the place to start on divisional charts and why their precision matters.

Questions readers ask

Does Astro Clock AI use Swiss Ephemeris?

No. Positions come from the formulas named on the page above, and the ayanamsa is a single Lahiri function with no alternative setting. The project's rules document names Swiss Ephemeris as the intended source of truth for a future replacement and lists the anchors any replacement must still satisfy, such as the Lagna at sunrise standing just behind the Sun.

Why does another program show a different degree for my Mars or Jupiter?

The five planets come from orbital elements, not a full ephemeris: checked against the VSOP87 theory for 1950 to 2030 they are within 0.02 degree for Mercury, Venus and Mars and 0.16 degree (about 10 arc-minutes) for Jupiter and Saturn. Until October 2026 they also carried a frame error of about 50 arc-seconds for every year from 2000 (8 arc-minutes for 1990, 42 for 1950), since removed; a saved chart from before then can differ by that much. A different ayanamsa adds its own shift to every point.

Why is my Rahu different elsewhere?

Most likely the other chart uses the true node. This engine uses the mean node everywhere. The engine's own diagnostic for the gap between them returned 28 arc-minutes for the 1910 chart above, and its code notes about 1.88° as the largest value it found between 1970 and 2015.

I do not know my birth time. What can still be relied on?

The numerology numbers, which use the calendar date only, and the signs of the slower grahas unless one stands at the very edge of a sign. The Moon's sign and nakshatra hold if the two ends of your birth date agree, which the date-only calculators report. The Lagna, the houses and the strength scores are not computed at all, and a decision returns WAIT. Start with the free birth chart and add the time when you have it.

How exact are the dasha dates?

The division is exact; the input is not. In the 1910 book chart the Moon differed from print by 15 arc-seconds and the opening balance by one day. The rules document's more cautious figure for the Moon, 0.1 to 0.3 of a degree, would correspond to weeks or months, so read a boundary as a band and not as a single day.

Can the AI change a number or a verdict?

It is given the numbers and verdicts and computes none of them, so the tiles, tables and scores are the same whether or not a model runs. What it writes about them is not checked by code. By default the chat is written by the rule-based engine and no model is called.

How can I check a result myself?

Work a numerology number by hand; it is digit sums. For a birth near sunrise, see that the Lagna and the Sun are close together. Compare one known date: the engine puts the Sun's entry into sidereal Aries in 2026 at 09:11 IST on 14 April, which can be set against any printed panchang. Or enter a chart from a book that prints its positions, as this page did, using the birth chart calculator.