Evidence level: B
人格匹配失败,不等于事件预测也失败——这是两个不同的考场。要分出胜负只有一种办法:让从业者在事前把“谁会遇到什么事、落在哪个时间窗里”写死,然后再去等结果。
一、Carlson 协议考的是“配对”,不是“预测”
Carlson 1985 年那个双盲实验,任务其实是配对:给占星师一张星盘,让他从三份人格档案里挑出对应的那一份。28 位资深占星师的结果是 33.7%,几乎就是 1/3 的随机水平。
但配对和预测是两回事。配对任务里答案已经存在(三份档案里必有一份是对的),从业者做的是辨认;预测任务里答案还没发生,从业者做的是断言——在结果未知时说出一个可被证伪的东西。两个考场的成绩不必然相关。所以命理圈那句辩护——“人格描述测不出来,我们测的是事件”——不能靠 Carlson 的数据直接驳回,它是另一个问题。
二、先看人类在这个考场上的基线
在断言“命理一定输”之前,得知道这个考场的正常分数。最硬的数据来自领域专家预测自己最熟的事:2022 年一项研究让 740 位临床试验主要研究者(PI)在试验注册后七个月内,对主要终点能否阳性、招募能否完成、能否按时结束给出概率预测。用 Brier 分数衡量(越小越准,“永远猜 50%”的无信息基线是 0.25),PI 对主要终点的平均分正好是 0.25,与无信息预测持平;对招募和时间线更差(0.38 与 0.52)(PMID 35134071)。
另一项研究找 71 位神经科专家预测三个即将揭盲的随机对照试验,最终只有 18% 的专家表现出超过无信息基线的预测技能(PMID 32546652)。
结论很关键:掌握全部数据的领域专家,在结构化事件预测上平均也只是打平随机。 临床与机械预测的经典元分析给的是同方向结论:机械(统计)组合比人判断平均准约 10%,且这个优势不随任务、判断者、经验多少而改变(PMID 10752360;早期综述见 PMID 1408625)。
三、但也不能假设命理必输——这个考场有可识别的“能手”
如果事件预测是纯噪声任务,这套协议就没有检验价值——谁都过不了,也就测不出差别。好在它不是。美国情报部门资助的地缘政治预测锦标赛证明:把历年表现最好的人挑出来组成精英队,他们能在数百个问题上持续保持高准确度,而且连续两年都没有向均值回归——预测能力一部分是“被发现”的,一部分是“被造出来”的(PMID 25987508)。
也就是说,事件预测是有真实信息量、能分出高下的任务,拿它当考场是公平的。如果命理从业者真如自己所说“测事件准”,就该在这里显示出超过随机的判别力,而不是只在现场让人点头。
四、协议要怎么写才算“可证伪”
关键是把三样东西在事前锁死:断言、时间窗、判据。
断言要具体到可核验,比如“此人未来五年内会发生一次职业变动”。不能是“感情会有波折”——那样事后怎么解释都成立。
时间窗必须在事前预注册,这是整套设计最要紧的一条。事后随便指一个时间点,谁都“说得中”,因为人生够长、事件够多;窗口前移锁定后,“什么时候”就不再能从数据里挑。这与后见之明偏误同源:结果一旦已知,人就会觉得它本来更容易预料(PMID 26168501)。
判据要双轨:命中率(事件是否发生、是否落在窗内),加上校准与 Brier 分数——“发生了”和“我说过会发生”是两个数。校准近年甚至被称为预测研究的“阿喀琉斯之踵”:判别力好不等于概率数值准(PMID 31842878)。
还要三样配套。随机对照臂:同一批命盘,一半用真盘、一半打乱(从业者不知哪半是自己);真盘命中率若不显著高于打乱盘,那“事件预测准”就只是基准发生率在起作用。多重检验校正:一次断的事很多,不矫正的话纯随机数据也会冒出“显著”关联——用安大略省 1067 万居民全量数据按星座扫描 223 个常见住院诊断,确实找出两个“显著”关联,校正后却全部消失(PMID 16895820;机制见 PMID 27879038)。从业者分层:资历、流派、断言是事前给出还是事后追认,都要单独记录。
五、易学这一侧的落点:应期传统本来就是“事件预测版”
这套协议不必被当成外来物。传统术数内部一直有一条“应期”传统——先立断言,后候应期。斗数讲流年断事、六壬讲课应期、奇门讲应期,结构上都是“事件 + 时间窗”的断言。也就是说,古人本来就主张自己在这一栏上能打,而这一栏恰恰从没被对照过。这套设计做的事,就是把传统最看重的那一栏第一次挂上随机基线。
反过来说也成立:如果从业者在人格匹配上失败、却在事件预测上显著高于随机,那说明的恰恰是人格匹配这个考场选错了——信息通道若存在,也不在“性格描述”里,而在“事件与时间”里。这是一个可赢也可输的检验,不是预设结论的批判。
还要提醒:这类“异常信息传递”研究的复制记录本身分裂——有元分析未能复制(PMID 10414223),也有用更大数据库合并后得出相反结论的(PMID 11393304),更宽的综述把整个领域列为争议(PMID 29792448, PMID 33041926)。正因如此,事件预测协议才必须走预注册——否则任何一方的“显著”都不过是研究者自由度开出来的花。
证据与局限
已验证: 临床与机械预测的准确度差异有元分析支撑(PMID 10752360, 1408625);领域专家的结构化事件预测平均与无信息基线持平(PMID 35134071, 32546652);后见之明偏误稳定存在并污染事后判断(PMID 26168501);事件预测任务本身可分离出高技能个体(PMID 25987508);多重检验可造出虚假关联且校正后消失(PMID 16895820, 27879038);概率预测的校准与判别是两个独立维度(PMID 31842878)。
部分验证: “随机打乱命盘”的对照臂在中文命理语境中尚无执行数据;已有相关设计只测过人格描述与占星摘要的辨认(PMID 18649494)。异常信息传递类研究的复制状态仍有争议(PMID 10414223 与 PMID 11393304 相互对立)。
待验证: 命理从业者在“预注册断言 + 时间窗 + 双轨判据”协议下,事件命中率是否显著高于随机对照臂;该判别力是否高于人格匹配任务;判别力与从业年限、流派、断言给出方式(事前 vs 事后)之间的关联结构;时间窗长度与命中率之间的函数形态。
相关阅读
- 八字与紫微斗数的盲法检验:Carlson 协议在中文命理体系上的系统执行
- 命理“感觉准”与“真的准”差多少:巴纳姆基线的定量分离
- 命理与占星预测的盲法检验:科学如何看待“出生时刻决定命运”
- 研究池子 P2:时空频率拓扑
证据等级: B(科学类比/间接证据——事件预测的基线、判据工具与偏误机制均有直接实验支撑,但“预注册断言 + 时间窗”协议尚未在中文命理从业者中执行)
参考文献: PMID 10752360, PMID 1408625, PMID 35134071, PMID 32546652, PMID 25987508, PMID 26168501, PMID 31842878, PMID 16895820, PMID 27879038, PMID 18649494, PMID 10414223, PMID 11393304, PMID 29792448, PMID 33041926
免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。
Can \”Event Forecasting\” in Fate Reading Survay Blind Testing: Fixing Assertions and Time Windows in Advance
Evidence level: B
Failed personality matching does not equal failed event prediction—these are two different examinations. The only way to determine the winner is to have practitioners hardcode “who will encounter what event and within which time window” in advance, then wait for the results.
I. Carlson’s Protocol Tests “Matching,” Not “Prediction”
Carlson’s 1985 double-blind experiment was actually a matching task: given an astrological chart, the astrologer had to select the corresponding profile from three personality files. The results of 28 senior astrologers were 33.7%, nearly matching the random level of 1/3.
However, matching and prediction are distinct. In a matching task, the answer already exists (one of the three files must be correct); practitioners perform identification. In a prediction task, the answer has not yet occurred; practitioners make assertions—stating something falsifiable when the outcome is unknown. Scores in these two examinations are not necessarily correlated. Therefore, the defense in the Yijing Studies community—”personality descriptions cannot be measured; we measure events”—cannot be directly refuted by Carlson’s data, as it addresses a different issue.
II. Establishing the Baseline for Humans in This Examination
Before asserting “Yijing Studies must fail,” one must know the normal score for this examination. The most robust data comes from domain experts predicting their own expertise: a 2022 study asked 740 clinical trial principal investigators (PIs) to provide probability predictions for whether the primary endpoint would be positive, recruitment would be completed, and the trial would end on schedule within seven months of registration. Measured by the Brier score (lower is better; an uninformed baseline of always guessing 50% yields 0.25), PIs averaged exactly 0.25 for the primary endpoint, matching the uninformed prediction; performance was worse for recruitment and timelines (0.38 and 0.52 respectively) (PMID 35134071).
Another study found that among 71 neurology experts predicting three upcoming open-label randomized controlled trials, only 18% of experts demonstrated predictive skills exceeding the uninformed baseline (PMID 32546652).
The conclusion is critical: Domain experts with full data still average no better than random in structured event prediction. Classic meta-analyses of clinical and mechanical prediction offer conclusions in the same direction: mechanical (statistical) combinations are approximately 10% more accurate than human judgment, and this advantage does not change with task type, judge, or experience level (PMID 10752360; early review see PMID 1408625).
III. But We Cannot Assume Yijing Studies Must Fail—This Examination Identifies “Experts”
If event prediction were a pure noise task, this protocol would lack validation value—nobody would pass, and no differences could be measured. Fortunately, it is not. A geopolitical prediction tournament funded by the US intelligence agency demonstrates that selecting the best performers from past years to form an elite team allows them to maintain high accuracy across hundreds of questions, and there was no regression to the mean over two consecutive years—predictive ability is partly “discovered” and partly “created” (PMID 25987508).
In other words, event prediction is a task with real information content and discernible hierarchy; using it as an examination is fair. If Yijing Studies practitioners truly claim “accurate event prediction,” they should demonstrate discriminative ability exceeding random here, rather than merely gaining nods in the field.
IV. How to Design the Protocol for “Falsifiability”
The key is to lock down three elements in advance: assertions, time windows, and criteria.
Assertions must be specific enough to verify, such as “this individual will experience one career change within the next five years.” Vague statements like “there will be emotional turbulence” allow post-hoc explanations regardless of outcome.
Time windows must be pre-registered; this is the most critical element of the design. Arbitrarily selecting a time point after the fact allows anyone to “hit the mark” because life is long and events are numerous; once the window is locked forward, “when” cannot be cherry-picked from the data. This stems from the same mechanism as hindsight bias: once results are known, people perceive them as more predictable (PMID 26168501).
Criteria must be dual-track: hit rate (whether the event occurred and fell within the window) plus calibration and Brier scores—”did it happen” and “did I say it would happen” are two distinct numbers. Calibration has recently been termed the “Achilles’ heel” of prediction research: high discriminative ability does not equate to accurate probability values (PMID 31842878).
Three supporting elements are also required. Randomized control arm: half of the same charts use real charts, half use shuffled ones (practitioners do not know which half is their own); if the hit rate of real charts is not significantly higher than shuffled charts, then “accurate event prediction” merely reflects the base rate. Multiple testing correction: many assertions are made at once; without correction, purely random data can yield “significant” associations—scanning 223 common hospital diagnoses against zodiac signs in Ontario’s full population of 10.67 million residents did indeed find two “significant” associations, which vanished after correction (PMID 16895820; mechanism see PMID 27879038). Practitioner stratification: seniority, school affiliation, and whether assertions were given in advance or retroactively recognized must all be recorded separately.
V. The Yijing Studies Landing Point: The “Ying Qi” Tradition Is Essentially an “Event Prediction Version”
This protocol need not be viewed as foreign. Within traditional divination, there has always been a tradition of “Ying Qi” (Applicable Period)—first establishing an assertion, then awaiting the Ying Qi. Dou Shu (Star Chart) discusses judging events via annual cycles, Liu Ren teaches Ying Qi, and Qimen Dunjia focuses on Ying Qi; structurally, they all assert “event + time window.” In other words, ancient practitioners already claimed competence in this category, yet this category has never been subjected to control. This design places the traditional system’s most valued category for the first time against a random baseline.
Conversely, if practitioners fail in personality matching but significantly exceed random in event prediction, it indicates that the personality matching examination is flawed—if an information channel exists, it lies not in “character description” but in “events and timing.” This is a test that can be won or lost, not a predetermined critique.
Also note: replication records for such “anomalous information transfer” studies are split—some meta-analyses failed to replicate (PMID 10414223), while others using larger merged databases reached opposite conclusions (PMID 11393304), and broader reviews list the entire field as controversial (PMID 29792448, PMID 33041926). Precisely because of this, the event prediction protocol must follow pre-registration; otherwise, any “significance” is merely a researcher’s degree of freedom blooming.
Evidence and Limitations
Verified: Meta-analyses support accuracy differences between clinical and mechanical prediction (PMID 10752360, 1408625); domain experts’ structured event prediction averages match the uninformed baseline (PMID 35134071, 32546652); hindsight bias is stable and contaminates post-hoc judgments (PMID 26168501); event prediction tasks can isolate high-skill individuals (PMID 25987508); multiple testing can create false associations that vanish after correction (PMID 16895820, 27879038); calibration and discrimination are independent dimensions in probability prediction (PMID 31842878).
Partially Verified: Control arms using “randomly shuffled charts” lack execution data within the context of Chinese Yijing Studies; related designs have only measured recognition of personality descriptions and astrological summaries (PMID 18649494). The replication status of anomalous information transfer studies remains controversial (PMID 10414223 opposes PMID 11393304).
To Be Verified: Whether Yijing Studies practitioners achieve significantly higher event hit rates than the random control arm under a “pre-registered assertion + time window + dual-track criteria” protocol; whether this discriminative ability exceeds that of the personality matching task; the structural relationship between discriminative ability and practitioner years, school affiliation, and assertion timing (advance vs. retroactive); the functional form between time window length and hit rate.
Related Reading
- Blind Testing of Bazi and Ziwei Dou Shu: Systematic Execution of the Carlson Protocol in Chinese Divination Systems
- How Much Difference Between “Feeling Accurate” and “Truly Accurate” in Divination: Quantitative Separation of Barnum Baselines
- Blind Testing of Divination and Astrology Prediction: How Science Views “Birth Time Determines Destiny”
- Research Pool P2: Spatiotemporal Frequency Topology
Evidence Level: B (Scientific Analogy/Indirect Evidence—baselines for event prediction, criteria tools, and bias mechanisms have direct experimental support, but the “pre-registered assertion + time window” protocol has not yet been executed among Chinese Yijing Studies practitioners)
References: PMID 10752360, PMID 1408625, PMID 35134071, PMID 32546652, PMID 25987508, PMID 26168501, PMID 31842878, PMID 16895820, PMID 27879038, PMID 18649494, PMID 10414223, PMID 11393304, PMID 29792448, PMID 33041926
Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.
Leave a Reply