Wenxin Yiyan Bazi: Fixed Table of Hidden Stems in Earthly Branches

文心一言 is a general-purpose language model. When it is asked to calculate the eight characters of a birth time, it may produce a set of four pillars that looks complete, yet the labels attached to those pillars can still be wrong. The term 文心一言算八字 refers not to a single fixed operation, but to whatever procedure the model happens to perform in response to a bazi request. Because the model can imitate the surface form of a bazi chart, errors may occur in layers that are not immediately visible. One such layer is the hidden stems and the ten gods derived from them.

藏干与十神是第二层核对

The first layer of a bazi reading is the four pillars themselves: year, month, day, and hour, each written as one heavenly stem and one earthly branch, giving eight characters in total. A model can produce these eight characters correctly and still make mistakes in the second layer. That second layer consists of the hidden stems within each earthly branch and the ten gods assigned to the visible and hidden stems.

In this layer, each earthly branch contains a fixed set of hidden stems. Their presence and their order of weight do not depend on the birth date. The same branch always hides the same stems in the same relative order. This means the hidden-stem layer can be checked against an unchanging table. If a model assigns the wrong hidden stems to a branch, the error is not a matter of interpretation; it is a deviation from that fixed correspondence.

The ten gods are then derived from the relationship between the day master and each of the other stems, including both the visible heavenly stems and the hidden stems inside the branches. Because the hidden stems are fixed, and because the ten-god rules are also fixed, this entire second layer can be verified step by step.

地支藏干的固定对应

Each earthly branch has a fixed set of hidden stems. The contents are not altered by the year, month, day, or hour. For example, the branch 子 contains 癸. The branch 丑 contains 己，癸, and 辛, with 己 placed first as the principal qi, followed by 癸 and 辛. The branch 寅 contains 甲，丙, and 戊. The branch 卯 contains 乙. The branch 辰 contains 戊，乙, and 癸. The branch 巳 contains 丙，庚, and 戊. The branch 午 contains 丁 and 己. The branch 未 contains 己，丁, and 乙. The branch 申 contains 庚，壬, and 戊. The branch 酉 contains 辛. The branch 戌 contains 戊，辛, and 丁. The branch 亥 contains 壬 and 甲.

The order within each branch matters because the first hidden stem is treated as the principal qi of that branch. When a model lists hidden stems in the wrong order, it can change which stem is treated as the dominant one. That in turn can change the emphasis of a reading even if the ten gods themselves are otherwise labeled correctly.

Because this layer is fixed, the correct way to check it is not to ask the model whether its answer is right. The correct way is to compare its output with the fixed table. If the model says that 午 contains 丙 rather than 丁 and 己, the error is immediately visible. The same applies to any other branch.

十神是怎么从日主推出来的

The ten gods are produced by comparing the day master with another heavenly stem. The comparison uses two properties: the five-element relationship and whether the yin-yang polarity is the same or different.

The five possible element relationships are: same element, the day master produces the other element, the day master controls the other element, the other element controls the day master, and the other element produces the day master. These five relationships are then each divided into two by polarity, giving ten gods in total.

The ten gods are called 比肩，劫财，食神，伤官，偏财，正财，七杀，正官，偏印, and 正印.

The grouping rule is: if the other stem is the same element as the day master, the god is a 比劫 type, divided into 比肩 and 劫财. If the day master produces the other stem, the god is a 食伤 type, divided into 食神 and 伤官. If the day master controls the other stem, the god is a 财 type, divided into 偏财 and 正财. If the other stem controls the day master, the god is a 官杀 type, divided into 七杀 and 正官. If the other stem produces the day master, the god is a 印 type, divided into 偏印 and 正印.

Within each pair, the division depends on whether the yin-yang polarity of the two stems is the same or different. For example, when the other stem is the same element as the day master, same polarity gives 比肩 and different polarity gives 劫财. When the other stem is produced by the day master, same polarity gives 食神 and different polarity gives 伤官. When the other stem is controlled by the day master, same polarity gives 偏财 and different polarity gives 正财. When the other stem controls the day master, same polarity gives 七杀 and different polarity gives 正官. When the other stem produces the day master, same polarity gives 偏印 and different polarity gives 正印.

The day master itself is one of the ten heavenly stems. Each of the other nine stems can therefore be classified relative to it. The same nine relationships repeat for every stem that appears in the chart, including the hidden stems. What changes is the day master, not the rule.

这一层错了为什么看不出来

A bazi chart can be written with four pillars that are correct, while the ten-god labels beneath them are wrong. The page still looks complete because the structure is familiar: four columns, each with a stem, a branch, hidden stems, and a series of ten-god names. If the model writes plausible names in the expected positions, a reader may not notice that some of those names do not follow from the day master.

The reason this layer is hard to catch by inspection is that the ten gods are derived values. They are not part of the eight characters themselves. The eight characters can be copied correctly from a calendar calculation, while the derived layer below them is generated separately by the model. A model may handle the first layer well and the second layer poorly, because the second layer requires applying a fixed relational rule to many pairs of stems.

When the model simply prints a chart without showing the derivation, there is no visible sign of the mistake. The output still contains the right number of labels. The error is hidden in the relationship between one stem and another, which is not obvious unless the reader recomputes it.

要求模型把推导写出来

The practical way to check this layer is to ask the model to write out the derivation of each ten god. For every label, the model should state: what the day master is, what the other heavenly stem is, what the five-element relationship is, and whether the polarity is the same or different.

For example, if the day master is 甲 and the other stem is 辛, the derivation should say that 甲 is yang wood, 辛 is yin metal, metal controls wood, and the polarity is different, therefore the label is 正官. If the model only writes 正官 without these four steps, the reader cannot tell whether the label came from a correct derivation or from a guess.

When the derivation is written out, the reader is checking four separate logical steps rather than one final label. If the day master is misidentified, the error appears in the first step. If the element relationship is wrong, it appears in the second step. If the polarity is wrong, it appears in the third step. If the final label does not match the first three steps, it appears in the fourth step. This is far easier to verify than looking at a finished chart and trying to judge whether the whole thing is correct.

The instruction to write out the derivation also reduces the chance that the model will produce a fluent but unfounded answer. A model asked only for the chart may generate the chart in one pass and the ten gods in another, without connecting them. A model asked to derive each label must produce the intermediate steps, and those steps either agree with each other or they do not.

本站引擎按上述规则逐位标注十神。It labels both the visible heavenly stems and the hidden stems within the earthly branches. Because the rules are applied mechanically, the same input produces the same output each time. The labels do not vary between runs.

This layer does not change the eight characters of the four pillars. It changes only the labels attached beneath those eight characters. The day master, the other stems, and the branches remain the same. What varies is the set of ten gods shown under them. That is why a bazi chart can have correct pillars and incorrect annotations at the same time.

When the model writes out its derivation, the reader can check the four-step logic for each label. That is a different task from checking a conclusion. A conclusion can be accepted or rejected, but a four-step derivation can be examined at each point where it might fail. For 文心一言算八字, this second-layer check is the part that separates a chart that merely looks complete from one whose hidden stems and ten gods are actually consistent with the day master.