Why Free BaZi Calculators Disagree

Put the same birth into three BaZi sites and you can get three different charts. Not always. But often enough that people notice, and the usual explanation — some of them are just wrong — is only about half right.

Some of the disagreement is error. A site with a bad table produces a bad chart, and no amount of interpretive skill downstream will recover it. But a substantial part is not error at all. It is two sites resolving the same genuine fork in opposite directions, both defensibly, and neither telling you they made a choice.

Learning to tell those apart is worth more than finding the one site that is "correct."

The chart is computed, not looked up

The eight characters are not stored in a table somewhere, waiting to be retrieved by birth date. They are the output of a chain of calculations that starts with a timestamp and a location and ends with four pairs of characters. Every link in that chain is a place where two implementations can part company.

Here is the chain, in the order the errors compound.

Which clock

Standard time is a political artefact. Time zones are drawn to give large regions one convenient clock, and the sun does not consult them.

Korea keeps its standard time on the 135th meridian. Seoul sits some seven and a half degrees west of it. That is about thirty minutes of real solar time — when the clock in Seoul reads noon, the sun does not reach the meridian for another half hour. Korea's own National Archives put it plainly: twelve thirty is the real noon.

This matters because hour branches are two-hour windows. A birth in the first half hour of a window slides into the previous one once the correction is applied, which means that for one Seoul birth in four, the hour pillar depends entirely on whether the site bothered to correct for longitude at all. The hour pillar is a quarter of the chart.

Country is not enough. Busan and Incheon are more than two degrees apart — nine minutes of solar time, and the two of them fall on opposite sides of any single national figure. A site applying one correction for the whole country is right in the middle and progressively wrong toward either edge.

This site corrects by city rather than country, from stored coordinates.

When the month changes

The month pillar does not change on the first of the calendar month. It changes at a solar term — one of the twelve 節 that divide the solar year at fixed points of the sun's apparent longitude. 立春, the beginning of spring, falls in early February and marks the boundary that matters most.

Two distinct errors live here, and they have different sizes.

The year pillar does not change at the lunar new year. It changes at 立春. These are usually within a few weeks of each other and occasionally on either side, so a site that uses the lunar new year produces a wrong year pillar for everyone born in the gap — every year, for a window of days to weeks. This is a straightforward error, not a school position.

A solar term is a moment, not a day. 立春 happens at a specific hour and minute, computed from the sun's position. A site working from a table with day-level granularity has to round, and for anyone born on the boundary day the rounding decides their month pillar — which decides the season the day master is read against, which is the single heaviest factor in whether that day master reads strong or weak.

This one is worth being concrete about, because this engine had it wrong. Its solar terms came from an approximate table, and when that table was replaced with an astronomical computation, six of the twelve terms moved by a full day. Every chart on those boundary days had carried the wrong month pillar until then. The output had never looked broken. It looked like a chart.

This site computes solar terms from solar longitude rather than reading them from a table, and marks a reading when the birth falls close enough to a boundary for the month pillar to be genuinely in doubt.

When the day changes

Here the disagreement stops being error.

The 子 hour runs from 23:00 to 01:00, which straddles midnight. So when does the day pillar advance — at midnight, or at 23:00 when 子 begins?

Both answers have long traditions behind them, and the difference is not academic. Take the chart from earlier posts, a birth on 28 February 1990. That day's pillar is Jia Zi 甲子. Move the birth to 23:30 that evening and the two conventions split: one keeps the day pillar at 甲子, the other advances it to Yi Chou 乙丑.

The day master is therefore 甲 or 乙. Yang Wood or Yin Wood — the same element, opposite polarity.

That sounds like a small difference and it is not. Polarity is what distinguishes the Ten Gods from each other. 庚 Metal standing beside 甲 is 칠살, Seven Killings — pressure, confrontation, the thing that sharpens or grinds. The same 庚 standing beside 乙 is 정관, Direct Officer — structure, position, legitimate authority. Same character in the same place, and a different relationship to the person the chart is about.

Every stem in the chart flips this way. Two readings built on the two conventions are not slightly different. They are describing different people.

Neither site is wrong. They inherited different answers to a question the tradition never settled.

What counts as present

Once the four pillars are fixed, the question becomes what is in them.

The visible eight characters are not the whole chart. Each of the twelve branches contains one to three hidden stems — 寅 holds 甲, 丙 and 戊; 子 holds 壬 and 癸 — and the hidden stems are where most of the actual substance sits. A branch that looks like the wrong element on its face may hold a root of the right one.

Sites differ on all of it. Some count only what is visible. Some count hidden stems but treat each as one unit. Some weight each hidden stem by the portion of the month it governs, so a primary hidden stem counts for several times what a residual one does.

These three approaches produce three different pictures of the same chart, and the difference is not marginal. A chart that looks Wood-dominant when you count characters can be Fire-dominant once the hidden stems are weighted.

This site counts hidden stems, weighted by the share of the month each governs.

How the tests are weighted

Then comes the judgement that most affects what a reading says: is the day master strong or weak.

Classical practice looks at whether the day master holds the season, whether it holds its own seat, whether it has roots elsewhere, and what the rest of the chart is doing. Everyone agrees on the four tests. Nobody agrees on what they are worth relative to each other.

Weight the season heavily and a day master born in its own month reads strong even in hostile company. Weight the surrounding characters heavily and the same chart reads weak. Both are defensible readings of the same tradition, and they produce opposite verdicts — which then inverts the interpretation of every other character in the chart.

There is a second choice hiding behind the first: where the lines fall. Strong, weak, and the band between them are cut at particular thresholds, and those thresholds are chosen, not derived. Two sites using identical weights can still disagree about a chart sitting near a boundary, purely because one drew its line a little further out.

This site weights the season most heavily by a wide margin, and the day master's own seat second.

What feeds the verdict

The last fork is the one readers almost never hear about.

Branches interact — they combine, they clash, they pull each other toward elements neither started as. A chart where two branches form a half-combination toward Fire is arguably not the same chart as one where they sit inertly side by side.

The question is whether those interactions feed the strength verdict, or sit alongside it as a separate layer of the reading. Some implementations fold them in and let a clash weaken a root. Others assess strength from season, seat and roots alone, and report the combinations and clashes separately.

On this site, combinations and clashes are reported as part of the chart's structure and do not alter the strength verdict itself.

Which disagreements are errors

Sorting the list:

The top half is fixable. A site in that group is not offering you an alternative tradition; it has a bug, and the chart it hands you is wrong in a way its own authors would concede if shown the arithmetic.

The bottom half is not fixable, because there is nothing to fix. Two careful implementations will keep disagreeing about charts near those boundaries for as long as the schools do.

What to ask a calculator

You cannot audit a site's source. You can look for four things that are visible from outside.

Does it ask for a city, or only a country? A site that asks for a country has decided longitude does not matter, which is a statement about its accuracy that it is making without saying so.

Does it tell you when you are near a boundary? Births close to a solar term, or in the hour before midnight, are the cases where methods diverge. A site that knows this can say so. A site that reports every chart with the same flat confidence either does not know or is not telling.

Does it show the reasoning behind the strength verdict, or only the label? "Strong day master" is an assertion. "Holds the season, does not hold its seat, one root elsewhere" is an argument, and you can check an argument against another site's.

Can you tell which of its answers were computed and which were chosen? This is the real test. Every site makes choices. The useful ones let you see where.

A chart that agrees with three other sites is not thereby correct — they may share a common bug, and popular libraries propagate the same rounding errors across dozens of sites. A chart that disagrees is not thereby wrong. The question is always whether the site can show you why.