Is Qi Men Dun Jia Accurate? What You Can and Can't Verify
Is Qi Men Dun Jia accurate? One of our own AI misses reviewed step by step, a published case that fails once ages are recounted, the Barnum effect, and what you can verify.
Accuracy has two layers: whether the chart follows its stated rules, which you can audit, and whether a reading comes true, which needs a specific claim written in advance and checked later. We publish one of our own misses and have no hit rate to claim.
Asking “Is Qi Men Dun Jia accurate?” can mean two different things. Did the site calculate the chart according to its stated rules? Did a reading help someone reason about a real decision? The first is reproducible. The second needs a specific question and an honest record of what happened afterward. Neither means that a chart proves the future.
QimenMind chart for Beijing time 2026-09-29 23:00. Chart fields can be reproduced; the six-column review log requires real evidence to evaluate the reading.
What can be audited
Record the input in the chart tool, then inspect the four pillars, solar term, bureau, Zhifu (Chief Star) and Zhishi (Chief Gate). This site's question chart treats manual time as Beijing time and derives its automatic bureau by Chaibu. If you are comparing apps, the chart-difference checklist helps isolate a mismatch before you compare prose.
What depends on the question
Two people asking at the same hour may receive the same underlying chart. Their subjects, options and chosen focal points can differ, so the reading can differ too. State the decision, the people involved, the deadline and the evidence you already have. The question-writing guide gives a practical structure. Keep the initial reading and compare it with what later occurred rather than remembering only the parts that seemed right.
Medical symptoms, legal rights and investments require qualified professional advice and primary evidence. A Qimen reading does not replace a clinician, lawyer or financial analysis. The chart is available without an account. Dunyuan AI reading requires sign-in; the first three AI questions are free in total per account. You should still check its claims against the chart and the world.
Two claims require two kinds of evidence
A reproducible chart and a useful interpretation should be evaluated separately. The first claim is computational: given a date and minute, a time convention, a day boundary and a stated Ju method, can another run reproduce the same pillars, term, configuration and Zhifu and Zhishi placements? QimenMind exposes those fields on its chart, and someone else can recast it with the current hour chart or a manual time. For a manually entered question time it uses Beijing time, and its automatic Ju follows Chaibu. The field comparison guide shows how to find the first disagreement. Passing this check says that a particular chart follows the declared method. It does not demonstrate that an event will happen.
The second claim concerns a reading of one event. The reader must identify the people or objects in the question, select the Useful God (Yong Shen), compare relevant palaces and describe which real-world conditions affect the judgment. Two people can cast at the same hour and share the same base chart while asking about different events. A statement such as “this gate is favorable” has little testable content until it says whose action, against which target, and by what deadline. Keep those interpretation choices visible so the later outcome can be compared with the original statement rather than with a convenient memory of it.
A real miss: will a missing child be found?
In a missing-person case review we keep on file, someone asked whether a missing child would be found. Our AI first answered that the child would "probably be found". It read the Fei Fu path as Declining qi and took that to mean the missing-person situation was easing. The record states the actual outcome: the child had drowned. The problem was not the word "found" but the optimistic reading, which missed clear signs of physical danger in the chart.
Reviewed afterward against our lineage's San Yuan method, the reading went wrong in four places:
- It applied Declining qi to the wrong subject. Declining qi describes a strength that is ebbing. In this retrospective review, because the question concerned the missing child's safety, it should have been read for the child's condition first. In that chart the path ran from Yuan Shen (Xun palace, Gestating) through Zhifu (Kun palace, death) to Tian Yi (Qian palace, Depleted), which the review treated as a danger signal to check first, not as an easing situation.
- It skipped the Gua Qi Palace. The San Yuan method reads the Gua Qi Palace first to set the overall direction. Here it fell in Qian, in a Depleted state, held Tian Rui (lodged with Tian Qin), and had the Scenery Gate in tomb, with its element controlling the palace (Gate Pressures Palace): an adverse baseline. The AI went straight to the Dui palace of the child's birth-year stem instead.
- It chose only the mild images. The Dui palace held the Obstruction Gate, the Soaring Snake and the Ji-over-Yi pattern "Tomb Spirit Unclear". The AI read them as hiding and being frightened; the review held that, in a safety question, this combination should first be checked as a sign of being blocked or trapped. The Kan palace, the palace of water, held the Death Gate, and the reading never mentioned it. No single symbol confirms a death or its cause; the review relied on several layers pointing the same way.
- It reached a conclusion first. It started from "will be found" and picked supporting symbols, instead of reasoning in order from the Gua Qi Palace to the Useful God palace, related palaces and the Fei Fu path.
After this review, we added dedicated missing-person rules to the AI in April 2026 and refined them in August. They require that a real disappearance is reported to the police first; that a sign of contact or return is not treated as proof of safety; and that any life-or-death conclusion has agreement across several layers of the chart, with the level of risk stated when the evidence is insufficient, instead of a promise of safety. These are rules the AI is given, and it can still break them. We publish this record because seeing exactly which step failed helps anyone check a reading system for the same mistakes. In a real disappearance, medical emergency or accident, contact the police, a doctor or a search team first. Do not wait for a chart.
A published case that partly fails under our age rule
The 2022 Qimen collection Dun Jia Ji Zhang includes a destiny-chart case: in one year the subject earned several hundred thousand from live-streaming; the next year an online crackdown led to frequent account bans and falling income. The book explains the contrast with the personal annual palace of each year, and our calculator reproduces the book's birth chart, starting age and first two ten-year cycles.
The book, however, counts age as the difference between calendar years, while QimenMind follows our lineage's rule: age one at birth, changing at Lichun. Under our rule, each year's personal annual position moves forward by one earthly branch: the profitable year moves from Hai in Qian to Zi in Kan, and the year of restrictions from Zi in Kan to Chou in Gen. The book's reasoning for the profitable year no longer applies, and the evidence for it is now insufficient. The book explains outcomes that were already known and offers no verifiable prediction written in advance, so it cannot show that Qi Men predicts accurately. It also shows how a case can stop holding up once the age rule changes.
Why "it felt accurate" is not evidence
Psychologists call the tendency to accept vague, general statements as a personal fit the Barnum effect (see the APA Dictionary of Psychology). Lines like "you have been under pressure lately" or "someone around you will help" can be fitted to many people's lives. A reading is only testable if it made a specific claim in advance that could turn out wrong: who it concerns, the deadline, and what outcome would count as a match.
Does QimenMind publish a hit rate?
No. We have no prediction hit rate to publish, and we will not substitute a few hand-picked successes for one. Open a saved question to add an outcome: when you look back, mark it as matched, partly matched, did not match or not happened yet, and note what actually happened. These records help you review your own questions first. If we ever publish a summary, we will state how many of all records have feedback, how many are still unresolved or unclear, and why voluntary feedback cannot be read as a hit rate.
Ask a question whose result you can observe
Compare “Will my career go well?” with “I have completed two interviews; a competing offer expires this week. Should I keep waiting for the first employer before the deadline?” The second question identifies one event, its stage, the decision and a time window. A later record can say whether a reply arrived, when it arrived and which choice the person made. This is an example of question design, not a prediction about an employer. The question guide offers further examples of keeping one chart tied to one matter.
For a relationship question, an observable event might be whether a planned conversation occurs and what each person explicitly says. A chart cannot directly establish another person's unspoken thoughts. Words, agreements and sustained conduct provide evidence about the real relationship. If a question has no deadline or observable outcome, record it as an open reflection rather than scoring a reading as right or wrong afterward.
A practical review log has six columns: the original question, the casting time and its time convention, facts already known at casting, the answer and the chart fields it cited, the action taken, and the outcome at the agreed review date. Mark each claim as supported, contradicted or still unresolved. Do not rewrite “may be delayed” into “will arrive on Friday” after learning the result. Such a log cannot produce a universal accuracy percentage from one person or a few events, but it can make a single conclusion accountable.
Why the same hour can lead to different readings
The base pillars, term and palaces may be shared. The Useful God and host-guest relationship depend on the question. One person may ask about entering a company, another about a partner's payment, and a third about whether to send a message. Calling the same star “good” for all three skips the role it plays in each event. A reading should name the selected role and palace before describing the relationship, the strength or weakness of that palace, and conditions such as emptiness or a pressed gate. The public Zhifu and Zhishi guide likewise treats the Chief Star and Chief Gate as coordinates for analysis rather than automatic verdicts.
A follow-up about the same matter can continue from the locked chart. If the subject changes to another person or another event, define a new question and, when appropriate, cast a new chart. Recasting the same question repeatedly until one answer sounds reassuring weakens the record: it becomes unclear which timestamp and answer are being evaluated. When two readings conflict, first compare their inputs, Useful Gods and known facts, then ask which claim actually differs.
Where a reading should stop
Medical symptoms and treatment require clinical records and qualified care. Legal rights require the relevant documents and professional advice. Investment decisions require product details, downside risk and financial judgment. Urgent safety situations call for direct real-world help. A Qimen reading can help organize a question, but those decisions need evidence that the chart does not contain. The same boundary applies to AI prose: a fluent answer can misread a field or invent a background detail the user never supplied.
For Dunyuan AI reading, first cast the chart and check the time, Ju, Zhifu and Zhishi the answer cites. Ask which statements are chart observations and which are conditional interpretations. Charting does not require sign-in. AI reading does; the first three AI questions are free in total per account. That allowance describes access, not the reliability of a particular conclusion.
FAQ
Is there a real example of a Qi Men reading being wrong?
Yes. This article reviews one of our own AI's misses: in a missing-person question it read Declining qi as the situation easing and missed signs of physical danger, and the record states that the person had died. The review found four errors, and we added dedicated missing-person rules afterward.
Can a chart prove that an event will happen?
No. You can audit how the chart was calculated, but the chart alone does not establish a future outcome.
Why can two people in the same hour get different readings?
Their base chart may match while the question, subject and focal point of interpretation differ.
Can this replace medical or legal advice?
No. Seek qualified advice based on the relevant records and circumstances.
