How accurate is this almanac?

Every time on this site is computed, not copied from a table. Here is what it is computed from, what it was checked against, and — just as importantly — what it does not do.

What the positions come from

• A planetary ephemerisPositions come from the DE405 planetary ephemeris — the class of data used for spacecraft navigation. Not a table, not an approximation formula — the ephemeris itself.
• Lahiri ayanamsaThe Chitra Paksha ayanamsa, with nutation applied. This is the Indian government's official value, and the one every published Indian panchang uses, so a tithi here is the tithi your family's calendar names.
• Your own horizonSun and Moon are computed topocentrically — from where you stand on the surface, not from the centre of the Earth. The Moon's parallax is about a degree, which is two Moon-widths, and it is the whole reason an eclipse is total in one town and partial in the next.
• Drik GanitaTrue computed positions rather than the mean positions of the Surya Siddhanta. Drik Ganita is the approach and the ephemeris is the data — they are not rival methods, and a panchang saying “Drik Ganit” is not thereby less accurate.

The science behind the calculation

All five limbs of the panchangam come from two numbers: where the Sun is, and where the Moon is. Everything else is arithmetic on those two angles — which is why a single error in one of them shows up in several places at once, and why the fields that depend on the Moon alone agree most closely with other engines.

• TithiMoon − Sun, divided by 12°A lunar day is how far the Moon has pulled ahead of the Sun. Thirty steps of twelve degrees make one lunation, so the tithi number is that gap divided by twelve. Because the Moon's speed varies, a tithi runs anywhere from about 21½ to 26 hours — which is why it does not line up with a calendar day and why the app gives you the exact moment it ends.
• NakshatraMoon, divided by 13°20′The Moon's own position against the fixed stars, in 27 equal parts. It depends on one body rather than two, which is why it is the field that agrees most closely with other engines — two seconds, here.
• YogaMoon + Sun, divided by 13°20′The sum of the two longitudes, in the same 27 parts. Not a conjunction of anything visible; an arithmetic combination with its own named cycle.
• KaranaHalf a tithiSixty per lunation, so two per tithi. Both the current one and the next are shown, because a karana changes mid-day far more often than a tithi does.
• VaraSunrise to sunriseThe weekday changes at sunrise, not at midnight — so does Rahu Kaal, choghadiya and hora, all of which divide the day from sunrise to sunset and the night from sunset to the next sunrise. Get sunrise wrong and every one of them moves.

The corrections that do the work

Raw positions are not enough. Five corrections separate a number that looks right from one that survives comparison with a published almanac.

• Sidereal, not tropicalWestern positions are measured from the moving equinox; a panchangam is measured against the fixed stars. The difference is the ayanamsa — currently about 24°, a whole sign either side of wrong if you skip it. We apply Lahiri (Chitra Paksha), with nutation.
• Topocentric, not geocentricThe Moon is close enough that where you stand on the Earth changes where it appears by up to about a degree — two Moon-widths, or roughly four minutes of tithi. Positions are computed from your own point on the surface.
• Apparent, not geometricLight takes time to arrive and the Earth moves while it does. Aberration and light-time are applied, so the positions are where the bodies appear now rather than where they were.
• Delta-TThe Earth's rotation is not a perfect clock, so dynamical time and universal time drift apart — about 69 seconds today. It is applied before any comparison with a published table, which is the step that makes a correct eclipse time look correct.
• Solved, not steppedTransition moments — when a tithi ends, when the Sun crosses the horizon — are found by root-finding on the underlying function, not by sampling every minute and taking the nearest. That is the difference between a time good to the second and one good to the minute.

Checked against a reference almanac

The reference is a widely used online almanac that computes from the same planetary ephemeris rather than from printed tables. Compared field by field for Louisville, Kentucky on 14 September 2026:

• Varjyam0sThe start of the avoided window, to the second.
• Amritkalam−1sThe nectar window.
• Nakshatra end−2sThe Moon’s own position, effectively identical.
• Durmuhurtam+2sThe inauspicious slot.
• Moonrise+14s
• Tithi and karana end+49sA known offset — see below.
• Sunrise / sunset−69s / +65sA known offset — see below.

Two differences remain, and both are single causes rather than scattered error. Sunrise and sunset differ by about a minute — a rise/set convention difference that every eight-part division of the day inherits. Tithi and karana run 49 seconds late; since the nakshatra matches to two seconds, the Moon is effectively identical and the difference sits on the solar longitude.

Checked against the eclipse canon

Greatest-eclipse times against the standard published canon of solar eclipses for 2001–2100. The canon quotes Terrestrial Dynamical Time; the engine's own delta-T is applied before comparing.

• 2017 Aug 21+1sThe American total eclipse.
• 2024 Apr 08+5s
• 2026 Aug 12−4s
• 2027 Aug 02+1sThe long one over North Africa.

Checked against what people saw

Agreeing with another calculation is not the same as agreeing with the sky. These are computed local circumstances against what was actually observed and published.

• Hopkinsville, KY — 2017100%On the path. Computed magnitude 1.004; totality was observed.
• New York — 201759%Off the path. Computed magnitude 0.668; a deep partial was observed.
• New York — 202489.9%Computed maximum at 3:26 PM. Published figures: about 89.6% at about 3:25 PM.

Checked across thirty years

Festival rules are easy to fit to the year you are looking at. These were re-run against years they were never tuned on — 13 of 13 correct for 2026, 8 of 8 for 2027, and 8 of 8 across 2000, 2010, 2020 and 2030.

What it does not do

• Lagna shuddhiWhether the rising sign at your chosen hour is afflicted. Astrologers call this the deciding test for a muhurta, and it needs a chart cast for the exact moment. The rising sign is shown; it is not judged.
• Birth chart doshasNeither person's chart is read for Mangal dosha or anything like it.
• Regional variationFamilies and regions genuinely differ on which nakshatras a wedding admits. Each page names the list it used rather than presenting it as the only one.
• Holi, in some yearsHolika Dahan can land a day out. Pradosh is the right principle but not the complete rule, and the obvious fix — skipping days whose pradosh falls in Bhadra — was tried and is wrong.

Your data

The almanac runs on your device, so most of what you do here never reaches us at all. Birth details typed on the date-picking pages stay on the phone — the module that holds them has no network access — and there is no analytics SDK, no advertising SDK and no tracker anywhere in the app. Signing in uses an emailed code rather than a password, so there is no password to lose. The privacy page lists every field that leaves the device, where it goes, and what is not claimed.

Questions people ask

What the numbers rest on

• A planetary ephemerisDE405 — the class of planetary data used for spacecraft navigation. It is the basis of every position on this site.
• The published eclipse canonThe standard catalogue of solar eclipses for 2001–2100. Every eclipse peak here was compared against it.
• A reference almanacA widely used online panchangam computed from the same ephemeris, measured against field by field.

Want to see it working? Open the almanac — every page computes its times for the place you set, and says which place that is.