The Three Definitions Missing From Every Accuracy Rate

A few weeks ago, I came across a table circulating on a public discussion board. Someone had put together a tidy comparison chart listing six different systems of traditional Chinese divination, assigning a neat accuracy percentage to each one. One system was listed at eighty to eighty-five percent, another at eighty-five to ninety, and one line read eighty to eighty percent, the same number written on both ends of the range. Beneath the post, dozens of readers engaged in spirited debate over which system was superior, but not a single commenter asked how those numbers had been calculated. The creator of that table was almost certainly not attempting to deceive anyone. He was reporting a genuine personal impression in good faith. The problem is that a personal impression is not a measurement, and treating an impression as a statistical fact distorts how these systems actually function.

There is a structural flaw visible on the surface of those figures. A genuine empirical measurement does not produce a padded range like eighty to eighty-five percent. If you run a hundred controlled trials and record eighty-four hits, your measured result is eighty-four percent. It is not an elastic window. When someone quotes a range instead of a single integer, they are sharing a memory, a feeling, or a casual estimate. They are not presenting a recorded tally.

To convert a personal feeling into a meaningful percentage, you must first establish three baseline definitions. The first definition is what constitutes a single prediction. If a consultant speaks with you for an hour and utters forty sentences, how many distinct predictions were made? Is every sentence counted as one unit? Is every calendar year considered one unit? Or is every discrete real-world event treated as one unit? Consider a hypothetical reading containing thirty descriptive sentences that together outline two major life events, such as a career change and a marriage. If twenty-four of those thirty sentences sound broadly accurate to you, but both major life events turn out completely false, what is the accuracy rate of that session? If you calculate by sentence count, the accuracy is eighty percent. If you calculate by event count, the accuracy is zero percent. Without fixing both the numerator and the denominator beforehand, any mathematical ratio you produce is entirely arbitrary.

The second definition requires establishing what constitutes a hit, who makes that judgment, and when the judgment occurs. There is a fundamental difference between evaluating a claim against a strict rubric set down before an event and letting a subject decide whether a statement felt accurate after the fact. Human minds possess a well-documented tendency to take ambiguous descriptions and map them onto personal experiences. This phenomenon is stable, easily reproducible in psychological settings, and has nothing to do with dishonesty or weak character.

When a reading includes statements such as you experienced noticeable emotional pressure two years ago, or you have a close relative with an unyielding temper, the listener immediately searches their personal history for a corresponding match. If the authority to declare a hit remains exclusively with the person receiving the reading, and if there is no hard temporal limit on when the event must manifest, a sufficiently flexible statement will almost always find a landing spot eventually. That outcome does not mean the prediction struck a narrow, specific target. It simply means the target was broad enough to cover the entire room.

The third and most critical definition is the baseline comparison. In formal evaluation, this is known as the base rate. Consider another hypothetical example where a reading states that you will experience interpersonal friction at work during the coming twelve months. For a working adult in any modern organizational setting, the baseline probability of encountering at least one workplace dispute over the course of an entire year is already very high before any chart is drawn. If an astrological reading predicts that outcome and it happens, achieving high accuracy on paper is trivial because the background probability was already overwhelmingly high.

A prediction only possesses real information value if its accuracy significantly exceeds the natural base rate of the event in question. The true measure of a prediction lies not merely in whether it happens to be correct, but in how many plausible possibilities it successfully rules out in advance. If a statement applies to eight out of ten people selected randomly off the street, learning that it also applied to you provides zero evidence regarding the validity of the tool used to generate it.

If you want to test the accuracy of a reading without relying on subjective memory, there is a practical test you can run yourself. Take the transcript or notes from a consultation, extract ten specific descriptive statements, and strip away all identifying markers including names, specific dates, and geographical references. Then hand those ten isolated statements to three people who have never met you and know nothing about astrology. Ask each of them to mark how many of those ten statements accurately describe their own lives.

The average percentage of statements accepted by those three uninvolved strangers represents your empirical base rate. The vital condition of this test is that the ten statements must be copied down before you assess the reading. Selecting statements after you already know the life outcomes invalidates the sample. If your reading achieves an eighty percent agreement rate from you, but complete strangers also agree with seventy-five percent of the exact same statements, the actual predictive signal generated by the reading is essentially non-existent.

To understand why sweeping accuracy claims break down, it helps to dismantle what people casually group together as fortune telling into three distinct operational layers.

The first layer is chart calculation. Having written chart generation software engines myself, I can confirm that this layer is pure, deterministic computation. Given a specific set of birth parameters, the placement of stars, palaces, and temporal cycles follows a strict sequence of mathematical logic. If two separate computer programs execute the exact same algorithm on identical birth data, they will output identical charts every single time. At this foundational level, the process is entirely reproducible.

The second layer is symbolic interpretation. This step involves translating spatial and structural relationships into probabilistic descriptions of personal tendencies, behavioral patterns, and potential life themes. In theory, these probabilistic assertions could be subjected to rigorous empirical testing over large sample sizes. In practice, however, virtually no one in the field conducts controlled, double-blind trials to measure their statistical validity.

The third layer is concrete prediction, where a practitioner claims to foretell specific, named future events, such as winning a lawsuit in May or purchasing a home in a specific city. The core difficulty with this third layer is not whether it is accurate or inaccurate. The difficulty is that the criteria for what constitutes a wrong prediction are never defined. If a reading asserts that you will change jobs in autumn, and you instead receive a major promotion with new responsibilities at your current firm, is that a hit or a miss? Because the boundary conditions for failure are never specified in advance, calculating an accuracy rate becomes logically impossible. Where there is no operational definition of error, the concept of accuracy cannot exist.

Even my earlier assertion that the first layer, chart calculation, is completely deterministic comes with a necessary qualification. Chart calculation is only reproducible once every foundational convention has been explicitly agreed upon in advance.

When converting modern birth records into a traditional chart structure, a developer or practitioner must resolve several methodological ambiguities. Do you adjust standard clock time to true solar time based on the exact geographic longitude of the birth location? If you make that adjustment, which specific geographic coordinate system do you use? How do you treat the midnight boundary, specifically whether the hour of the Rat between eleven at night and midnight belongs to the current calendar day or the following day? When a leap month occurs in the lunar calendar, do you assign its calculations to the preceding month or the subsequent month? Each of these questions has at least two established traditional schools of thought. If two software programs or two practitioners do not share the exact same conventions, they will generate completely different star placements for the exact same individual. The calculation layer is deterministic, but only within a chosen set of rules. If those rules remain unstated, reproducibility fails before you even begin the process of interpretation.

This lack of rigid, universal uniformity is not a modern innovation. Historical texts on Ziwei Doushu explicitly warn against expecting a single, static formula to apply across all circumstances.

In the classical text 《紫微斗数捷览》, the author writes 「其福与祸亦不可执一例而推」, which conveys that good fortune and bad fortune cannot be deduced by rigidly applying a single rule across all cases. Having read through classical texts extensively, it is clear that even historical masters recognized that attempting to force complex human outcomes into a uniform, mechanical rule fails to reflect reality.

Similarly, in the Taiwei Fu section of 《紫微斗数全书》, the text states 「星有同躔，数有分定，须明其生克之要」, noting that even when stars reside in the same palace, their structural allocations and elemental interactions dictate different readings. Because identical star configurations require distinct interpretations depending on context, the notion of a single, global accuracy percentage is conceptually meaningless even within the internal logic of the classical system itself.

When you evaluate claims about traditional systems, you do not need to enter an argument over belief or accuse anyone of bad intentions. You only need to run a brief internal audit consisting of three straightforward questions. First, how is a single prediction defined and counted? Second, who determines whether a prediction was a hit, under what pre-declared criteria, and over what time horizon? Third, what is the base rate probability that an uninvolved stranger who knows nothing about astrology would accept the exact same description by sheer chance?

Whether a system of thought withstands rigorous analytical scrutiny and whether an individual finds personal value or comfort in it are two completely separate matters. One can admire the intricate symbolic geometry of an ancient traditional framework without accepting unverified percentage claims. If a claim cannot define what counts as a clear failure, it is not offering a measured scientific fact. It is simply sharing an appealing story. Does quoting an accuracy rate mean someone is lying

Usually not. Most of the time the number is an honest report of a personal impression rather than an invented figure. The trouble is that the number carries no unit, so the yardstick in the speaker's head and the one in yours are probably not the same yardstick.

How do I estimate the base rate myself

Copy ten statements out of a reading before you evaluate it, strip out names, places and dates, then ask three people who have never had a reading whether each statement fits their own life. The share they accept is your base rate. Copying them first is the whole point, because picking statements afterwards means grading your own paper.

If chart calculation is deterministic, why do two programs produce different charts

Because several human choices sit in front of the calculation. Whether to convert to true solar time, which longitude to use, whether the late night hour belongs to today or tomorrow, and which side of a leap month the calculation falls on. Each has more than one traditional answer. Once both sides write their conventions down, the charts line up.

What is left after those three questions

The part worth discussing. Which layer is computation, which layer is inference, and which layer has no definition of error yet. Once those are separated you at least know what you are trading for what.