Verification

How accurate is BaZi, really?

The chart is arithmetic and can be tested to the minute. The reading cannot be tested that way. Here is exactly where the line falls — and why most answers to this question dodge it.

Updated November 2026

One word, two questions

Search for this and you will find two kinds of answer. One says the system is precise mathematics, four pillars, sixty-year cycles, exact to the minute. The other says it is a personality test with a thousand years of marketing behind it. Both are usually written in good faith, and both are usually answering a different question from the one being asked.

There are two questions here, and they have almost nothing to do with each other:

• Did the calculator produce the right chart? Given a birth moment and a birthplace, the four pillars have one correct answer. This is calendar arithmetic. It can be checked in public, against other implementations, case by case.
• Does that chart describe a person? This is the interpretation. It cannot be checked the same way, and pretending otherwise is where this subject normally goes wrong.

Merging the two is convenient for anyone selling a reading, because a correct calculation lends its credibility to the part that was never arithmetic. So we will take them separately: the first with numbers, the second without pretending.

Part one: the chart is arithmetic, and it can be audited

The rules for the four pillars are not ambiguous. The year and month come from solar terms computed astronomically, the day from a continuous sixty-day count, the hour from the day stem. Implemented correctly, they give the same answer every time.

The mistakes do not happen in the rules. They happen in everything before them: which timezone offset was in force on that date, whether daylight saving was running, how far the birthplace sits from the meridian its clock is set to, and which hour the tradition treats as the start of the day. Each of those is a place where a calculator can be quietly wrong while looking completely confident.

Concrete sizes, all taken from the thirty reference charts we publish:

• Clock time is not solar time. Across those thirty charts the gap between the clock on the wall and the sun at the birthplace has a median of 16 minutes. The largest is Madrid, at 85 minutes — Spain keeps Central European Time while sitting at 3.7° west.
• What a few minutes costs. A shift of s minutes moves a uniformly-timed birth across a two-hour pillar boundary with probability s/120. At the mean shift of 20 minutes, that is about 17% of charts landing in the wrong hour pillar, and 1.4% in the wrong day pillar, for any calculator that skips the correction. Not a rounding error: a different reading.
• The day starts at 23:00. The day pillar advances at 23:00 local true solar time, not at midnight. Someone born at 23:30 carries tomorrow’s day pillar. That is roughly one birth in twenty-four.
• The year starts at 立春, not 1 January. Anyone born between New Year and the first week of February carries the previous year’s animal — almost one birth in ten. This is the single most common reason people say their Chinese zodiac sign is wrong.

None of those are exotic edge cases. They are ordinary birthdays, and each one moves the answer by a whole pillar.

So the calculation can be tested, and we test it. Thirty birth moments spanning 1949 to 2026, six continents, both hemispheres, daylight saving on and off, the 23:00 late-Rat hour, a leap day, a daylight-saving gap and a daylight-saving overlap. All four pillars agree with an independent implementation that shares no code with ours — Julian Day Number arithmetic for the day, Meeus’ apparent solar longitude for the solar terms — 30 out of 30. Element weights reproduce within 0.6 percentage points, 30 out of 30.

That is the entire accuracy claim for the calculation. It is narrow, and it is falsifiable, which is the point. You can run it yourself at Algorithm & accuracy, or download the data and the harness from the benchmark page and run it against our numbers with our code nowhere in the loop.

Part two: the reading is not measurable the same way

Here is where most sites on this subject get vague, so let us be plain. We are not aware of peer-reviewed evidence that a BaZi chart describes personality, behaviour or life outcomes better than chance. We are not going to pretend there is some, and we are not going to cite a study we have not read.

That is not the same as saying the system is worthless. It is a thousand-year-old vocabulary for talking about temperament and balance, and it is a genuinely interesting one: it is structured, it is internally consistent, it has a long commentarial tradition, and it is unusually specific about where a person is unbalanced rather than merely what they are like. Used as a set of questions to think with, it does work that a magazine horoscope cannot, because it is not trying to flatter you.

But its statements are, as usually phrased, unfalsifiable. “Strong Wood tends to over-extend” cannot be wrong, because any Wood-dominant person who does not over-extend can be described as Wood that has found its Metal. That is a feature of the vocabulary and a problem for the word “accurate”.

The element percentages make the same point visible. “Fire 25.3%” looks like a measurement. It is a convention: we weight stems at 1.0, branches at 0.8 and each hidden stem at 0.3, then normalise to 100%. Another school weights them differently and gets different numbers from the identical chart. Neither is wrong, because the coefficients are a modelling choice, not a fact about the person. We publish ours so that at least the number is reproducible.

So the honest summary: the chart can be right or wrong, the weights can be conventional, and the reading is a reading.

Why a reading feels accurate even when it is not evidence

If you have had a chart read and thought “that is unnervingly close”, you should know about the Forer effect. In 1949 Bertram Forer gave every student in a class the same generic personality sketch, assembled from an astrology column, and asked them to rate how well it described them personally. The class averaged 4.26 out of 5. The experiment has been repeated many times since, with the same result.

BaZi makes this stronger than a magazine horoscope does, for three reasons.

The chart is real arithmetic. A reading that opens with a verifiable calculation borrows credibility for the part that is not verifiable. The correct pillars do not validate the paragraph underneath them, but they feel as if they do.

You supplied the input. You know your own history, so almost any half-specific sentence can be fitted to something that happened. The reader does not need to be right; you will do the matching.

Memory is asymmetric. The sentences that landed stay with you. The ones that missed are forgotten within the hour, and never enter the assessment of whether it was accurate.

We cannot take any of that out of a reader’s head. What we can do is refuse to exploit it. There are no percentages in a WuWay report, no “98% match”, no dates, no health claims, and no sentences built to be unfalsifiable. Every element gets its failure mode written down next to its strength, because a reading that only flatters you is not doing the job.

What we claim, and what we do not

The whole position, in one table.

How accurate does your birth time need to be?

The hour pillar is a two-hour window, so what matters is not precision but distance from the boundary. If your recorded time could be wrong by X minutes, the chance it actually belongs on the other side of a boundary is about X/120:

• correct to 5 minutes — roughly a 4% chance the hour pillar is the other one
• correct to 15 minutes — roughly 13%
• correct to half an hour — roughly 25%

The practical rule: more than thirty minutes from the nearest boundary, and the hour pillar is safe. Within ten minutes of one, the honest thing is to read the chart both ways — the day master does not move, only the hour pillar — and see which version survives contact with you.

Where the time came from matters as much as its precision. A birth certificate or a hospital record with an hour on it is usually reliable. “Early morning, before the sun was up” is not a time, it is a memory, and it can be wrong by two hours. If there is genuinely no time at all, drop the hour pillar: the day master comes from the day pillar, so the core of the reading survives without it. What is lost when the birth time is missing.

How to check us instead of believing us

Everything above is a claim, so here is where each one can be broken:

• Algorithm & accuracy — runs the real engine in your browser against all thirty cases, side by side with an independent implementation, and re-runs on every visit.
• Reference charts — the same thirty in plain HTML, with the true solar time and the element weights, no JavaScript required.
• Boundary cases — five worked examples with real output: the day rolling at 23:00, true solar time shifting a birthday across a pillar, and 立春 flipping the year in New York.
• Benchmark — the data, a reference engine and a harness, MIT code and CC BY data, so the check does not depend on our site being up.

If you find a case where we disagree with a printed reference or with another implementation, send it to us through contact. A disagreement that can be reproduced is the only kind of accuracy question that can actually be settled, and we would rather know than not.

Questions people actually ask

How accurate is BaZi?

It depends which part you mean. The chart is calendar arithmetic with one correct answer per birth moment, and it can be tested — ours agrees with an independent implementation on all thirty published cases. The interpretation is not measurable in the same way, and there is no published evidence that it predicts personality or events better than chance. A site quoting a single accuracy percentage is describing neither of those.

Is BaZi scientifically proven?

No. The astronomy behind the chart is exact — timezone offsets, solar terms, the sexagenary cycle are all well defined — but that is the input, not the reading. We are not aware of peer-reviewed evidence that a chart describes a person better than chance, and we do not claim otherwise. WuWay is a reflective tool, not a predictive one.

Why do different BaZi calculators give different results?

Almost always conventions, not arithmetic. The chart depends on the timezone offset in force on that date, whether daylight saving was running, the birthplace’s longitude correction, and whether the day rolls over at 23:00 or at midnight. Two calculators that disagree on any one of those give different pillars from identical input. The element weighting is a separate choice again.

How accurate does my birth time need to be?

Distance from a boundary matters more than precision. If your recorded time could be wrong by X minutes, the chance it belongs on the other side of a boundary is about X/120: 4% for five minutes, 13% for fifteen, 25% for half an hour. Beyond thirty minutes from a boundary, the hour pillar is safe.

Does BaZi predict the future?

We do not, and we do not read charts that way. The system describes balance and tendency — where a chart is strong, where it is stretched thin — not events. Nothing in a WuWay report gives a date, a prediction or an outcome, and there are no health, medical or financial claims.

Why does a reading feel so accurate?

Partly the Forer effect: in 1949 a class of students rated the same generic sketch 4.26 out of 5 for personal accuracy, a result reproduced many times since. BaZi adds two things that strengthen it — a real calculation in front of the unverifiable sentence, and an input you supplied yourself, which makes any half-specific line easy to fit. Feeling accurately described is not evidence of accuracy.

What is the most accurate BaZi calculator?

Nobody can answer that from outside, because accuracy depends on timezone history and coordinate data you cannot audit. What you can check is whether a calculator shows its work — birthplace, the correction to true solar time, the pillars as characters, and published test cases. That is a different question from accuracy, and it is the one that can actually be answered: six checks you can run on any calculator.