If you were born in late January or the first half of February, there is a real chance the animal on your paperwork is not your animal in the traditional count. Not because of anything mysterious — because two different calendars are used to answer the question "when does the zodiac year change", and they disagree for a window of up to three weeks every single year.
We needed the exact answer for every year between 1924 and 2043, couldn't find it as clean data, computed it, and open-sourced it:
github.com/hengwenpu/chinese-zodiac-year-boundary — MIT licensed, two CSV files, one zero-dependency Node.js lookup helper.
Two calendars, one zodiac
| Convention | Year switches at | Used by |
|---|---|---|
| Lichun (立春) — the sun reaches celestial longitude 315° | ~Feb 3–5, varying by year | The ganzhi (干支) calendar, BaZi / Four Pillars, traditional almanacs |
| Lunar New Year (正月初一) | The lunar calendar, Jan 21–Feb 20 | The civil lunar calendar; the Chinese national standard GB/T 33661-2017 defines the zodiac year of the lunar calendar this way |
Both are internally consistent. Both are "correct". They just answer different questions — and anyone born between the two dates gets a different animal depending on which one you ask. Roughly one person in twelve is affected.
1996 is a vivid example: Lichun fell on February 4, Lunar New Year on February 19. Someone born on February 18, 1996 is a Rat in the ganzhi count and would be counted an Ox (or Pig, depending on the year) by a calculator that keys off Lunar New Year. Fifteen days of disagreement.
Why this is an engineering problem
Most zodiac calculators on the internet use one of three shortcuts: January 1 (wrong), a fixed "February 4" (wrong — the date wobbles between Feb 3 and 5), or Lunar New Year (valid for the civil calendar, wrong for anything ganzhi-related).
For casual use, that's a shrug. But if you compute BaZi / Four Pillars charts, the year pillar comes from the Lichun boundary — get the boundary wrong and the year pillar is wrong, which propagates into every derived pillar. There is no fuzzy match; a date is on one side of the line or the other.
The astronomy, in one paragraph
Lichun is the moment the sun's apparent geocentric longitude crosses 315°. Because Earth's orbit is elliptical and its speed varies (Kepler's second law), the instant drifts by minutes year to year, and the date lands on February 3, 4, or 5 depending on leap-year rhythms and where you are on the orbit. In 2026 the instant is February 4, 04:01 Beijing time. In 2024 it was February 4, 16:27. Same day both years, but not the same day every year — and not the same day in every decade.
What's in the repo
data/zodiac-years-1924-2043.csv — one row per ganzhi year, two full 60-year cycles, contiguous by construction:
ganzhi_year,ganzhi,ganzhi_cn,animal,animal_cn,starts_lichun,ends_next_lichun 1924,Jia-Zi,甲子,Rat,鼠,1924-02-05,1925-02-04 1925,Yi-Chou,乙丑,Ox,牛,1925-02-04,1926-02-04 2026,Bing-Wu,丙午,Horse,马,2026-02-04,2027-02-04 data/lichun-1900-2100.csv — every Lichun instant from 1900 to 2100, with an approximate Beijing-time clock reading and a near_midnight flag for the only cases where minute-level accuracy could ever change the date.
index.js — a zero-dependency lookup:
const zodiac = require('chinese-zodiac-year-boundary'); zodiac.animalFor('2026-02-03'); // Snake — still the Snake year zodiac.animalFor('2026-02-05'); // Horse — the Horse year began at Lichun, Feb 4 zodiac.animalFor('1996-02-18'); // Rat — even though Lunar New Year 1996 fell on Feb 19 zodiac.yearInfo(1984); // { ganzhi: 'Jia-Zi', animal: 'Rat', startsOn: '1984-02-04', ... } A birth on the Lichun date itself counts toward the new year; if you need to split that day, the approximate instant is in the Lichun CSV. Dates outside 1924–2043 return null rather than a guess.
Accuracy, stated honestly
- Day-level dates are the claim of this dataset. They were computed from the sun's apparent longitude and spot-checked against published values from the Purple Mountain Observatory (紫金山天文台) — including 2024-02-04 16:27 and 2026-02-04 04:01 — with no date-level discrepancies, including the two years in the range closest to midnight (2046, 2050).
- The minute column is approximate and may differ from the observatory's official instant by up to about 7 minutes. Do not build anything on the minutes. That's exactly why
near_midnightexists — it flags the only rows where that gap could ever touch the date, so you know where to double-check. - Corrections with an authoritative source are welcome as pull requests.
Which convention should you use?
Whichever one matches your product's claim:
- Serving the civil/lunar calendar view (what most people mean by "Chinese New Year")? Use the Lunar New Year boundary. GB/T 33661-2017 legislates it for the civil calendar, and it is the right answer for that calendar.
- Serving ganzhi, BaZi, almanac, or solar-term contexts? Use Lichun. That's the boundary those systems actually run on.
We publish on the Lichun convention because that's what our own charts are computed on. The point of the dataset is not that one convention wins — it's that the two are different, and the difference is knowable in advance, for every year, down to the day.
Why we built it
We're Xuanji (璇玑) — we compute BaZi charts from true solar time and publish the boundaries our own engine runs on, because a chart you can't audit isn't worth much. We also make physical zodiac cards with printed reports (yushucarp.com) — that's the commercial interest, disclosed. The free chart tool requires no signup: tool.yushucarp.com.
The dataset is MIT — use it, ship it, cite it. A link back is appreciated but not required. If you find a date that disagrees with an authoritative source, open a PR; that's the whole point of publishing it.
