Why Two BaZi Calculators Disagree: The Six Conventions Behind Different Charts

01The six conventions, in the order they usually bite

First, the day rollover: a birth at 23:30 is either the current day pillar (midnight convention) or the next (23:00 convention). Second, the year: 立春 versus Lunar New Year, which differ for up to about two weeks each year. Third, the month: the exact sectional-term instant versus the calendar date — indistinguishable except on twelve days a year. Fourth, true solar time: with or without longitude and equation-of-time correction, which shifts the hour pillar near boundaries and can shift the day pillar near midnight. Fifth, historical time zones: whether 1986–1991 daylight time and older offsets are reconstructed. Sixth, the ephemeris: two astronomical implementations can disagree about a term instant by a minute or two.

Each convention is defensible. None is universal. The problem is silence: software that does not state its choices forces users to compare charts without a common reference.

02Worked example: five births, three engines, one table

The matrix below uses five probe births chosen to isolate one convention each. Columns show the four pillars under a convention-aware engine (this family's baseline), lunar-javascript 1.7.7, and tyme4ts 1.5.4, all run on 31 October 2026 and archived with the page.

23:30 on 1 December 1990: midnight-rollover engines return 庚午 丁亥 庚子 戊子; 23:00-rollover engines return 辛丑 for the day pillar. 10:20 on 4 February 1990: with true solar time the year and month pillars are 己巳 丁丑; with wall clock they are 庚午 戊寅, because 立春 fell at 10:14 and the correction is about −14 minutes. 09:14 on 5 April 1990: true solar time keeps the month at 己卯; wall clock has already crossed the 09:12:56 term instant and shows 庚辰. 17:30 on 1 December 1990 in Ürümqi: true solar time gives 甲申; wall clock gives 乙酉. 03:30 on 15 April 1990: DST-aware reconstruction gives 丁丑; an uncorrected clock gives 戊寅.

Every difference in the table is explained by one of the six conventions. No engine is wrong; each is explicit or silent about a different set.

03Reconciling two charts in four steps

Step one: compare day pillars. If they differ, check whether the birth time is between 23:00 and 23:59 and whether one tool uses the 23:00 rollover.

Step two: compare year and month pillars. If the day matches but the year or month differs, check whether the birth falls within two weeks of 立春 or on a sectional-term day, and whether one tool applies true solar time.

Step three: compare hour pillars. If the first three pillars match, check birthplace longitude and whether the birth year falls in a daylight-saving window.

Step four: if a difference remains only at a term instant, compare the ephemeris: two implementations can legitimately differ by a minute, and a birth within that minute has two candidate month pillars. Anything outside these four categories is worth reporting as a possible bug.

04Limits: what a convention comparison is not

A convention is not a quality ranking. A tool that uses wall-clock time is not worse at arithmetic; it is answering a different question. This page describes behaviour and states our own choices; it does not claim that one set of conventions produces better readings.

The comparison also cannot validate an uncertain birth record. If the time is unknown to the hour, no convention can recover it, and the chart should be treated as ambiguous.

Finally, a convention baseline is versioned. Engine versions change, ephemerides improve, and historical time-zone data is revised; every comparison must carry a date, as this one does.

05Checklist

• The birth time's position relative to 23:00 was checked first
• The birth date's position relative to 立春 was checked next
• The birth date's position relative to the nearest sectional term instant was checked
• The birthplace longitude and historical time zone were checked for the hour pillar
• Any remaining difference was attributed to the ephemeris, with versions and dates recorded