How a BaZi chart is calculated

A birth time on its own is ambiguous. 09:30 in London and 09:30 in Tokyo are eight or nine hours apart, depending on the season, and the same city can change its offset from one season or decade to the next. Thaat looks up the UTC offset that was in force at the birthplace on the birth date in the IANA time zone database, the record of the world’s clock changes that computers use, including past daylight saving time. Its records are reliable from 1970 on; for earlier births it covers many places, but not every local clock change.

Each row below is what the engine reads for a birth at noon on that date:

The BaZi year starts at Lichun (立春, also written Li Chun: Start of Spring), the moment the sun reaches 315° along the ecliptic. That falls on 3, 4 or 5 February. It is not 1 January, and it is not Chinese New Year, which follows the moon and moves between 21 January and 20 February. Someone born on 1 February belongs to the previous BaZi year, and so does their zodiac animal.

Lichun is an instant, not a date. Thaat compares the birth instant with the term instant, so the answer is the same in every time zone, even when the local date is still the day before.

Example: Lichun 1990 fell at 4 Feb 10:14 Beijing time, which was 21:14 on 3 February 1990 in New York. A baby born in New York at 21:00 that evening has the year pillar Ji Si (己巳), the Snake; one born at 21:30 has Geng Wu (庚午), the Horse, although the local date is the same.

The solar year is divided into 24 solar terms, one for every 15° the sun moves along the ecliptic. Every second term, a jie (節), starts a BaZi month, so the months follow the sun, not the calendar. Each month has a fixed branch, starting with Yin (寅) at Lichun. The month’s stem follows from the year’s stem.

The instants shift by about a day between years and fall at any hour, so a calculator that switches months on fixed dates gets everyone born near a boundary wrong. These are the twelve jie of the BaZi year that began in 2026, as Thaat’s engine computes them:

Days carry a stem-branch name in a cycle of 60 that runs on without regard to months or years. The day pillar is that name for the date of birth in local solar time. It changes at local midnight.

Schools differ about the hour from 23:00 to midnight. Thaat keeps it on the same date: someone born at 23:40 local solar time keeps that date’s day pillar, and only the hour pillar moves to the next day’s count. Other schools start the new day at 23:00.

The hour pillar is one of twelve two-hour blocks. Thaat measures it by the sun at the birthplace, not by the clock, as many BaZi practitioners do; not every school applies this correction. Clock time is the average solar time of the time zone’s central meridian, so two corrections are applied:

Daylight saving time drops out on its own, because the offset in step 1 already includes it. In wide time zones the gap between clock and sun can reach two hours. The hour’s stem then follows from the day’s stem.

Without a birth time Thaat does not guess an hour. The chart has three pillars, and the element balance and every other count are made from those three.

The engine’s calendar comes from lunar-typescript, an open-source library. We check its output, and our own time conversion around it, against sxtwl (寿星天文历), a separately written calendar library in C++ and Python. The test set has 2,233 birth moments on 1,389 dates from 1924 to 2030, chosen to hit the traps: Lichun on several days of early February, Chinese New Year before Lichun, every month’s start days, and the hours just before and after each jie instant from 1950 to 2030. All four pillars must match in every case.

What that test does not cover: every case is set in Beijing time on the 120° east meridian, where the longitude correction is zero, and every hour is even, the middle of a two-hour block, so the hour boundaries themselves are not tested there. Time zones, daylight saving time and solar time have their own tests: historical offsets for New York, London, Sydney, São Paulo, Tokyo and Bangkok; the same instant written in two different zones must give the same chart; and Lichun 1990 seen from New York.

Both libraries also build on the same family of astronomical algorithms, so agreement between them mainly proves that the pillar logic and the time handling are right. As a check from outside, our tests compute the sun’s position with a separate, simpler formula at every jie instant from 1924 to 2030, 1,284 in all. Each one lands within 0.01° of its term, about a quarter of an hour of the sun’s motion. And the engine’s Lichun for 2024, 2025 and 2026 is the minute the Hong Kong Observatory publishes: 4 Feb 2024 16:27, 3 Feb 2025 22:10 and 4 Feb 2026 04:02 Hong Kong time.

Correct pillars make a chart that matches the calendar. They say nothing about whether BaZi describes people well; that part is a traditional interpretive framework, and Thaat offers it for self-reflection, not as a prediction of events.

A chart is only as exact as the birth time you enter. If your recorded time may be rounded, or sits within a few minutes of an hour boundary, look at both neighbouring hour pillars.