K-MUDANG Philosophy Column - Why we do not let AI calculate the chartA reading laid on a wrong chart does not sound odd at all.2026-06-04 (rev. 2026-09-09)

✶ This is the K-MUDANG philosophy column. It sets out the boundary between what the code calculates and what Yong Hae-in reads. The references at the foot of the page name the classics and the astronomical data every calculation rests on.

The first thing Saju looks at is the Day Master. The stem of the day you were born, which is you.

Say the Day Master is Yang Fire. Born in the depth of winter it is a fire about to go out. Born at the height of summer it is a fire already spilling over. The first has to be kept alive with wood and fire. The second has to be held down with water. This is climate balance.

The prescriptions are opposite. One character, the month of birth, decides which.

The K-MUDANG engine returns eight climate verdicts. Extreme Cold, Cold, Cold Damp, Cool Damp, Balanced, Dry, Extreme Dry, Water Excess.

One piece of code picked out the cold ones. It checked whether the word Cold appeared in the verdict. Extreme Cold matches. Cold Damp matches. Cool Damp does not.

Cold charts were counted as not cold. Only the Cool Damp ones, and quietly.

The Vital Element can flip and the text still reads fine. Nothing turns awkward, so reading will not catch it. It surfaced when we put the same chart through the Korean and English editions and held the two answers side by side.

What makes the AI Saju going around dangerous is not that the reading is strange. It is that the chart is wrong to begin with, invented rather than worked out. And the trouble is that a reading laid on a wrong chart does not sound odd at all. A language model cannot calculate, but it tells a story well.

■ A language model does not calculate. It continues.

A language model is a machine for stringing together the character that plausibly comes next. It is good at plausible comfort.

But opening an almanac and working out when a solar term begins is a different job. Hand it a birth date and time and it pulls the wrong Day Pillar. It reverses the order of the luck cycles and attaches stars that are not there.

The trouble is what comes after. A reading laid on a wrong chart still reads smoothly. The person reading it has no way of knowing it is not their chart.

■ There is no room for taste in a saju calculation

The chart, the true-solar-time correction, the luck cycles, the Vital Element, the interactions between characters. The answer is already fixed.

For Seoul you subtract thirty-two minutes, the longitude gap from the standard meridian at 135 degrees east. Daylight saving is unwound as well, from whatever your birthplace was actually running that year. Korea had it in 1948 to 51, 1955 to 60 and 1987 to 88, and every other country carries its own record. The luck-cycle number is the real count of days from birth to the next solar term, divided by three.

Thirty-two minutes is Seoul's figure. Yours depends on where inside your own time zone you were born, not on how far that is from Korea. True solar time in Madrid runs seventy-five minutes behind the clock. In Warsaw, on the same clock, it runs twenty-four minutes ahead. New York sits within four minutes of its own meridian and Tokyo nineteen minutes east of it. The further east you sit inside a zone, the earlier the sun crosses overhead. So there is a separate field for the place of birth, 463 cities across 120 countries.

The hour can also be entered as an hour and a minute instead of picked from a two-hour block.

Most of the time you do not need the minute at all. It only matters when the moment of birth sits on the boundary between two branches, and that boundary moves with the birthplace and the date. So we do not say that adding the minute makes it more accurate. We work out the stretch of minutes that actually splits the result for that date and that birthplace, and put one line in front of the people who could land inside it.

Nowhere in any of this is there room for the reader's taste. The rules only went into code exactly as the classics set them down.

■ Where the day actually turns

The date turns at midnight, but the Saju day turns at the Rat Hour. The Rat Hour opens late at night, crosses midnight, and closes in the small hours. The boundary of a day sits inside a ninety-minute door.

Anyone born after that door opens has a Day Pillar that could belong to either date. The heaviest pillar in the chart moves a full day.

This is where the schools part. The Early Rat Hour counts it as the next day the moment the hour opens. The Late Rat Hour keeps it as today until midnight. Neither is wrong. They start from different premises.

We use the Early Rat Hour. It is the majority view, and picking one and holding it seemed better. If you want the Late Rat Hour, change it in the settings. The chart is rebuilt the moment you do.

The boundary misses the hour for the same reason. Your clock is set to one meridian while your birthplace sits on another, so the sun crosses overhead early or late by that gap and the twelve branch boundaries slide with it. In Seoul the gap is thirty-two minutes late. In Tokyo it is nineteen minutes early.

■ Yong Hae-in speaks only on numbers already fixed

The finished chart goes to Yong Hae-in exactly as it is. She does not calculate, and she does not build the chart. She takes the settled chart, the luck cycles, the Vital Element and the stars, and works out only what they mean in this person's life.

Why separate the calculation from the reading. Because whoever interprets, if they also hold the calculation, will want to nudge the chart toward what sounds good. People do it. Models do it. Better to put it out of reach than to ask anyone to resist.

Finishing the calculation and handing the result to a language model inside a fixed frame is itself filed as a patent. Application 10-2026-0053341, “System and Method for Preventing LLM Hallucination through Structured Prompt Transformation of Domain-Specific Computation Results.”

“I do not reckon the chart. I only read the letters already written in the heavens, in the earth, and on your face.”
— Yong Hae-in

■ What changes because of it

The chart does not come out wrong. There is nowhere for guesswork or hallucination to enter.

The original shape is not bent by anyone's mood or by sales pressure. The rules sit in code as the classics wrote them, and code has no mood.

The same birth data yields the same chart, whoever reads it and whenever.

That last one is the hardest. Code keeps changing, and a change in one place can move a calculation somewhere unrelated. So we pin 121 charts in each of Korean, English and Japanese as the right answers and check against them on every change. One character off and the commit is blocked. The Cool Damp problem was caught right there.

■ It does not end with one question

A fortune-teller reads you once and it ends, or charges you again to ask more. There is no way to keep the reading in a note. And the AI Saju going around invents a different chart, a different person, every time you ask again.

At K-MUDANG the chart is already fixed, so you can keep asking after the first reading. Whether to change jobs this year, how things sit with that person, whether to take this road. Yong Hae-in answers each one from your chart, your luck cycles, your Vital Element, never from a horoscope line sent to everybody.

The chart is fixed, so the readings stay consistent. And you can copy any of them out as plain text and keep it.

■ A good day is not the same day for everyone

Date selection works the same way.

Picking a good day out of the calendar looks like it has one answer for everyone, and half of it does. Ten day pillars counted among the gravely inauspicious are avoided whatever you are doing. Moving house and opening a business look to the days the spirits are away. A ground-breaking rite looks to the days the old almanacs mark open. That far, everyone gets the same answer.

The other half is decided by your own chart. If that day's pillar carries your Vital Element it is a good day. If it carries the element you avoid, skip it. For a wedding we check whether that day's branch harmonises with your birth-day branch. For an exam we look to the Support Element first.

So the same date can be the best day of the month for one person and a day to avoid for another. That is where this parts from a calendar handing everyone the same lucky days.

Yong Hae-in does not calculate this either. Twelve months of scores come out first. She takes the table and works out only why it should be that day. The basis is the date-selection theory of the imperial almanac, with Vital-Element day scoring laid on top.

■ We handle face reading the same way

When face reading opened after Saju, the first thing settled was the same boundary. What the camera reads off a face is landmark coordinates, nothing more. The ratio of forehead to jaw, the angle at the outer eye, the width of the nose. All of it comes out as numbers. Calculation stops there, and what the numbers mean goes to Yong Hae-in.

A face is not a chart, though. A birth moment never moves once it is set, while a face can measure differently with the device, the light and the angle. So face reading does not stand in for Saju. The chart is the reading and the face is material laid on top of it.

We went through this here too. When face reading first opened, the same person came out as one element type on a webcam and another on a phone.

It was the calculation, not the interpretation. Facial coordinates arrive normalised separately, horizontal against the frame width and vertical against its height, and at the point where face length is divided by face width the two were mixed, so the camera's aspect ratio was multiplied straight into the result. What we fixed was the coordinate, not the sentence. Multiply the horizontal by the frame ratio and both axes are measured with the same ruler.

There is one reason Saju never has this problem. A birth date and time reads the same on any device.

Saju is not divination. It is the oldest mirror for looking at yourself. A mirror has to be accurate and clean. In front of a clouded one, even the kindest words end up being about somebody else.

So the polishing of the mirror and the sitting down to look into it were left in different hands, all the way through.

✶ The chart is calculated in code from the almanac, the solar terms and true solar time, and every change to that code is checked back against 121 golden-master charts in each of Korean, English and Japanese.