Type a birth date into any Zi Wei Dou Shu (紫微斗数) calculator and a chart appears in under a second. What happened in that second is usually invisible, and that invisibility is the source of most of the confusion around these tools: people compare outputs without knowing that the two tools made different decisions on the way in.
This page takes the calculator apart. It covers what such a tool actually computes, where the choices are made, where calculation ends and interpretation begins, and why the same birth data can produce different charts in different tools. It ends with a short list of tests you can run on any calculator, including ours.
Whatever the interface looks like, a Zi Wei Dou Shu calculator performs five jobs in order. Four of them are mechanical. The fifth is where the tool stops being a calculator.
| # | Job | What it produces |
|---|---|---|
| 1 | Convert the input | A lunar birth date and a birth double-hour |
| 2 | Locate the Life and Body Palaces | The anchor position for the whole chart |
| 3 | Fix the Five Element Bureau | A number from 2 to 6 that star placement depends on |
| 4 | Place the stars | Fourteen main stars and an auxiliary layer |
| 5 | Produce text | A written reading of the chart |
Jobs 1 to 4 are deterministic: given the same decisions, the same chart comes out every time. Job 5 is not, and treating it as though it were is the root of a lot of mistrust — in both directions.
This is the least visible stage and the one that causes most disagreements between tools. Three decisions happen here before any star is placed.
Zi Wei Dou Shu is built on the lunar calendar, so the first translation is from a civil date to a lunar year, month and day. That translation is not mechanical in the way it sounds: lunar months begin at astronomical new moons, and different sources can differ by a day at a boundary. There is also the question of intercalary (leap) months — whether a birth in a leap month is counted in its own right or folded into the preceding month.
A birth time recorded on a clock is not the sun's time. Longitude within a time zone, historical clock changes, and — in some methods — the equation of time all shift the actual solar moment. Some calculators correct for this; some do not; some correct for longitude only. If a birth falls close to the edge of a double-hour, this single decision can move the chart. On our own charts, the correction is applied, and the correction itself is explained on the true solar time page.
The birth hour is expressed as one of twelve double-hours. Because the boundaries fall at odd clock times, a birth at, say, 10:50 sits in a different double-hour from a birth at 11:10 — and the correction above can push one across the line. This is also where the late Zi-hour question lives: births between 23:00 and midnight can be assigned to the current day or the next, and the two readings give different charts.
Test it yourself: enter the same birth with the clock shifted by twenty minutes and see whether the chart changes. If it does, the tool is applying a time correction of some kind. If it does not, it is using clock time as given. Neither answer is a defect — but you should know which one you are looking at.
Once the inputs are fixed, the chart is built by rules — the same rules a person would use with a pen. Nothing in this stage depends on interpretation or on language.
Each of those steps has a full write-up: the sequence as a whole on how a chart is calculated, the anchoring step on the Life Palace calculation, and the star rules on star placement formulas.
Here the tool stops being a calculator, and it is worth being precise about the seam, because it is where most of the industry's confusion lives.
A chart is a fixed object. A reading is language, and language is produced by a model of some kind — historically a person, now often a language model. The two are not the same kind of output, and they do not deserve the same level of trust:
| The chart | The reading | |
|---|---|---|
| Produced by | Rules | A model of language |
| Reproducible | Yes — same inputs, same chart | Not word for word |
| Checkable by hand | Yes | Not in the same way |
| Varies between tools | Only via input decisions | Freely — and this is expected |
Two things follow. First, a bad reading does not imply a bad chart, and a good reading does not prove the chart was right. Second — and this is the practical point — if a tool shows you only the prose and never the chart, you have no way to tell which of the five jobs above went wrong when something looks off. A reading is much easier to evaluate when the structure it is reading is visible next to it.
This is the design choice we made for our own charts: the calculation layer runs as rules and produces the structure, and the language layer works from that structure rather than inventing it. The two layers are deliberately kept separate, so that a difference of opinion about wording does not quietly become a difference of opinion about facts.
This is the question that brings most people to this page. If a chart is calculated, why do three tools give three results?
Because the disagreements happen in stage one and stage two — and they are disagreements about inputs and conventions, not about the craft of interpretation. In practice, they come from a short list:
Notice what is not on that list: nobody is disagreeing about what a star means. Every item is a decision made before interpretation begins. That is why comparing final readings is such a poor way to compare tools — if the charts differ, the readings are describing different objects, and comparing them tells you nothing.
The better habit is to compare the inputs and the anchor first: same birth data, same time-correction policy, same Life Palace branch? If those three line up and the charts still differ, you have found something genuinely worth investigating. If they do not line up, you have simply found two different conventions.
These tests are not a way to grade a tool's taste. They are a way to separate two things that get tangled up constantly: whether the chart was built the way it says it was, and whether you agree with the reading. The first is a factual question with a checkable answer. The second is a matter of judgement, and it gets much easier once the first is settled.
The fastest way to understand any calculator is to give it a birth date whose chart you have already worked out by hand. If you would rather start from the worked example used across this series — 12 September 1992, 10:20, female, Life Palace in Chen — you can follow every step from the five-step walkthrough and then compare it against any tool you like, ours included. If the two disagree, you will now know exactly where to look.
What does a Zi Wei Dou Shu calculator actually do? It performs five jobs in order: take the input, build the chart, then turn the chart into text. Four of those steps are mechanical.
Why do two calculators give different charts for the same data? Usually because of choices made on the way in: how the calendar is converted, how true solar time is handled, and which convention is used for edge cases.
How can I test a calculator? Compare its output on a chart you can verify by hand, and ask it which fixed rule produced a given statement. A rule-based tool can answer; a guess cannot.
Is the explanation part automated too? In a good tool, the chart is calculated deterministically and the text interprets that fixed result, so the reading can be traced back to specific placements.