Korean Saju · free to start
Your Four Pillars,
calculated from your birth date.
Explore your Four Pillars, five-element balance, natural strengths, emotional rhythm, and practical growth direction.
How this chart is calculated, and what the paid report adds
What each field in the form becomes
Set the calendar field to lunar and your date, leap month included, is converted to its solar date first. From there the four pillars are built by two different routes. The year and month pillars turn over at solar terms, so the moment you enter is first moved to the solar-term reference clock and read there. The day and hour pillars are read from the civil clock of the place you were born. Tick “fine-tune birth time” and only those two are rebuilt from a clock shifted by the representative longitude of the timezone you picked plus the equation of time; leave it unticked and the civil clock stands unchanged. Having two references inside one chart is a rule we chose, and it is written here rather than left implicit.
The stem of the day pillar is your Day Master. The whole reading starts from that one character: its element and its yin or yang polarity decide the sentences the personality section is built on. Every other stem, and every stem hidden inside a branch, is then read as a relationship to it — a Ten God. The branch characters themselves are given no Ten God of their own; a branch is read through the stems it holds.
If the birth time is unknown, no hour pillar is built at all — the chart has three pillars, and the rest of the calculation proceeds at midday. Gender is used only for the ten-year luck pillars, and it settles two things there rather than one: which way they run, and the age the first pillar begins, since both fall out of the same call. The eight characters of the chart itself do not move with it. Leave it unset and the chart carries no luck cycle, and the paid report is instructed to say nothing about one rather than invent a decade or an age.
Say the time is exact and the engine rebuilds the pillars five minutes either side; say approximate and it uses sixty. Either way both pairs are recomputed — the year and month, and the day and hour — and a pillar that comes back different flags the chart as sitting near a boundary instead of presenting one answer as settled. Say unknown and only the year and month are retested, across twelve hours; the day and hour test is recorded as excluded rather than run, because a chart with no time to test would cross into the neighbouring day every time and a flag that is always on says nothing.
How the free reading is assembled
The numbers come first. The element balance counts every visible stem and branch at a weight of 2 and every stem hidden inside a branch at a weight of 1. When the five rounded figures do not total a hundred, the difference is applied to the largest of them — added when rounding fell short, taken back when it overshot. That is why the bars always read 100, and that too is a decision of ours.
The Day-Master support figure is a separate count: the weight of your own element plus the element that generates it, plus 8 if the month branch is your element or 5 if it generates it, plus 3 for each pillar rooted in your element. At 60 percent and above the chart is called support-heavy, below 40 support-light, and between them balanced. Those two cut points are numbers we chose, and they are stated here rather than buried.
Ten Gods are counted for every stem except the Day Master's own and for every hidden stem, then grouped into five families; the largest family becomes the lens the career and money section is written through. Your pillars are also checked against each other in seven tables — the stem combination, and the branch six-combination, six-clash, six-harm, punishment, destruction and Wonjin — and tallied as connecting or tension signals. What selects a sentence is not how many of each there are but whether each kind is there at all: connecting only, tension only, both, or neither. The sentence for that tone opens the relationships section and the life-direction section, and the two counts themselves are quoted as numbers inside the relationships text. The two lightest elements become the balancing candidates the lucky-elements section offers.
The free reading is those numbers choosing sentences. No language model is involved in writing any of it, so the same birth details return the same words however many times you ask — details meaning the whole form, not only the date: change the time, its stated accuracy, the timezone, the calendar or the fine-tune box and the pillars move, so the sentences do too.
What changes in the paid report
The chart does not change. The writer does. The paid report is written by a language model handed the facts calculated above. Two things travel with them that are not calculations: the nickname it addresses you by, and one archetype line keyed to your heaviest element, taken from the same hand-written list the free reading titles itself with. Everything else is computed — the pillars, the element percentages, the Ten Gods and the support ratio are the engine's own values, and the model only interprets them. It is instructed to quote each figure exactly as given, never to add, average or derive a new one, and not to pad a short section by repeating the same numbers.
It is also given facts the free text does not use: the ten-year luck pillar you are currently in, how many years into it you are and the one that follows; the 공망 branches read from your day pillar and which of your pillars land on them; and where the Day Master stands on each branch under 십이운성. Where the Day Master is yin the schools read every branch differently, so the second reading is handed over as well, and that difference is to be named once in the whole report rather than in every section.
Generated readings are stored against the calculated signals, so a chart that computes the same as one already read reuses the words already written and nothing is generated and billed twice. If the generation fails, the free reading is shown instead — someone who has just paid should see their chart rather than an error screen. The one failure we do not paper over that way is the daily ceiling on generations — 400 a day across every paid reading on the site: there you get a plain “briefly unavailable, please try again shortly” rather than a substitute, because what stopped is a spending decision of ours and not anything about your chart, and quietly handing over the free text would hide that. Separately, asking for the paid generation more than six times in a single minute from one connection drops the calls past that limit to the free text.
No free screen ever calls the model. That is not only a policy but a test: it walks every route on the site and fails if one that can reach the model is missing the check that this request was paid for.
How much weight to give it
This is a calculation, not a prediction. The model that writes the paid report is told explicitly not to predict specific events, dates, illness, pregnancy, accidents, lawsuits, exam or investment results, and to give no medical, legal or financial advice. What it may write about is a tendency visible in the structure of the chart, which is not a fate you are held to.
The calculation keeps its own limits too. This engine reports itself as beta, and it does not declare a single governing element. The two lightest elements are offered as balancing candidates only, because that is not a declaration that they are the right answer.
What you can read free is roughly the first 30 percent of every section — the opening one included, which is cut exactly like the rest. That cut is made on the server, so the withheld text never reaches your browser at all: it is not blurred out, it was never sent. The preview card then shows roughly the first 40 percent of that already-shortened opening, so the first thing on screen is nearer a tenth of the section than a third. A section short enough that 30 percent of it falls under 40 characters is left whole.
The birth details you type are used for this response and not stored. Before anything is sent back, the response is walked for the input fields themselves — date of birth, birth time, timezone, calendar type — and the free response is refused rather than delivered if one is still attached. It is a check on the structure leaving the server, not a search through the prose for something that looks like a date.
Read the full rules — calendar, true solar time, 공망 and 십이운성 →