Where a Bazi Compatibility Score Actually Comes From
Where a Bazi Compatibility Score Actually Comes From
You put in two birth dates and a tool gives you a number. Seventy one out of a hundred. Maybe a percentage with a coloured bar behind it. The obvious question is whether the number is any good, and the more useful question, the one almost nobody answers, is where the number came from in the first place.
Here is the plain answer before anything else. No classical Bazi text prescribes a percentage compatibility score. Not one. The percentage is a modern invention, in every language and in every product on the market, and that includes ours. What the tradition actually produced, when it matched two people at all, was a named category, and it read far less of the chart than you would expect.
That is not a reason to throw the tools away. It is a reason to know what you are holding.
How is Bazi compatibility calculated?
Differently by every tool, and most of them do not tell you.
We read the pages rather than the search results. Here is what the calculators say about their own methods, quoted from the pages themselves.
xuanseal publishes its weights. Its 0 to 100 score is composed as "The Day Master layer (40% of the score) compares the two day stems ... The Five Elements layer (30%) measures how well the two charts together cover Wood, Fire, Earth, Metal, and Water ... The Chinese Zodiac layer (30%) uses the year-branch relationships". Whatever you think of those weights, publishing them is a real act of good faith, and it is more than most tools do. It is also more than we publish about our own internal number, which is worth saying before criticising anyone.
openfate names its layers and declines to claim a pedigree. It calls its approach an "OpenFate proprietary compatibility method", which is the accurate word and almost nobody else in the segment uses it. Its disclaimer is the best we found anywhere: the score "is not a measured probability, a scientific forecast, or advice to marry or separate".
Others name the layers without the weights. ba-zi.ai lists day masters, five elements, branch interactions, ten gods, spouse palace and timing, and says plainly that "their relationship meaning is a traditional interpretation, not a scientifically validated assessment".
And some give a number with nothing behind it. bazi-lab describes its score as "an overall assessment of how well your energies naturally complement each other based on BaZi principles". Which principles, weighted how, from which source, is not stated.
Now the thing that is true of all of them, including the careful ones, and including us until this page. Not one cites a classical text for two chart matching. The disclaimers in this segment have got genuinely good. The provenance has not moved at all.
Did the classical tradition score couples?
Yes, and this is the part that gets left out.
The oldest widely transmitted two chart matching system is 吕才合婚法, attributed to 吕才, one of the eighteen scholars of the Tang court under Taizong. It works, in the words of a Chinese source describing it, 「以三元大运的男女宅命宫来看，并不是以属相来看」, from a 命宫 derived through the three cycles, and explicitly not from the animal sign.
What it returns is not a number. It returns one of eight named outcomes, and those eight are sorted into three grades:
- 上婚, the upper match: 生气, 延年 (also called 福德), 天医.
- 中婚, the middle match: 伏位 (归魂), 祸害 (绝体), 六煞 (游魂).
- 下婚, the lower match: 五鬼, 绝命.
Look at the shape of that rather than the content. A couple comes out of it holding a name, and that name sits in one of three bands. Nobody is told they are a 73. There is no scale, no percentage, and no implied precision that the method cannot support.
There is a second thing worth noticing, and it is less flattering. The whole system runs off the birth year. One pillar. Two characters out of eight. Everything the tradition's most durable matching method knows about a couple, it knows from the year.
So the honest summary of the classical position is not that the old masters were more sophisticated than the calculators. On this particular question they read less of the chart than a modern tool does. What they did do is refuse to dress the result up as a measurement.
What does 《三命通會》 actually say about matching two charts?
Nothing, and this matters because 三命通会 is the text most often waved at in this segment.
Our own classical reference is explicit about it. 《三命通會》 contains no dedicated two chart 合婚 chapter. Its relationship material sits inside a single chart, chiefly 卷七 論女命, which judges a woman's fate through her officer star and child star, and 論六親, which handles the six kinships. That is a reading of one person's eight characters, not a comparison of two people's sixteen.
Classical two chart matching lives elsewhere in the tradition: in 年命纳音 pairing, in 神煞 matching, in 生肖 clash and combination tables, in 《渊海子平》 and in the almanacs. Those are real lines of practice with real transmission. They are simply not the same book, and attributing a modern pair score to 三命通会 is a citation nobody can check.
We hold ourselves to that too. Our own compatibility engine is tagged in our reference as a modern construct, with a standing instruction written into the file: do not let reading copy claim that the classics score this couple at anything.
The thirty per cent problem
Here is the one criticism in this piece that is technical rather than a matter of taste, and it can be checked on both sides.
xuanseal weights the year branch at 30 per cent of its score. That is your animal sign, and the relationships between two animal signs are genuinely real mechanics, the six harmonies and the clashes and the harms that run through the twelve earthly branches.
Meanwhile, in Chinese, practitioners writing about the same question describe year branch matching as 「基本偏于粗糙型」, basically the crude sort, and as 「以偏概全型」, the sort that takes a part for the whole. The same source is unenthusiastic about 纳音 matching too, calling it 「非常鸡肋」, chicken ribs, the idiom for something not worth the trouble of eating.
So one segment builds thirty per cent of its verdict on the layer the other segment names as its crudest method. Both halves of that are published, in the open, in two languages, and they contradict each other. This is not an accusation of bad faith against anyone. It is a straightforward information asymmetry, and it is the third time this year that the Chinese language material has turned out to hold a caution the English language material never received.
The underlying point is one we have made before in a different place. Your animal sign is roughly one eighth of your chart, and it is not the eighth that stands for you. The character that does is your day master, which is why a compatibility method that leans on the year is leaning on the wrong pillar for most of what a couple wants to know.
What would a score out of 100 have to mean?
This is the question that dissolves most of the segment, and it has nothing to do with metaphysics.
A number out of 100 is a claim about a population. Seventy one only means something if you know what the other numbers were. Seventy one against what field, measured how, across how many pairs. Without that, a 71 and a 62 are two pieces of typography.
Almost no calculator answers this. What several of them do instead is publish a pass mark, which is a much bigger claim than a score. One widely circulated Chinese page states it flatly: 「60分比较适合，70分为较好，80分90分为很好，100分为最好！60分以上可结婚」. Sixty and above is fit to marry. And the pedigree offered for that threshold is 「先哲」 and 「先贤」 across 「数千年」, sages across thousands of years, with no text named at all.
That is a sentence telling a real person whether they may marry, resting on an authority that is not cited because it does not exist in that form. It is the most consequential thing we found in any language this run, and it is worse than silence, because silence at least leaves the reader their doubt.
The Russian language segment fails differently and more quietly. The calculators are there, the methods are not published, and access is priced. One sells its compatibility calculator at 4,800 roubles for a year. Another is candid that its free tier 「проверяет только Господина дня」, checks the day master only. Honest about the limit, silent on the method.
What The Eight does instead
We have a number. You never see it.
Inside the engine there is a 0 to 100 traction score, and the comment sitting next to it in the source says what it is for: it derives the tier and it runs the tests, and the interface shows the tier, never the number. What you are shown is one of five named tiers. Spark, Candlelight, Torchlight, Hearthfire, North Star.
The reason is written into the product itself, in the reading copy for a romantic bond: a score would flatten what an affinity is, so The Eight names the light between you instead.
We did not know, when that decision was made, that it lands in the same shape as 吕才合婚法 eleven centuries earlier. Named classes rather than a measurement. We are claiming no lineage for that. It is a coincidence of good sense, and we would rather report it as one than dress it up.
Here is the part that is a genuine claim, and the part we think the segment is missing.
The tiers are calibrated, and calibrated against something stated. The cuts are the measured population quintiles for each kind of bond, the twentieth, fortieth, sixtieth and eightieth percentiles, computed by running the engine over 6,000 random chart pairs for each bond kind with a fixed seed. The script that regenerates them is in the repository. So a tier has a definition you can hold in one sentence: it tells you which fifth of the field a pair sits in.
That is a much smaller claim than a percentage, and it is the whole point. It is a statement about where you sit among pairs. It is not a statement about what will happen to you, and no arrangement of two charts has ever been able to make that second statement.
Two further things the engine does that we have not seen elsewhere. The weighting changes with the kind of bond, because the same relation does not mean the same thing in two different relationships. A clash at the day branch strikes the spouse palace, which is severe for a marriage and much more survivable between business partners or friends, so romantic bonds weight the day chamber heaviest, working partnerships weight the month, and a parent and child bond counts the nourishing current between the two cores double. And the tiers are applied uniformly, never tuned to a particular couple.
If you already have a score
Three questions worth asking of any compatibility number, ours included.
What is it measured against? If the tool cannot tell you what population its scale represents, the number is decoration. Ask this one first, because it eliminates most of them.
What is it weighting, and how much of that is the year? If a third of the verdict rests on two animal signs, you are reading a zodiac table with extra steps.
What is it claiming to know? A description of friction is knowable from two charts. A verdict on whether you should marry is not, and no method in any of the sources we read, classical or modern, has ever been able to produce one. If a tool has quietly crossed from the first to the second, that is the moment to stop trusting it.
And then the thing the number was standing in for. What two charts genuinely show is texture, the specific friction between two people and where it comes from, which parts of it are structural and which are situational, and which of it you can actually live with. That is smaller than a percentage and considerably more useful, because it is the only version of the answer you can act on.
The door
If you want to see this rather than read about it, put two charts side by side. You will get a named tier, you will be told which fifth of the field it is, and you will be shown which chamber the friction is sitting in and where it comes from.
You will not be given a percentage, and now you know why.