Computed, classical, believed Which parts of this almanac are astronomy, which are a classical rule faithfully applied, and which are belief. The three are not the same kind of claim, and the app should not print them as though they were.

A panchangam page gives you a sunrise, a Rahu kaal and a verdict on the day, one under the other, in the same type. They look alike and they are not alike. The sunrise is a measurement and you can check it against the horizon. The Rahu kaal is arithmetic on a rule that no measurement produces: the day is cut into eight and the fifth part belongs to Rahu on a Thursday because the tradition says so. The verdict is a belief about what those two mean.

Everything this app computes is one of those three. This page says which is which.

These are quantities with a right answer that does not depend on anyone's tradition. They are tested against Swiss Ephemeris 2.10.03 reading the JPL DE431 files — the library Jagannatha Hora and most professional Indian software are built on — and against JPL Horizons and the published eclipse circumstances. The figures below are the worst disagreement measured, not a target.

Planetary longitudes

Apparent geocentric positions from VSOP87D for the planets and the ELP series for the Moon, with light-time, aberration and nutation applied. Cast as a sidereal chart from civil time — so the figure includes ΔT, the ayanamsa and the sidereal convention, not the theory alone — the worst disagreement with the Swiss Ephemeris over 546 instants spread across 1800–2200 is 0.82 arc-seconds, and the ascendant alone is within 0.027. Repeated across all nine ayanamsas the app offers, 10,206 separate positions, the worst is the same 0.82 and the mean 0.09. A sign is 30 degrees and a nakshatra pada is 3°20′, so this is finer than anything read off it by four orders of magnitude.

Sunrise and sunset

Within 1.7 seconds of the Swiss Ephemeris rise and set (upper limb, standard refraction). At latitudes where the Sun neither rises nor sets the app says so rather than inventing a time.

Tithi, nakshatra, yoga and karana boundaries

The instant each one ends, over 3,000 days at six places from Chennai to New York, 1900–2100, each on that place's own historical clock: worst 0.011 minutes, under a second. Which limb is running at sunrise matched the Swiss answer on all 12,000 checks.

Sankramana and the Moon's entries

The instant the Sun enters a sidereal sign: within 2.8 seconds. The Moon entering a nakshatra: within 0.53 seconds. These are what the solar calendars and the monthly birth star are built on.

Eclipse circumstances

Every solar and lunar eclipse whose greatest phase falls in 1900–2100 — 454 solar and 459 lunar — is found, with none missing and none spurious. Greatest eclipse is within 9.7 seconds (solar) and 8.2 seconds (lunar) of the Swiss instant, and the solar magnitude within 0.0027. A local observer's contacts land within about a minute of the published circumstances for 2017, 2019, 2023 and 2024.

Time itself

ΔT, the gap between the clock and the Earth's rotation, is within 0.2 seconds of the Swiss model over 1800–2200. The ayanamsas reproduce the Swiss definitions to within 0.001 arc-seconds, so the sidereal zero point is theirs rather than an approximation of theirs.

Two honest limits on this column

The convention is a choice, not a measurement. Which ayanamsa, which sunrise definition, mean node or true node — each of these changes the numbers and none of them is settled by observation. The app's defaults are named on the settings page and every one of them can be changed. What is computed is the answer given the convention.

The birth time is coarser than the mathematics. Four minutes of clock moves the lagna a degree. A Moon half a degree from a nakshatra boundary turns on whether a remembered time is right, and the pages say so when a position sits that close. Beyond about 1800–2200 the series are outside the span they were sized for and every chart carries a flag saying so.

These are not measurements of anything. They are arithmetic on top of definitions a tradition wrote down, and the app's claim about them is narrow and complete: it applies the rule as the sources state it. That is a claim about the app, not about the world. A Rahu kaal is exactly right as a Rahu kaal and says nothing whatever about the sky between half past one and three.

Each table below was checked against an independent reimplementation written from the classical texts and the published sources, not a translation of this app's own code, cell by cell over the whole table rather than on samples.

The ten Tamil poruthams

All 36 × 36 star-and-rasi pairs, 12,960 grade cells, against the rule text of the Tamil and Kerala sources. Where two sources give different rules — Dina by the published list of counts or by the tara remainder, Yoni by the seven enemy-animal pairs or by the 14 × 14 points, Rasi by the Tamil reading or by the older "the 7th or beyond" — the almanac tradition's reading is the default and the other is offered as an option rather than being hidden.

Ashta Koota, the 36 points

All 108 × 108 pada pairs: 93,312 koota cells and 11,664 totals. The points are a scoring scheme, not a measurement; the app adds them up correctly.

Rahu kaal, Yamaganda, Gulika and Abhijit

Sunrise to sunset cut into eight equal parts, each window taking the part fixed for the weekday by the standard almanac rhyme. Abhijit is the 8th of the day's fifteen muhurtas, and almanacs do not count it on a Wednesday. The eighths are exact. Which eighth belongs to Rahu is a rule, and the rule is all there is.

The muhurtham point tables

A day is graded by adding and subtracting points for the star, the tithi, the weekday, the lagna and the yogas, by the tables the Tamil almanacs and the Muhurta Chintamani print. Two almanacs weight them differently and will grade the same day differently; the app names the tables it uses.

The dashas

Vimshottari divides 120 years among nine lords in fixed proportions and starts the clock at the Moon's position within its nakshatra. Against an exact rational-arithmetic reimplementation over 2,000 charts, the maha, antar and pratyantar dates agree to 1.4 × 10⁻⁹ days, about a tenth of a millisecond. Ashtottari, Yogini, Kalachakra and Chara are checked the same way. The divisions are arithmetic; that 120 years is the span of a life, and that Saturn's nineteen of them mean what the texts say they mean, is not.

Ashtakavarga and the divisional charts

The eight bindu tables reproduce BPHS row by row, the benchmark totals come out 48, 49, 39, 54, 56, 52, 39 and the Sarvashtakavarga 337, and 5,000 random charts show no disagreement. The sixteen vargas match at every one of 7,200 division boundaries, tested a billionth of a degree either side of each, so a planet at exactly 3°20′ never lands in the wrong ninth.

KP's subdivisions

The zodiac cut into 249 subs in Vimshottari proportions: boundaries correct to 2.3 × 10⁻¹⁰ arc-seconds. That the sub lord decides the matter is the system's own claim.

The full statement of every rule, and the source named for each, is on the method page.

Everything in this third group the app computes exactly as its tradition defines it, and takes no position at all on whether it is true. This is not a disclaimer bolted on the end. It is the honest description of what a calculator can and cannot establish, and saying it plainly is the reason this page exists.

• That a dosha harms. Sevvai dosham is a real classical rule with a real definition, and the app applies it, its exceptions and its cancellations correctly. Whether Mars in the seventh house from the lagna makes a marriage go badly is not something the app knows, and nothing in it has been shown to.
• That a remedy works. The gem for a graha, the mantra counted a given number of times, the temple for an affliction — the app can tell you what the tradition prescribes and why. It cannot tell you that it will do anything.
• That an hour is lucky or dangerous. Nalla neram, Rahu kaal, the choghadiya, the yogas of the day. The windows are computed to the second; their meaning is the tradition's.
• That a score predicts a marriage. Twenty-six points out of thirty-six is a number produced by a stated procedure. It is not a measurement of two people, and two well-matched charts have ended badly and two badly matched ones well, many times, within every family reading this.
• That a nakshatra shapes a temperament. The readings this app prints for a star, a rasi or a dasha are the tradition's own descriptions, reported. They are not findings.

Two things follow from that, and both are meant kindly. If you keep these practices, the app will get the arithmetic right for you and never argue with you about them. If you do not, and you are here because a wedding or a death in the family has put you in a room where they matter, you can use this page to understand exactly what is being said without having to agree with it. Taking part out of love rather than belief is a coherent position, and the app is written so that it stays one.

It does not predict events

It will not tell you that you will get a job in March, that a child is coming, that a journey will fail, or when or how anyone will die. It reports what a period is called and what the texts say about it. It will not tell you anything frightening, because it is not in a position to know anything frightening.

No financial or medical advice

Nothing here is advice about money, health, a legal matter or a medical decision, personalised or otherwise. A "good hour" for business is a classical grading of an hour, not a view on your business. Choosing the hour of a caesarean is a real and common question, and it is a medical decision with an obstetrician in the room; the app will compute an hour you name and will not recommend one.

It does not choose between traditions

Where two almanacs honestly differ, both are shown with the rule behind each. Your wall calendar and this one can disagree about the day a month begins, and the reason is printable: the thirukanitha Sun of modern astronomy against the vakya Sun of the classical almanac, a sankranti before sunset against one after it, an amanta month against a purnimanta one, one ayanamsa against another. The app says which reckoning produced each date instead of pretending there is only one.

It does not claim the vakya almanac's own numbers

The vakya Sun as printed is not one formula but a family of them: houses round differently and correct by hand, so published month starts scatter by two or three hours around any one computation. This app's vakya reckoning reproduces 34 of the 36 Tamil month starts of 2025 to 2027, and the two it misses are named in the test file beside the almanac pages they were read from, so anyone can check both the agreements and the failures. A wider run over eleven years scored 114 of 132, but that table is not in the repository and so is not something you can check here. Where your almanac differs, its own reckoning is the one to keep.

It does not replace your astrologer

A chart on a screen is a set of positions and the readings that go with them. Reading one for a particular person, weighing what matters against what does not, is a judgement, and the people who make it well have spent a long time learning to. Where a position sits on a boundary, or a birth time is uncertain, the app says so and stops.

None of the figures above is a marketing number. Each comes from a test in this project that compares the engine with the Swiss Ephemeris or with an independent reimplementation of the classical rules, and each test prints the distribution it measured when it runs. The method page states every rule and names the sources it was taken from, including where two sources disagreed and which reading was chosen. The settings page holds every convention that changes a number — the ayanamsa, the house system, the nodes, the sunrise — so you can put the app on the same footing as the almanac you already keep and see whether the two then agree.

If a sentence on this page stops being true of the app, the sentence is the bug. Corrections are welcome, and the page carries no claim its own tests do not make.