Blog

  • 浮沉弦滑数迟,各由频谱的哪一维说话

    Evidence level: B

    六个脉象词不是一个六选一的选择题,而是五个能独立变化的物理量各自在频谱上找位置——真正要裁决的不是”能不能分类”,而是”哪个要素归哪根轴”。

    正文

    上一篇把问题设成六分类:浮、沉、弦、滑、数、迟,看判别 AUC。这个设计能出结果,但埋了陷阱——分类器准确率再高,也不能证明脉象的物理含义被找对了。它完全不需要知道”弦”是什么,只要在特征空间里给”弦”划一块地,哪怕那块地的物理意义是”这批人刚好都六十岁以上”,AUC 也会很好看。

    中医脉象本来也不是六个互斥的类。临床上”弦滑”并见是常态,”沉细”是常态。强行单标签分类,等于要求一个同时具备多个属性的人只报一个——信息在这一步就被压掉了。而且六个词不在同一层级:浮/沉说深度,弦说张力,滑说光滑度,数/迟说频率,虚实说振幅。这是五个可自由组合的维度。所以要做的不是又一个分类实验,而是逐要素裁决:拆回五个物理量,各自去频谱上找最对应的那根轴。判据不是”AUC 高不高”,而是”五个要素同时进模型时,某要素是否只在它自己那根轴上有独立增量“。

    四根候选轴

    轴一:能量重心(C1 占比)。 最”像”浮沉,但证据在往下拆。2026 年一项 919 人(6-92 岁)桡动脉研究用 Womersley 数传播模型算了前四阶谐波的复数波速:第一谐波是主导能量贡献者,也是稳定性最高的一阶;随年龄变化的是虚部 Y1(代表衰减)显著下降,而实部 X1(相速度)相对稳定,能量向低阶重新分配,老年人脉形因此更宽更满(PMID 42521805)。低频能量占比主要跟着血管老化走,而不是跟着手指按的深浅走。 采集端证据更直接:横移或加压时波形振幅都是先大后小,接触压力超过 100 mmHg 后模板相关系数掉到 0.983 以下(PMID 20092621)。“浮沉”物理上最直接对应的是”最佳接触压力落在哪一档”——这是采集几何量,不一定是频谱量。 这是本篇最需要诚实面对的可能结局:浮沉也许是四要素里唯一”不在频谱里”的那个。

    轴二:谐波衰减率(振幅随阶次的下降斜率)。 弦脉最像这根轴。手感”绷直、按不住”,物理上对应管壁张力高、波速快、反射叠加明显。919 人研究给的正是直接支撑:代表衰减的 Y1 是最敏感的老化量(PMID 42521805)。2026 年一篇桡动脉谐波综述把链条铺得更宽:谐波参数与动脉僵硬度、内皮功能、自主神经失调、血管重构一致地相关(PMID 42490820)。张力→衰减率,是四根轴里物理链条最短的一根。

    轴三:高阶谐波结构(相邻阶比值、某阶振幅变异系数)。 滑脉最像,也最易误判。傅里叶级数确实分得开:24 名受试者的三维脉图里,正常、弦、滑三组在第一谐波比值 r1 与比值差 Δr1 上显著不同(PMID 31885653)。但另一项 105 人、3×4 通道阵列研究给了反向线索:弦脉阵列脉搏体积 APV 显著区别于正常(6.26 vs 6.04)与滑(6.26 vs 6.07),而正常与滑之间没有显著差异(P=0.75)(PMID 29790062)。两者合起来只允许一种解释:滑脉的签名不在空间分布上,只在时间序列的高阶层。 这也解释了”滑”为什么难摸——它不是形状,是变化率。旁证:第 4 阶谐波振幅变异系数(C4CV)在 2324 例 2 型糖尿病者中独立预测大血管事件(HR 2.13)、微血管事件(HR 2.08)与心血管死亡(HR 3.03),校正年龄、血压、HbA1c、病程后仍成立(PMID 31591828)。

    轴四:基频本身(心率)。 数/迟直接就是它。但这里有个必须挑明的污染:45 例高血压患者的自动评级与血流动力学对照显示,浮脉的判定阈值竟然是心率 ≥ 58.25 次/分,数脉是心率 ≥ 69.6 且 PWV ≥ 9.185 m/s,强脉是脉压 ≥ 62.5 mmHg(PMID 26988431)。浮脉的自动判定落在心率上——这不是中医的问题,是”用统一压力档位测所有人”的协议问题:心率快的人舒张期回缩少,同档压力下波形显得更浮。不把心率剥掉,浮沉会一直跟数迟混在一起。

    裁决怎么做

    三件事同时成立才算裁决成功。①要素独立评级:≥3 位医师对五要素各自打分(深度/张力/光滑度/力度各 1-5,频率直接用心率),允许”弦+滑”同时高分,kappa ≥ 0.6 入组。②关联矩阵 + 增量检验:先算”五要素 × 四轴特征”完整相关矩阵(看谁天然纠缠),再做逐步模型,每加一个要素看它是给自己那根轴带来独立增量(ΔAUC/ΔR²),还是给别人的轴带来增量(说明定义交叉)。③强制协变量:年龄、心率、平均压、接触压力档位先进模型——上面四段证据至少三段会被这几个协变量改写结论。

    判据提前写死,避免事后解释:某要素只在预设那根轴上有独立增量且效应量最大 → 该轴可作为该要素的物理定义候选;某要素在所有四根轴上都没有独立增量 → 它不在频谱里,属采集几何或纯触觉维度,原表述必须改写。 后者不是失败——把”浮沉”从频谱模型请出去、改挂到”最佳接触压力”上,本身就是一次有价值的定位纠正。

    本篇不重复 #1160 的”找交接阶次”(差分),不重复 #501 的”三方份额分解”(方差归因),也不重复原六分类研究(整体判别)。它做的是从”类”退回”要素”再重新分配轴——这一步错了,前面所有分类 AUC 都是在为一个定义不清的标签服务。

    证据与局限

    已验证: 桡动脉前十一阶谐波重复性优秀(观察者间 ICC 0.83-0.96,>90% 谐波值落在 Bland-Altman 两倍标准差内,PMID 27170904);正常/弦/滑三组第一谐波比值显著不同(PMID 31885653);弦脉 APV 区别于正常与滑,正常与滑无差异(PMID 29790062);浮沉强弱快慢与血压、心率、PWV 显著关联(PMID 26988431);C4CV 独立预测糖尿病大/微血管事件与心血管死亡(PMID 31591828);第一谐波主导且最稳、Y1 随龄显著下降而 X1 稳定(PMID 42521805);接触压力 >100 mmHg 破坏波形完整性(PMID 20092621);谐波参数与僵硬度和自主神经功能一致相关(PMID 42490820)。

    部分验证: “五要素 × 四轴”对应关系只有零散单点证据(弦↔衰减、滑↔高阶比、数迟↔基频),从未在同一批受试者身上同时测过;”浮沉不在频谱里”只是从接触压力实验与 919 人老化研究推出的合理怀疑,尚无直接对照。

    待验证: ①五要素独立评级下的”要素×轴”关联矩阵,哪个要素在哪根轴上增量最大;②控制年龄、心率、接触压力后浮沉的谐波判别力是否归零(归零即证实其为采集几何量);③弦滑并见的复合脉象在频谱上是两轴叠加还是产生新交互项。

    相关阅读

    证据等级: B(要素-血流动力学关联、弦滑谐波/阵列差异、谐波重复性、高阶谐波变异预测结局各有直接文献;”五要素逐一对轴裁决”本身无先例,属研究提案)
    参考文献: PMID 42521805, PMID 42490820, PMID 31885653, PMID 29790062, PMID 26988431, PMID 31591828, PMID 27170904, PMID 20092621


    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。

    Floating, Sinking, Stringing, Sliding, Numbers, Delay: Which Dimension of the Spectrum Do They Speak From

    Evidence level: B

    The six pulse terms are not a six-option multiple-choice question, but rather five independently variable physical quantities locating themselves on the spectrum. The true criterion is not “can we classify,” but “which element belongs to which axis.”

    Main Text

    The previous article framed the problem as a six-classification task: floating, sunken, string-like, slippery, rapid, and slow, evaluating discrimination AUC. This design yields results but embeds a trap—a classifier’s accuracy does not prove the physical meaning of pulse terms has been correctly identified. It requires no knowledge of what “string-like” actually is; as long as a region is assigned to “string-like” in the feature space, even if that region physically corresponds to “this specific cohort happens to be over sixty,” the AUC will still look excellent.

    Traditional TCM pulse diagnosis is not inherently six mutually exclusive categories. In clinical practice, “string-like and slippery” appearing together is common; “sunken and thin” is also common. Forcing single-label classification is equivalent to requiring a person possessing multiple attributes to report only one—information is lost at this step. Furthermore, the six terms do not exist on the same hierarchy: floating/sunken describe depth, string-like describes tension, slippery describes smoothness, rapid/slow describe frequency, and full/empty describe amplitude. These are five freely combinable dimensions. Therefore, the task is not another classification experiment, but element-by-element adjudication: decompose back into five physical quantities and identify the most corresponding axis for each on the spectrum. The criterion is not “high AUC,” but “whether an element contributes independently only on its own axis when all five elements enter the model.”

    Four Candidate Axes

    Axis 1: Energy Centroid (C1 proportion). Most resembles floating/sunken, but evidence requires further decomposition. A 2026 study of 919 participants (ages 6–92) using a Womersley number propagation model calculated the complex wave speeds of the first four harmonics: the first harmonic is the dominant contributor to energy and also the most stable first-order mode; the imaginary part Y1 (representing attenuation) decreases significantly with age, while the real part X1 (phase velocity) remains relatively stable, with energy redistributing toward lower orders, resulting in wider and fuller pulse forms in the elderly (PMID 42521805). Low-frequency energy proportion follows vascular aging rather than finger pressure depth. Direct evidence from the acquisition end is more immediate: waveform amplitude increases then decreases during lateral displacement or pressure application, dropping below a template correlation coefficient of 0.983 when contact pressure exceeds 100 mmHg (PMID 20092621). “Floating and sunken” physically correspond most directly to “at which pressure tier the optimal contact pressure falls”—this is an acquisition geometric quantity, not necessarily a spectral one. This is the most honest outcome we must face in this article: floating and sunken may be the only element among the four that is not “in the spectrum.”

    Axis 2: Harmonic Decay Rate (amplitude decline slope with order). The string-like pulse most resembles this axis. The tactile sensation of “tight and unyielding” corresponds physically to high wall tension, fast wave speed, and obvious reflection superposition. The 919-person study provides direct support: Y1, representing attenuation, is the most sensitive aging indicator (PMID 42521805). A 2026 review on radial artery harmonics broadens the chain: harmonic parameters are consistently related to arterial stiffness, endothelial function, autonomic dysfunction, and vascular remodeling (PMID 42490820). Tension → decay rate is the shortest physical chain among the four axes.

    Axis 3: Higher Harmonic Structure (ratios of adjacent orders, amplitude coefficient of variation for specific orders). Slippery pulse most resembles this and is most prone to misjudgment. Fourier series indeed separate them: in 3D pulse maps of 24 subjects, normal, string-like, and slippery groups showed significant differences in the first harmonic ratio $r_1$ and difference $\Delta r_1$ (PMID 31885653). However, another study with 105 subjects and a 3×4 channel array provided counter-evidence: string-like pulse array pulse volume APV significantly differed from normal (6.26 vs 6.04) and slippery (6.26 vs 6.07), but there was no significant difference between normal and slippery (P=0.75) (PMID 29790062). Combined, these allow only one interpretation: the signature of the slippery pulse is not in spatial distribution but in higher-order harmonics within the time series. This also explains why “slippery” is hard to feel—it is not a shape, but a rate of change. Supporting evidence: the amplitude coefficient of variation (C4CV) of the fourth harmonic independently predicts major vascular events (HR 2.13), microvascular events (HR 2.08), and cardiovascular death (HR 3.03) in 2,324 type 2 diabetes patients, remaining valid after adjusting for age, blood pressure, HbA1c, and disease duration (PMID 31591828).

    Axis 4: Fundamental Frequency (Heart Rate). Rapid and slow are directly this. However, there is a contamination that must be explicitly acknowledged: automatic rating of 45 hypertensive patients compared with hemodynamic controls showed that the threshold for judging floating pulse was heart rate ≥ 58.25 bpm, rapid pulse was heart rate ≥ 69.6 and PWV ≥ 9.185 m/s, and strong pulse was pulse pressure ≥ 62.5 mmHg (PMID 26988431). The automatic judgment of floating pulse falls on heart rate—this is not a TCM issue, but a protocol problem of “using a unified pressure tier for all subjects”: individuals with fast heart rates have less diastolic recoil, so under the same pressure tier, the waveform appears more floating. Without stripping away heart rate, floating and sunken will always mix with rapid and slow.

    How to Adjudicate

    Three things must hold simultaneously for successful adjudication. ① Independent Element Rating: ≥3 physicians rate each of the five elements individually (depth/tension/smoothness/strength on a 1–5 scale, frequency directly via heart rate), allowing “string-like + slippery” to score high simultaneously; kappa ≥ 0.6 for inclusion. ② Correlation Matrix + Incremental Testing: First calculate the full correlation matrix of “five elements × four-axis features” (to see who is naturally entangled), then perform stepwise modeling, adding one element at a time to check if it brings independent increment to its own axis ($\Delta$AUC/$\Delta$R²) or brings increment to another axis (indicating definition crossover). ③ Forced Covariates: Age, heart rate, mean pressure, and contact pressure tier must enter the model first—the conclusions in the previous four sections are likely rewritten by these covariates.

    The criteria are fixed in advance to avoid post-hoc explanation: if an element has independent increment only on the preset axis and shows the largest effect size → that axis can serve as a physical definition candidate for that element; if an element has no independent increment on any of the four axes → it is not in the spectrum, belonging to acquisition geometry or pure tactile dimension, and the original expression must be rewritten. The latter is not a failure—removing “floating and sunken” from the spectral model and reattaching it to “optimal contact pressure” is itself a valuable positional correction.

    This article does not repeat the “finding transition orders” (differentiation) from #1160, nor the “three-party share decomposition” (variance attribution) from #501, nor the original six-classification study (overall discrimination). It performs a return from “class” to “element” and reallocation of axes—if this step is wrong, all preceding classification AUCs have been serving a definition-unclear label.

    Evidence and Limitations

    Verified: Excellent repeatability of radial artery first eleven-order harmonics (inter-observer ICC 0.83–0.96, >90% harmonic values within two standard deviations of Bland-Altman limits, PMID 27170904); significant differences in first harmonic ratios among normal/string-like/slippery groups (PMID 31885653); string-like pulse APV distinguished from normal and slippery, with no difference between normal and slippery (PMID 29790062); floating/sunken strength/speed significantly associated with blood pressure, heart rate, and PWV (PMID 26988431); C4CV independently predicts diabetic macro/microvascular events and cardiovascular death (PMID 31591828); first harmonic dominant and most stable, Y1 decreases significantly with age while X1 remains stable (PMID 42521805); contact pressure >100 mmHg compromises waveform integrity (PMID 20092621); harmonic parameters consistently related to stiffness and autonomic function (PMID 42490820).

    Partially Verified: The correspondence between “five elements × four axes” consists only of scattered single-point evidence (string-like ↔ decay, slippery ↔ higher-order ratios, rapid/slow ↔ fundamental frequency), never simultaneously measured on the same cohort; “floating and sunken not in the spectrum” is only a reasonable suspicion derived from contact pressure experiments and the 919-person aging study, lacking direct comparison.

    To Be Verified: ① The “element × axis” correlation matrix under independent element rating, identifying which element shows maximum increment on which axis; ② Whether harmonic discriminative power for floating/sunken returns to zero after controlling for age, heart rate, and contact pressure (zero return would confirm it as an acquisition geometric quantity); ③ Whether composite pulse patterns appearing as both string-like and slippery represent two-axis superposition or generate new interaction terms in the spectrum.

    Related Reading

    Evidence Level: B (direct literature exists for element-hemodynamic associations, string-like/slippery harmonic/array differences, harmonic repeatability, and higher-harmonic variation predicting outcomes; “element-by-axis adjudication” itself has no precedent, belonging to research proposal)
    References: PMID 42521805, PMID 42490820, PMID 31885653, PMID 29790062, PMID 26988431, PMID 31591828, PMID 27170904, PMID 20092621


    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.

  • 几何只画出户型,装修才决定回声:脉象谐波方差的”三方分账

    Evidence level: B

    同一张户型图,装修不同,回声就不同。”血管几何决定脉象频谱”要过关,必须回答的不是”几何重不重要”,而是”几何占几成”。

    正文

    上一篇把 1-D 模型的误差按谐波阶次排成了阶梯,发现账越往高阶越难对平。但它只回答”错多少”,没回答”错在谁身上”。这篇就做一次分账:把预测的谐波能量当成一个总方差,拆成三份——血管几何一份、管壁弹性一份、微循环边界一份,看各自能解释多少。

    先厘清这三者不是同一种东西。几何是半径、长度、分支位置,是”管子长什么样”,短期内基本不变;弹性是管壁有多硬、有多粘,中年以后缓慢变化;微循环边界是末梢那一大片小血管给主干的”出口阻力”,它受内皮功能、局部代谢、交感张力调控,几分钟内就能变。用房子的比喻:几何是户型图,弹性是墙体的材料和厚度,边界是每一个房间门开多大。三个都不改户型,但改回声的方式完全不同。

    第一笔账:弹性项有多大。上一轮引过的体外实验给了最干净的量级——37 条动脉全参数实测、零拟合的条件下,纯弹性管壁在末梢产生”欠阻尼的高频振荡”,加入 Voigt 型粘弹性后,压力波第 5-20 谐波幅度误差从 0.7% 降到 0.5%,流量侧从 7.0% 降到 3.3%(PMID 21724188)。注意这个 0.5% 的前提是”弹性已知”:一旦弹性从实测变成反演,误差就开始计入。Eck 等做过一个正面回答——在同一个 37 动脉网络、同一套三个临床可测输入(颈股脉搏波速度、平均动脉压、肱动脉截面积)下,比较三种常用管壁模型的预测质量,并量化了波速、压力、截面积三项测量不确定度传到 8 个位置压力流量波形上的不确定性(PMID 27890534)。也就是说,”壁模型选哪个”本身就是一个独立的方差源,它和几何误差不是一回事,也不能互相抵消。

    第二笔账:边界项有多大。Coccarelli 等把末端边界从传统三元件 Windkessel 换成”多孔弹性管”来代表微循环阻力,结果明确指出:边界模型的参数对系统的压力与流量分布有”深刻影响”,且换模型后仍能与实验数据对上(PMID 30900050)。这句话的技术含义是——同一个几何、同一套弹性,只换出口边界,波形就能改到你分不清是谁的功劳。肺循环的全局敏感性分析也给了同类结论:模型输出对远端血管阻力最敏感,而且参数影响力会随网络复杂度改变(PMID 31355521)。出口是最容易被随手设一个数的地方,也是最容易被记错账的地方。

    第三笔账:神经张力。这笔账常被漏掉,因为它在模型里表现为”边界参数”,在生理上却是一个活的旋钮。Nardone 等的综述汇总了微神经电图测的肌肉交感神经活动(MSNA)与颈股脉搏波速度、增强指数的关系,结论是交感神经系统能够在独立于当前血流动力学与血管运动张力的条件下调节大动脉硬度(PMID 33035441)。如果这句成立,那么”边界条件”在活人身上不是常数,是会随情绪、睡眠、体位变的值——中医说”弦脉随情志而变”,物理落点就在这里:不是几何变了,是那条出口被拧紧了。

    三个账都摊开了,接下来是方法论问题:怎么拆才不算错账。第一,参数之间在生理上相关。年龄大的人血管又粗又硬,几何与弹性天然共变;把二者当独立变量做朴素主效应分解,交互项会被错记到某一方头上。虚拟冠脉队列研究已经示范了处理办法——用能容纳”接受准则造成的输入相关性”的敏感性分析方法来分解变异来源(PMID 40172563)。第二,输出指标选谁,会改变”谁是主因”的排序。输出约束的全局敏感性分析发现,在血流储备分数的诊断灰区(0.75-0.85)里,侧支流量这类解剖参数的影响力显著,直接挑战了”狭窄程度主导一切”的常规认知(PMID 40758586)。第三,时间序列量不能直接用经典 Sobol 指数。Eck 等为此专门提出”时间平均敏感度指数”,并证明它比传统 Sobol 更准确地刻画参数对时变压力流量不确定性的贡献(PMID 27890534)。做随机不确定度量化的经典路线是随机配点加稀疏网格(PMID 23653286),但落到”逐谐波分账”上,还得把每一阶谐波单独当一次输出。

    顺带说明这套分账与相邻工作的分界:上一轮反推出的”两版预测差量找交接阶次”,只做两次预测相减、看误差在哪一阶换手;这篇做的是三方份额的方差分解——把几何、弹性、边界同时放进随机采样,估各自的主效应与交互项占比。一个是差分,一个是分解,不重复。

    判据也很干脆:若几何的主效应在高阶谐波上仍占多数(>50%),”血管几何决定脉象频谱”可以维持原表述;若高阶上弹性与边界合计过半,则这句话必须改写为”几何定骨架,弹性与边界定高频细节”,而高频细节恰是中医脉诊里辨”弦、紧、滑”的关键区间。

    证据与局限

    已验证: 管壁粘弹性是抑制高频欠阻尼振荡的关键项,其加入使第 5-20 谐波误差从 0.7% 降至 0.5%(PMID 21724188);壁模型选择与壁参数测量不确定度会显著改变 1-D 模型预测(PMID 27890534);末端边界模型参数对压力流量分布有实质影响(PMID 30900050);模型输出对远端阻力最敏感且敏感性随网络复杂度变化(PMID 31355521);交感神经可独立于血流动力学调节大动脉硬度(PMID 33035441);存在可处理相关输入与输出约束的敏感性分析方法(PMID 40172563、PMID 40758586、PMID 23653286)。

    部分验证: “几何、弹性、边界三方份额”尚未在”同个体 + 逐谐波输出”的框架下被同时估计过;上述敏感度结论多来自时域波形或单一标量输出(vFFR、AI、PWV),逐阶谐波上的份额分配属推断;”交感张力经由出口边界改写高频谐波”目前是机制通路,缺直接逐阶证据。

    待验证: ①同个体几何固定下,弹性/边界参数随机采样的谐波能量方差分解(各阶主效应+交互项占比);②含生理相关结构的采样(几何-弹性共变)与独立采样两种设定下份额估计的偏差量;③三方份额的跨个体异质性(是否随年龄分层单向漂移);④急性交感激活(冷压试验/握力)前后出口边界的等效参数变化与高频谐波响应。

    相关阅读

    证据等级: B
    参考文献: PMID 21724188, PMID 27890534, PMID 30900050, PMID 31355521, PMID 33035441, PMID 40172563, PMID 40758586, PMID 23653286


    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。

    Geometry Only Draws the Floor Plan, Renovation Determines Echo: The \”Three-Way Split\” of Pulse Wave Harmonic Variance

    Evidence level: B

    A single floor plan yields different echoes depending on renovation. “Vascular geometry determines pulse spectral content” must be validated not by asking “is geometry important,” but by answering “what proportion does geometry account for.”

    Main Text

    The previous article ranked the error of the 1-D model by harmonic order, finding that higher orders are increasingly difficult to reconcile. However, it only answered “how much is wrong,” not “who is responsible for the error.” This paper performs an accounting split: treating predicted harmonic energy as total variance, we decompose it into three shares—vascular geometry, wall elasticity, and microcirculatory boundary—to assess what each explains.

    First, clarify that these three are not identical. Geometry refers to radius, length, and branch location—the “shape of the pipe,” which remains largely constant in the short term. Elasticity refers to wall stiffness and viscosity, changing slowly after middle age. Microcirculatory boundary refers to the “exit resistance” provided by the extensive network of small vessels at the periphery, regulated by endothelial function, local metabolism, and sympathetic tone; it can change within minutes. Using a house analogy: geometry is the floor plan, elasticity is the building material and thickness of walls, and boundary is how wide each room’s door is open. None alter the floor plan, yet they modify the echo in fundamentally different ways.

    First Account: The Magnitude of the Elasticity Term. Previous in-vitro experiments provided the cleanest magnitude estimates—under full-parameter measurement of 37 arteries with zero fitting, a pure elastic wall generates “underdamped high-frequency oscillations” at the periphery. Incorporating Voigt-type viscoelasticity reduced the amplitude error of the 5th–20th harmonics in pressure waves from 0.7% to 0.5%, and flow-side errors from 7.0% to 3.3% (PMID 21724188). Note the premise of this 0.5%: “elasticity is known.” Once elasticity shifts from direct measurement to inversion, error begins to be attributed here. Eck et al. provided a positive answer—comparing prediction quality across three common wall models within the same 37-artery network and identical set of three clinically measurable inputs (carotid-femoral pulse wave velocity, mean arterial pressure, brachial artery cross-sectional area), they quantified how measurement uncertainties in wave speed, pressure, and cross-sectional area propagate to uncertainty at eight locations across pressure and flow waveforms (PMID 27890534). Thus, “which wall model is selected” constitutes an independent variance source; it is distinct from geometric error and cannot offset it.

    Second Account: The Magnitude of the Boundary Term. Coccarelli et al. replaced the traditional three-element Windkessel terminal boundary with a “porous elastic tube” to represent microcirculatory resistance, explicitly concluding that boundary model parameters have a “profound impact” on systemic pressure and flow distribution, and that switching models still aligns with experimental data (PMID 30900050). The technical implication is that keeping geometry and elasticity identical while only changing the exit boundary can alter the waveform to a point where it becomes indistinguishable whose contribution caused the change. Global sensitivity analysis of the pulmonary circulation yielded similar conclusions: model output is most sensitive to distal vascular resistance, with parameter influence varying with network complexity (PMID 31355521). The exit is the place most easily assigned a single arbitrary number and most frequently misrecorded.

    Third Account: Neural Tension. This account is often overlooked because it manifests as a “boundary parameter” in the model but represents a living knob in physiology. Nardone et al.’s review summarized relationships between muscle sympathetic nerve activity (MSNA) measured via micro-neurography and carotid-femoral pulse wave velocity and augmentation index, concluding that the sympathetic nervous system can regulate large artery stiffness independent of current hemodynamics and vascular tone (PMID 33035441). If this holds true, then “boundary conditions” in living humans are not constants but values varying with emotion, sleep, and posture—matching the traditional Chinese medicine concept that “the wiry pulse changes with emotional states.” The physical locus is not that geometry changed, but that the exit was tightened.

    With all accounts laid bare, the methodological question arises: how to split without double-counting. First, parameters are physiologically correlated. Older individuals have wider and stiffer vessels; geometry and elasticity naturally covary. Treating them as independent variables for naive main-effect decomposition misattributes interaction terms. Virtual coronary artery cohort studies have demonstrated a solution—using sensitivity analysis methods capable of accommodating “input correlations caused by acceptance criteria” to decompose variance sources (PMID 40172563). Second, the choice of output metrics alters the ranking of “primary causes.” Global sensitivity analysis with output constraints found that within the diagnostic gray zone for fractional flow reserve (0.75–0.85), anatomical parameters like collateral flow influence significantly, directly challenging the conventional认知 that “stenosis severity dominates everything” (PMID 40758586). Third, time-series quantities cannot be directly used with classical Sobol indices. Eck et al. specifically proposed a “time-averaged sensitivity index,” proving it characterizes parameter contributions to time-varying pressure-flow uncertainty more accurately than traditional Sobol indices (PMID 27890534). The classic route for stochastic uncertainty quantification involves random sampling combined with sparse grids (PMID 23653286), but for “harmonic-by-harmonic accounting,” each harmonic order must be treated as a separate output.

    Incidentally, clarify the boundary between this accounting and adjacent work: the previous iteration’s “inferred difference between two prediction versions to find transition harmonic orders” performed only subtraction of two predictions to identify where the error switched orders. This paper performs a three-way variance decomposition—simultaneously placing geometry, elasticity, and boundary into random sampling to estimate main effects and interaction term proportions. One is differential; the other is decomposition; they do not overlap.

    The criteria are straightforward: if geometry’s main effect remains dominant at higher harmonics (>50%), the statement “vascular geometry determines pulse spectral content” can be maintained. If elasticity and boundary collectively exceed half at higher orders, the statement must be rewritten as “geometry sets the skeleton, while elasticity and boundary determine high-frequency details,” where these high-frequency details correspond precisely to the critical range in TCM pulse diagnosis for distinguishing “wiry, tight, and slippery” qualities.

    Evidence and Limitations

    Verified: Wall viscoelasticity is the key term suppressing high-frequency underdamped oscillations; its inclusion reduced harmonic 5–20 errors from 0.7% to 0.5% (PMID 21724188); selection of wall models and uncertainties in wall parameter measurements significantly alter 1-D model predictions (PMID 27890534); terminal boundary model parameters have a substantive impact on pressure-flow distribution (PMID 30900050); model output is most sensitive to distal resistance with sensitivity varying by network complexity (PMID 31355521); the sympathetic nervous system can regulate large artery stiffness independent of hemodynamics (PMID 33035441); sensitivity analysis methods exist that handle correlated inputs and output constraints (PMID 40172563, PMID 40758586, PMID 23653286).

    Partially Verified: The “three-way share” of geometry, elasticity, and boundary has not been simultaneously estimated within a framework of “same individual + harmonic-by-harmonic output”; most sensitivity conclusions derive from time-domain waveforms or single scalar outputs (vFFR, AI, PWV), making harmonic-by-harmonic share allocation inferential; the mechanism linking “sympathetic tension rewriting high-frequency harmonics via exit boundary” currently lacks direct harmonic-by-harmonic evidence.

    To Be Verified: ① Harmonic energy variance decomposition under random sampling of elasticity/boundary parameters with fixed individual geometry (proportions of main effects and interactions); ② Bias in share estimation under two settings: sampling with physiological correlation structures (geometry-elasticity covariance) versus independent sampling; ③ Cross-individual heterogeneity of three-way shares (whether showing unidirectional drift with age stratification); ④ Changes in equivalent parameters of exit boundaries and high-frequency harmonic responses before and after acute sympathetic activation (cold pressor test/grip strength).

    Related Reading

    Evidence Level: B
    References: PMID 21724188, PMID 27890534, PMID 30900050, PMID 31355521, PMID 33035441, PMID 40172563, PMID 40758586, PMID 23653286


    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.

  • 高谐波为什么更难看准:1-D模型误差的’阶次阶梯

    Evidence level: B

    “血管几何决定脉象频谱”要成立,光有”波形像”不够,得逐阶对账。而账越往高谐波走,越难对平——难在哪里、是谁的错,就是这篇要拆的事。

    正文

    中医脉诊的现代解读里,最扎实的一条物理线索是谐波。桡动脉压力波做傅里叶分解,前 10 个谐波的相对能量(C1-C10)被用来对应脏腑与疾病状态——冠心病患者的谐波指标并入传统危险因素后,判别曲线下面积从 0.788 升到 0.856(PMID 31275406);2 型糖尿病患者第四谐波振幅的变异系数(C4CV)是独立的心血管事件预测因子(PMID 31591828);第一谐波 C1 升高也独立预测非致死性心梗与卒中(PMID 30381150)。谐波不是玄学,是能预报结局的信号。

    但”能预报”和”能解释”是两回事。王唯工体系的终极解释是:谐波能量由各器官供血动脉的几何共振决定——血管是琴弦,器官是弦上挂的钟。检验这条因果链只能靠闭环:同一受试者重建动脉几何 → 1-D 模型预测压力波 → 实测同一位置波形 → 两者都做 FFT → 逐阶核对能量。上一篇已论证过这个闭环的层级(PMID 17640653),这篇往下钻一层:预测误差是不是随谐波阶次单调爬升?高谐波误差的锅该记在谁的账上?

    先看已知事实。1-D 模型的谱级表现其实相当好,但有一个关键细节:误差在高频段更差,而且管壁粘弹性是最大的修正项。Alastruey 等在 37 条动脉的体外复刻模型上做过最干净的验证——所有模型参数直接实测、零数据拟合——纯弹性管壁模型在末梢血管产生”欠阻尼的高频振荡”,加入 Voigt 型粘弹性后,压力波平均相对误差从 3.0% 降到 2.5%、流量从 15.7% 降到 10.8%;而看频域更说明问题:第 5-20 谐波的幅度误差从 0.7% 降到 0.5%(压力)、流量从 7.0% 降到 3.3%(PMID 21724188)。注意这个数字的读法:压力波第 5-20 谐波只差 0.5%-0.7%,说明在”几何已知、弹性已知”的理想条件下,模型在高频段照样能压住误差——前提是把管壁粘弹性写进方程。

    问题恰恰出在”弹性已知”这四个字上。体外实验的管壁参数直接量出来,真实人体做不到:弹性模量要靠超声、MRI 反演,反演天然带不确定性。两篇量化研究把这事说得很清楚:肺循环 1-D 模型的全局敏感性分析发现,模型输出对远端血管阻力最敏感,而且参数影响力会随网络复杂度增加而改变——网络越细,哪根参数”说了算”越不稳定(PMID 31355521);分数阶粘弹性研究表明,管壁模型对血流动力学的影响主要由分数阶参数 α 控制,α 通过引入粘性耗散改变压力波传播(PMID 24414838)。两条一叠:几何(半径、长度、分支)与管壁特性(弹性、粘性)的测量误差传到高频谐波时会被放大——粘性耗散本就频率依赖,弹性管壁对高频振荡阻尼不足,恰是纯几何模型在高阶谐波上系统性偏高的机制入口。

    几何决定因子的第二层藏在反射位置里。波形的高频成分主要由反射波到达时刻决定——反射点近,高频抖动密;反射点远,抖得疏。1-D 模型预测高频准不准,取决于它把反射位点(胸主动脉、髂分叉、肾动脉开口)放得多准,而这又取决于分支几何的测量精度。体外复刻模型之所以高频误差也压得住,是因为 37 条动脉的长短粗细都是拿尺子量的;真实超声重建的半径误差常在 10% 量级,这一格误差进到高阶谐波,相位就开始乱。

    还有第三层,是测量端自己的账。谐波阶次越高,对应频率越高,信号能量本身越弱,信噪比越低;实测波形的高频段本来就抖。一篇 2016 年的方法学研究干脆提出:脉搏信号不该用固定谐波分解来刻,而该用”波形形状函数 + 同步挤压变换”这类自适应模型,因为脉搏形状随时间漂移,固定傅里叶框架会在高频段引入人为成分(PMID 27304979)。这提醒我们:误差矩阵里高阶谐波那一栏的”误差”,有一部分可能根本不是模型的错,而是”模型预测平均波形、实测抓的是某一拍”的错位。

    所以 #500 课题的正确打开方式,是把误差矩阵拆成三栏看:①阶次栏——预测与实测的相对能量误差按第 1-10 阶逐阶排开,先看是不是单调递增、有没有”误差拐点”阶次;②因子栏——同一受试者分别做”纯几何”(管壁弹性取人群均值)与”几何+个体化弹性”两版预测,误差差量就是几何外输入的贡献;③几何敏感性栏——对半径、长度、分支位置逐参数做扰动扫描,看哪一阶谐波对哪个参数最敏感。三栏一叠,就能回答”误差随阶次单调爬升吗、高谐波误差的几何决定因子是什么”——而这两问的答案,决定王唯工体系下一步该往哪修:如果几何+个体化弹性能把高阶误差压到临床可用,共振腔假说的物理载体就完整了;如果高谐波误差始终压不下去,说明”器官共振”的谐波映射还有几何之外的关键输入,脏腑对应表需要小修。

    这套检验还有一个反直觉的收益:它同时是”无创几何测量协议够不够细”的压力测试——阶次曲线一画出来,哪个测量环节拖后腿一目了然。它不替任何一方辩护,只把账摊开。

    证据与局限

    已验证: 1-D 模型在参数全实测条件下,压力波第 5-20 谐波幅度误差可压到 0.5%(PMID 21724188);管壁粘弹性是抑制高频欠阻尼振荡的关键修正(PMID 21724188、PMID 24414838);谐波能量携带疾病结局信息(PMID 31275406、PMID 31591828、PMID 30381150);1-D 模型输出敏感性随参数与网络复杂度变化(PMID 31355521)。

    部分验证: “几何测量误差在高频谐波被放大”是由粘弹性频率依赖特性与敏感性分析推得的机制性结论,尚无”同个体逐阶误差矩阵”的直接实测;”实测高频段含非模型性噪声”由自适应分解方法学研究支持(PMID 27304979),但错位误差的占比未被量化。

    待验证: ①同个体几何重建+1-D预测+实测FFT的谐波级误差矩阵(第 1-10 阶相对能量误差、相位差、归一化误差);②误差-阶次曲线的形状检验——单调递增还是存在”误差拐点阶次”;③”纯几何 vs 几何+个体化弹性”两版预测的误差差量(几何外输入贡献);④逐几何参数(半径/长度/分支位置/弹性)扰动扫描与逐阶敏感度排序;⑤跨个体误差矩阵的一致性(”阶次阶梯”是否人人同形)。

    相关阅读

    证据等级: B(1-D 模型谱级精度与谐波临床价值有直接文献支撑,但”逐阶误差矩阵跨阶次分布规律”本身无先例,属研究设计提案)
    参考文献: PMID 21724188, PMID 31355521, PMID 24414838, PMID 27304979, PMID 31591828, PMID 30381150, PMID 31275406


    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。

    Why High Harmonics Are More Difficult to Accurately Model: The ‘Order Staircase’ of 1-D Model Errors

    Evidence level: B

    “Vessel geometry determines pulse spectrum” holds only if we match term-by-term, not just by waveform resemblance. As we move up the harmonic ladder, matching becomes increasingly difficult. Understanding where the mismatch occurs and who bears responsibility is the core issue this paper addresses.

    Main Text

    Among the most robust physical clues in modern interpretations of Traditional Chinese Medicine pulse diagnosis is harmonic analysis. Decomposing the radial artery pressure wave via Fourier series, the relative energy of the first 10 harmonics (C1-C10) is correlated with internal organs and disease states. In coronary heart disease patients, incorporating harmonic metrics into traditional risk factors increased the discrimination area under the curve from 0.788 to 0.856 (PMID 31275406); in type 2 diabetes, the coefficient of variation for the fourth harmonic amplitude (C4CV) serves as an independent predictor of cardiovascular events (PMID 31591828); and elevated first harmonic energy (C1) independently predicts non-fatal myocardial infarction and stroke (PMID 30381150). Harmonics are not Yijing Studies in the sense of mysticism; they are signals capable of forecasting outcomes.

    However, “predictive capability” and “explanatory power” are distinct. The ultimate explanation in the Wang Weigong framework posits that harmonic energy is determined by geometric resonance within organs supplied by arteries—vessels act as strings, and organs are bells suspended upon them. Validating this causal chain requires a closed loop: reconstruct arterial geometry from the same subject → predict pressure waves via a 1-D model → measure the waveform at the same location → perform FFT on both datasets → verify energy term-by-term. The previous paper demonstrated the hierarchical nature of this closed loop (PMID 17640653); this paper drills down one level deeper: Does prediction error monotonically increase with harmonic order? Whose responsibility should be assigned to high-harmonic errors?

    First, let us review the known facts. The spectral performance of the 1-D model is actually quite good, but a critical detail exists: errors worsen at high frequencies, and wall viscoelasticity is the largest correction term. Alastruey et al. conducted the cleanest validation to date across 37 extracorporeal arterial replica models—all model parameters were measured directly with zero data fitting. Pure elastic tube wall models produced “underdamped high-frequency oscillations” in peripheral vessels; incorporating Voigt-type viscoelasticity reduced the average relative pressure wave error from 3.0% to 2.5% and flow error from 15.7% to 10.8%. Looking at the frequency domain reveals further clarity: magnitude errors for harmonics 5–20 dropped from 0.7% to 0.5% (pressure) and flow from 7.0% to 3.3% (PMID 21724188). Note the reading of these numbers: a difference of only 0.5%–0.7% for pressure wave harmonics 5–20 indicates that under ideal conditions where geometry and elasticity are known, the model can still suppress errors at high frequencies—provided wall viscoelasticity is included in the equations.

    The problem arises specifically regarding the phrase “elasticity known.” Extracorporeal experiments measure wall parameters directly, which is impossible in living humans: elastic modulus requires inversion via ultrasound or MRI, processes inherently carrying uncertainty. Two quantitative studies clarify this issue: a global sensitivity analysis of the pulmonary circulation 1-D model found that model outputs are most sensitive to distal vascular resistance, and parameter influence changes with network complexity—as networks become finer, which parameter “holds sway” becomes less stable (PMID 31355521); fractional-order viscoelasticity research indicates that the influence of wall models on hemodynamics is primarily controlled by the fractional parameter α, which alters pressure wave propagation by introducing viscous dissipation (PMID 24414838). Combined, these findings show that measurement errors in geometry (radius, length, branching) and wall properties (elasticity, viscosity) are amplified when transmitted to high-harmonic frequencies—viscous dissipation is inherently frequency-dependent, and elastic tube walls provide insufficient damping for high-frequency oscillations, constituting the mechanism entry point for systematic overestimation in pure geometric models at higher harmonics.

    The second layer regarding geometry lies within reflection points. High-frequency components of the waveform are primarily determined by the arrival time of reflected waves—closer reflection points yield denser high-frequency jitter; farther points yield sparser jitter. The accuracy of the 1-D model’s prediction at high frequencies depends on how precisely it places reflection sites (thoracic aorta, iliac bifurcation, renal artery openings), which in turn depends on the precision of branch geometry measurements. Extracorporeal replica models maintain low high-frequency errors because the lengths and diameters of 37 arteries were measured with rulers; in contrast, radius errors in real ultrasound reconstructions often reach the 10% level. When such an error enters higher harmonics, phase coherence begins to scramble.

    There is a third layer: the account of the measurement endpoint itself. Higher harmonic orders correspond to higher frequencies, where signal energy is inherently weaker and the signal-to-noise ratio is lower; high-frequency segments of measured waveforms are naturally noisy. A 2016 methodological study proposed outright that pulse signals should not be characterized using fixed harmonic decomposition but rather via adaptive models such as “waveform shape functions + synchronous squeezing transforms,” because pulse shapes drift over time, and a fixed Fourier framework introduces artificial components at high frequencies (PMID 27304979). This reminds us that a portion of the “error” listed in the high-harmonic row of the error matrix may not stem from the model itself, but rather from a misalignment between “model-predicted average waveform” and “measured single beat.”

    Therefore, the correct approach to Project #500 involves decomposing the error matrix into three columns: ① Order Column—rank the relative energy errors between prediction and measurement across orders 1–10 to first check for monotonic increase or identify a specific “error inflection point” order; ② Factor Column—generate two versions of prediction for the same subject (“pure geometry” using population mean wall elasticity vs. “geometry + individualized elasticity”), and the difference in error quantifies the contribution of geometric external inputs; ③ Geometric Sensitivity Column—perform perturbation scans on radius, length, and branch positions parameter-by-parameter to identify which harmonic order is most sensitive to which specific parameter. By overlaying these three columns, we can answer whether error monotonically climbs with order and what the geometric determinants of high-harmonic errors are. The answers to these two questions will determine the next direction for refining the Wang Weigong system: if geometry plus individualized elasticity can suppress high-order errors to clinically usable levels, the physical carrier of the resonance cavity hypothesis is complete; if high-harmonic errors remain unsuppressible, it indicates that the harmonic mapping of “organ resonance” involves key inputs beyond geometry, requiring minor revisions to the organ-correspondence tables.

    This validation protocol also offers an counter-intuitive benefit: it serves as a stress test for “whether the non-invasive geometric measurement protocol is sufficiently detailed.” Once the order curve is plotted, any bottleneck in the measurement process becomes immediately visible. It does not defend any side; it simply lays the accounts open.

    Evidence and Limitations

    Verified: Under full parameter measurement conditions, the 1-D model can reduce pressure wave harmonic 5–20 magnitude errors to 0.5% (PMID 21724188); wall viscoelasticity is the key correction for suppressing high-frequency underdamped oscillations (PMID 21724188, PMID 24414838); harmonic energy carries information regarding disease outcomes (PMID 31275406, PMID 31591828, PMID 30381150); 1-D model output sensitivity varies with parameters and network complexity (PMID 31355521).

    Partially Verified: The mechanism that “geometric measurement errors are amplified at high harmonic frequencies” is derived from frequency-dependent viscoelasticity characteristics and sensitivity analysis, lacking direct measurement of a “step-by-step error matrix per individual”; the presence of non-model noise in high-frequency measured segments is supported by adaptive decomposition methodological research (PMID 27304979), but the proportion of misalignment error remains unquantified.

    To Be Verified: ① Harmonic-level error matrix per individual combining geometric reconstruction + 1-D prediction + measured FFT (relative energy errors for orders 1–10, phase differences, normalized errors); ② Shape testing of the error-order curve—monotonic increase or existence of an “error inflection point order”; ③ Error difference between the two prediction versions (“pure geometry” vs. “geometry + individualized elasticity”) to quantify geometric external input contribution; ④ Perturbation scanning on geometric parameters (radius/length/branch position/elasticity) and ranking sensitivity by harmonic order; ⑤ Consistency of error matrices across individuals (whether the “step-ladder” shape is uniform).

    Related Reading

    Evidence Level: B (Direct literature support exists for 1-D model spectral precision and clinical value of harmonics, but no precedent exists for “cross-order distribution laws of step-by-step error matrices,” making this a research design proposal)
    References: PMID 21724188, PMID 31355521, PMID 24414838, PMID 27304979, PMID 31591828, PMID 30381150, PMID 31275406


    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.

  • 把样本切成三份之后,还比得出谁最准吗

    Evidence level: B

    三个无量纲指标对”大凶/中凶/小凶”的判别力,看起来是一场选秀。其实它是一场被样本量、度量方式和测量噪声三重夹击的考试。

    正文

    前面几步已经做完三件事:形煞与风压异常带的重合度算过了,几何线索的暗示效应分离过了,三类异常带谁在背锅也拆过了。现在轮到最直白的一步——三类形煞各有最贴切的无量纲指标:天斩煞对风速放大比 K,尖角煞对极值负压系数 Cp,min,反弓煞对压力脉动均方根 RMS。那就把三个指标和风水师的凶应评级摆在一起,看谁最配”大凶”。

    听起来像选秀:三个选手,同一批评委,谁相关性高谁赢。这个类比第一步就错了,而且一错,后面四步会连着错。

    一、三个人考的不是同一张卷子。 K 是气流被挤得多快,Cp,min 是某点被吸得多狠,RMS 是压力抖得多厉害——三种物理过程,三条致害路径。穹顶实测显示,同一几何下外压峰值在背风区对高跨比和湍流强度并不敏感,但内压以负压为主、内外净负压峰值被局部推高(PMID 34576493)。一栋房子至少有两套压力,指标不指定测点就没有意义。更麻烦的是 Cp,min 不是直接测的,是从有限长度的压力时程里外推的,阈值取在哪会实打实改变数值(PMID 33642655)。三个指标各带各的外推误差,”谁最准”就没有共同分母。

    二、样本一切三份,功效先崩。 检验只能分层做:天斩只用 K,尖角只用 Cp,min,反弓只用 RMS。假设总共 300 例、每类 100、凶级三档均分,每类里的”大凶”只剩三十来个;要在这规模上把 AUC=0.7 和 0.5 分开,功效已在边缘。再劈一半做跨地区验证,每格十几个,算出的置信区间宽到互相重叠,排序基本由抽样噪声决定。

    这就是第一个陷阱:“哪个指标最配大凶”的阴性答案,可能不是被数据否定的,而是被样本量否定的。 两者必须分开说,否则一次功效不足的研究会被读成”风压与凶应无关”,整条线被误杀。回归里”每个变量需要多少事件”的老规矩(PMID 12025266)在这里照样适用,只是事件数已被三层分类切过一刀。

    三、凶级是序数,不能拿二分类 AUC 硬套。 常见做法是把评级压成”大凶 vs 非大凶”,可这一压就把”中凶”和”小凶”合并了——而这两档恰恰是实务里”要不要动”的分界线,等于抹掉信息量最大的那道界。序数结局有自己的度量:多分类判别指数 PDI(PMID 33866577)、序数 c-index。一致性用加权 kappa 时必须写明线性还是二次权重:二次权重对”差一档”罚得轻、对”差两档”罚得重(PMID 25516203),正对应”大凶判成小凶”远比”判成中凶”严重。权重方案不是技术细节,是价值判断,必须在方法里先写死。 同一份数据还能给出三个不同的判别力,取决于协变量是当干扰、当对象还是当增量来源(PMID 20046933)。

    四、联合模型的”涨分”可能是重复计数。 医学上这套逻辑成立过:三个炎症标志物单独 AUC 为 0.766、0.716、0.768,联合到 0.894(PMID 41929599),涨 0.13,因为三者走不同通路。风压三指标不是这样——K 和 RMS 由同一湍流场驱动,尖角处 Cp,min 与 RMS 也来自同一层分离剪切。先做共线性诊断,再跑嵌套模型。判据可定硬一点:若联合模型相对最强单指标的 AUC 增量小于 0.03 且似然比检验不显著,则三指标实为一个维度,应合并成”风压强度综合因子”再谈判别,而不是摆三个数假装有三个证据。

    五、评委那边的噪声会把阴性结果放大。 风水师评级的一致性其实不差:108 条禁忌、11 位学者加 67 位从业者,跨性别、职业、学历、流派均无显著差异(PMID 37089290)。但同一研究承认重测信度不理想——隔半年再问,判断会漂。这是测量误差,会系统性压低关联:若评级可靠性 0.6、指标 0.9,真实相关 0.5 会被观测成 0.5×√(0.6×0.9)≈0.37,正好掉到门槛下(误差规划见 PMID 33410311)。必须做衰减校正并报两套数:校正后由不显著变显著,原阴性结论作废;校正后仍不显著,结论才站得住。

    六、换个城市,排名会不会翻。 三指标都是几何与来流的函数,换城市等于换粗糙度、风气候、建筑密度。缩尺实验还有门槛:室外流动要进入雷诺数无关区间,临界值约 27000,判据是相对变化率 ≤5%(PMID 34659649)。不达门槛,系数换个尺度就不成立。所以跨地区验证不是把两地数据凑一起算总 AUC,而是先确认两地都过了门槛。若 A 城 Cp,min 最强、B 城 RMS 最强,那就不存在普适的”最配大凶的指标”,只有”在什么风气候下哪个最配”。

    所以这篇真正的产出不是谁赢,是四道门:得分层后功效够不够、序数度量用对没有、联合增量是不是共线造的、评级噪声校没校正。四门有一门没过,答案就不可信。这也顺带说明了风水量化为何长期停在”建档”进不到”判别”——不缺数据,K、Cp,min、RMS 都是现成标准量;缺的是这套门。没有门,谁都能拿三百例样本宣布自己赢了。

    证据与局限

    已验证: K、Cp,min、RMS 是风工程标准无量纲量,内外压双套、测点位置决定系数值(PMID 34576493);极值峰值须从有限时程外推、阈值选择有客观判据(PMID 33642655);缩尺实验存在雷诺数无关门槛,室外临界值约 27000、判据为相对变化率 ≤5%(PMID 34659649);风水评级内部一致性良好但重测信度不理想(PMID 37089290);序数结局有 PDI 等专用度量(PMID 33866577),加权 kappa 权重方案有明确语义(PMID 25516203);多标志物联合可显著提升判别(PMID 41929599);测量误差致关联系统性衰减(PMID 33410311)。

    部分验证: “分层后功效不足”基于 300 例的推算假设,非实测样本;三指标实际共线性程度尚无风洞或实测直接报告,仅物理机制层面推断;回归样本量规则(PMID 12025266)原用于二分类结局,用于三档序数属外推。

    待验证: ①三类形煞无量纲指标的实际相关矩阵与方差膨胀因子;②分层后各类形煞的真实可用案例数;③风水师形煞凶应评级任务的可靠性系数实测值;④跨地区(不同地表粗糙度、风气候)下指标判别排名的稳定性。

    相关阅读

    证据等级: B
    参考文献: PMID 34576493, PMID 33642655, PMID 12025266, PMID 33866577, PMID 25516203, PMID 20046933, PMID 41929599, PMID 37089290, PMID 33410311, PMID 34659649


    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 we still determine who is most accurate after splitting the samples into three parts?

    Evidence level: B

    The discriminative power of three dimensionless indicators against “Great Misfortune/Medium Misfortune/Small Misfortune” appears to be a talent show. In reality, it is an examination subjected to triple pressure from sample size, measurement methodology, and measurement noise.

    Main Text

    The preceding steps have completed three tasks: we calculated the overlap between Form Sha and abnormal wind pressure zones, separated the suggestive effects of geometric cues, and decomposed which type of anomaly zone is taking the blame. Now comes the most straightforward step—each type of Form Sha has a most appropriate dimensionless indicator: the Velocity Amplification Ratio $K$ for Sky-Cutting Sha, the Minimum Coefficient of Pressure $C_{p,min}$ for Acute-Angle Sha, and the Root Mean Square (RMS) of pressure fluctuations for Reverse-Arrow Sha. We should then align these three indicators with the Feng Shui masters’ severity ratings to see which best matches “Great Misfortune.”

    It sounds like a talent show: three contestants, one panel of critics; whoever has the highest correlation wins. This analogy is flawed from the first step, and once wrong, the subsequent four steps will also be wrong.

    1. They are not taking the same exam. $K$ measures how quickly air is squeezed; $C_{p,min}$ measures how intensely a point is sucked; RMS measures how violently pressure fluctuates—three distinct physical processes, three distinct pathways of harm. Dome measurements show that under the same geometry, external pressure peaks in the leeward zone are not sensitive to span-to-height ratios or turbulence intensity, whereas internal pressure is dominated by negative values and the net peak negative pressure is locally elevated (PMID 34576493). A building has at least two sets of pressures; without specifying the measurement point, the indicators are meaningless. Furthermore, $C_{p,min}$ is not measured directly but extrapolated from pressure time histories of finite length, and the choice of threshold will concretely alter the value (PMID 33642655). Each indicator carries its own extrapolation error; therefore, there is no common denominator for “who is most accurate.”

    2. Splitting the sample three ways collapses statistical power first. Testing must be stratified: Sky-Cutting uses only $K$, Acute-Angle uses only $C_{p,min}$, and Reverse-Arrow uses only RMS. Assuming a total of 300 cases, 100 per category, and an equal split across three severity levels, the number of “Great Misfortune” cases within each category drops to only a few dozen; distinguishing between an AUC of 0.7 and 0.5 at this scale is already at the edge of feasibility. If we split again for cross-regional validation, with only a dozen per cell, the resulting confidence intervals will be wide enough to overlap, making the ranking determined largely by sampling noise.

    This is the first trap: The negative answer to “which indicator best fits Great Misfortune” may not be rejected by the data, but rather by insufficient sample size. These two factors must be discussed separately; otherwise, a study with insufficient power will be misread as “wind pressure is unrelated to misfortune,” leading to the erroneous elimination of an entire line of inquiry. The established rule regarding the number of events required per variable in regression (PMID 12025266) still applies here, though the event count has been sliced by three layers of classification.

    3. Severity levels are ordinal and cannot be forced into binary AUC. Common practice involves collapsing ratings into “Great Misfortune vs. Not Great Misfortune,” but this collapse merges “Medium Misfortune” and “Small Misfortune”—the very tiers that serve as the practical boundary for “whether to act,” thereby erasing the most informative cut. Ordinal outcomes have their own metrics: the Multi-class Discrimination Index (PDI) (PMID 33866577) or the ordinal c-index. When using weighted kappa for consistency, one must explicitly state whether linear or quadratic weighting is used: quadratic weighting penalizes a one-tier difference less severely than a two-tier difference (PMID 25510263), which corresponds precisely to the fact that misclassifying “Great Misfortune” as “Small Misfortune” is far more severe than misclassifying it as “Medium Misfortune.” Weighting schemes are not technical details but value judgments; they must be fixed in the methodology. The same dataset can yield three different discriminative powers depending on whether covariates are treated as confounders, subjects, or incremental sources (PMID 20046933).

    4. The “score increase” in joint models may stem from double counting. This logic holds in medicine: three inflammatory markers have individual AUCs of 0.766, 0.716, and 0.768, but jointly reach 0.894 (PMID 41929599), an increase of 0.13, because they follow different pathways. The three wind pressure indicators do not operate this way—$K$ and RMS are driven by the same turbulent field, and $C_{p,min}$ at acute angles and RMS also originate from the same layer of separated shear. First conduct a collinearity diagnosis, then run nested models. The criterion should be strict: If the relative AUC increment of the joint model compared to the strongest single indicator is less than 0.03 and the likelihood ratio test is not significant, then the three indicators represent a single dimension and should be merged into a “Comprehensive Wind Pressure Intensity Factor” before re-evaluating severity, rather than presenting three numbers as if they constitute three independent pieces of evidence.

    5. Noise on the panel side amplifies negative results. The consistency among Feng Shui masters is actually not poor: across 108 prohibitions, with 11 scholars and 67 practitioners, there were no significant differences across gender, profession, education level, or school (PMID 37089290). However, the same study admits that retest reliability is not ideal—judgments drift if asked again after half a year. This is measurement error, which systematically suppresses associations: if rating reliability is 0.6 and indicator reliability is 0.9, a true correlation of 0.5 would be observed as $0.5 \times \sqrt{0.6 \times 0.9} \approx 0.37$, dropping exactly below the threshold (error planning see PMID 33410311). One must perform attenuation correction and report two sets of numbers: If the corrected result becomes significant while the original was not, the original negative conclusion is void; if it remains non-significant after correction, the conclusion stands.

    6. Would rankings flip in a different city? All three indicators are functions of geometry and inflow; changing cities equals changing roughness, wind climate, and building density. Scaling experiments have thresholds: outdoor flow must enter the Reynolds number-independent regime, with a critical value around 27,000, judged by a relative change rate $\leq 5\%$ (PMID 34659649). Without meeting this threshold, coefficients do not hold across scales. Therefore, cross-regional validation is not simply pooling data from two cities to calculate a total AUC, but first confirming that both regions meet the threshold. If $C_{p,min}$ is strongest in City A and RMS is strongest in City B, then there is no universal “indicator best fitting Great Misfortune,” only “which indicator fits best under what wind climates.”

    Therefore, the true output of this paper is not who wins, but four gates: whether power is sufficient after stratification, whether ordinal metrics are applied correctly, whether joint increments are artifacts of collinearity, and whether rating noise has been corrected. If any one gate is not passed, the answer is unreliable. This also explains why Feng Shui quantification has long stalled at “filing” rather than advancing to “discrimination”—data is not lacking; $K$, $C_{p,min}$, and RMS are all standard dimensionless quantities. What is lacking is this set of gates. Without them, anyone can claim victory with 300 cases.

    Evidence and Limitations

    Verified: $K$, $C_{p,min}$, and RMS are standard dimensionless quantities in wind engineering; internal and external pressure sets and measurement point locations determine coefficient values (PMID 34576493); peak extrema must be extrapolated from finite time histories, and threshold selection has objective criteria (PMID 33642655); scaling experiments have a Reynolds number-independent threshold, with an outdoor critical value around 27,000 and a criterion of relative change rate $\leq 5\%$ (PMID 34659649); internal consistency among Feng Shui ratings is good but retest reliability is not ideal (PMID 37089290); ordinal outcomes have dedicated metrics such as PDI (PMID 33866577), and weighting schemes for weighted kappa have clear semantics (PMID 25510263); multi-marker combinations can significantly improve discrimination (PMID 41929599); measurement error causes systematic attenuation of associations (PMID 33410311).

    Partially Verified: “Insufficient power after stratification” is based on hypothetical assumptions using 300 cases, not actual samples; the actual degree of collinearity among the three indicators has not been directly reported by wind tunnel or field measurements, inferred only from physical mechanisms; regression sample size rules (PMID 12025266) originally apply to binary outcomes and applying them to three-tier ordinal outcomes constitutes extrapolation.

    To Be Verified: ① The actual correlation matrix and variance inflation factors for the dimensionless indicators of the three Form Sha types; ② The true number of usable cases per Form Sha type after stratification; ③ The measured reliability coefficient for the Feng Shui masters’ task of rating Form Sha severity; ④ The stability of indicator ranking for discrimination across regions (different surface roughness, wind climates).

    Related Reading

    Evidence Level: B
    References: PMID 34576493, PMID 33642655, PMID 12025266, PMID 33866577, PMID 25510263, PMID 20046933, PMID 41929599, PMID 37089290, PMID 33410311, PMID 34659649


    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.

  • 墙拐到多少度才算凶:断点是数据里的,还是你刀切的

    Evidence level: B

    风水讲”过某某度则凶”,工程也真在间距上找到过临界坐标。但”找到断点”和”存在断点”是两件事,中间隔着一道检验。

    正文

    一、临界值这件事,风水讲得比工程还早。

    尖角煞、天斩煞、反弓煞,老一辈的讲法里都带度数:角越尖越凶、缝越窄越凶、弯越急越凶,而且往往给一条线——过了这条线才算数。这套说法在工程上不算离谱,因为分离流、再附着、涡脱落确实会在某些几何条件下换挡,而不是一路平推。错列切角柱体的实测就给出过一个很像”临界”的坐标:间距小于 (4.0, 4.0) 时干扰效应明显,大于这个坐标,主柱体与孤立柱体的气动系数和非高斯区分布基本一致,干扰因子趋近 1(PMID 37845264)。这就是一条真实的换挡线。

    问题在于,从”存在换挡”到”存在你要的那个断点”,中间还有一大段路。

    二、分段回归最大的习惯:你不查它有没有断点,它一定给你一个。

    把角度、缝宽比、曲率半径做梯度扫描,得到一串风压系数,然后丢进分段回归。这套模型的设计目的就是”找到拐点并把线掰弯”,所以哪怕真实关系是一条光滑曲线,它也会在残差最小的地方交出一个断点、一个斜率差、一个 p 值。不给它加一道”断点是否存在”的零假设检验,所谓”临界角 47 度”就是拟合器的出厂设定,不是自然的台阶。

    真要下结论,得把断点当成参数来估:给它一个区间,而不是一个数。贝叶斯分层分段回归的做法是同时输出断点的点估计和可信区间(PMID 28587592)——这一步不做,”临界 47 度”和”临界 43 度”就没有区别,而这两度足以决定一栋楼要不要改。

    三、最狠的一条反证:把观测误差放进模型,四个阈值剩一个。

    生态学有过一次干净利落的示范。研究者先用常规分段 logistic 回归,在鸟类占据率数据里找到四条”阈值关系”;再把”物种没被看见”这件事显式建进模型,重跑——四条里有三条的阈值效应可信区间包含了 0,也就是不再成立(PMID 22352169)。忽略观测过程,会系统性地把平滑关系读成台阶。

    形煞这边完全对应:风压系数是从有限长度的压力时程里外推出来的,峰值要靠 Gumbel 分布外推到标准风暴时长,而现行的 ASCE 7-10 表在低层建筑围护上被证实严重低估外压系数(PMID 33551545)。也就是说,纵轴的每一个点都带着外推误差和低估偏差。这种噪声进到分段回归里,最典型的后果就是——断点位置被拉走,斜率差被放大,显著性被凭空造出来。

    四、几何实测给出的形状,不是台阶,是拐弯。

    同一批研究把三种角部处理(切角、圆角、凹角)都用上了,修改率固定 10%,雷诺数 22000,间距比从 1.2 扫到 8。结果不是”角度越小越凶”这种单调关系:圆角的遮档效应最强,切角的放大效应最强,而凹角在临界间距下,相邻立面的风压概率密度出现双峰(PMID 39953100)。双峰这件事很要命——它说明在那个几何上,压力不是围绕一个均值抖,而是在两个状态之间跳。这时候谈”平均风压对应几级凶”,等于用一个数描述两个世界。

    曲率半径这一头更老实:检索下来,圆角半径与分离/再附着的连续扫描在 PubMed 里几乎没有可直接引用的定量报告。这一格必须承认空白,不能用尖角的结论顶替。

    五、大凶是尾部事件,就该用分位回归,不是均值回归。

    分段线性回归拟合的是条件均值,而”大凶”落在分布的尾巴上。生态学那边比较过分段线性回归和分段分位回归:分位回归给出的断点偏差更小、置信区间更窄、预测带也更窄,样本小、误差变异大的时候优势更明显(PMID 33304555)。道理不难懂——均值被中间那一大堆”无所谓”的样本拽住,尾部换挡的信息被稀释了。要在尾部找断点,就得在尾部拟合。

    六、还有一头噪声在评委身上。

    断点要和风水师评级对齐,就要算 kappa。现实值不漂亮:睡眠内镜两套评分系统、五位评分者盲法,Fleiss kappa 落在 0.32–0.59,序数部分用线性加权 kappa 只有 0.37(PMID 39612111)。这类量级叫”中等偏弱”,意味着同一栋楼换个人看,可能从”中凶”滑到”小凶”——正好跨过你要检验的那道断点。评级噪声不会均匀地糊在数据上,它会专门把断点抹平,或者推向样本最密集的那一段。

    所以这篇要立的不是某个角度数字,是三道门。

    第一道,断点存在性:先做零假设检验,报断点的可信区间,不报单一数值。第二道,误差入模:把外推误差、阈值选择、评级信度显式写进模型,看断点还在不在——三条阈值剩一条和剩三条,是两种完全不同的结论。第三道,尾部对齐:大凶用分位回归找,别用均值回归找。

    三道门过了,才轮到说”过多少度算凶”。过不了,那句”临界角 47 度”就该改写成一句更诚实的话:在 47 度附近,风压系数变化最快。 变化最快不等于跨了台阶。这两件事,被混着讲了很多年。

    证据与局限

    已验证: 几何参数确实改变干扰效应格局,并存在”特征坐标”式换挡——间距小于 (4.0, 4.0) 干扰显著、大于则趋近孤立状态(PMID 37845264);角部处理方式(切角/圆角/凹角)在 10% 修改率下产生方向相反的效应,凹角在临界间距出现双峰压力分布(PMID 39953100);峰值压力须由时程外推至标准风暴时长,ASCE 7-10 在低层围护上被证实严重低估(PMID 33551545);忽略观测不完全会把平滑关系误读为阈值关系——四条阈值中三条在建模后不再成立(PMID 22352169);分段分位回归在小样本、大误差变异下对断点的估计优于分段线性回归(PMID 33304555);评分者间一致性可低至中等偏弱,Fleiss kappa 0.32–0.59、加权 kappa 0.37(PMID 39612111);贝叶斯分段回归可给出断点点估计与可信区间(PMID 28587592)。

    部分验证: 上述几何结论来自缩尺 LES 与风洞,雷诺数 22000,与真实城市边界层仍有差距;”断点被噪声拉走”是从生态学结论外推,形煞数据尚未实测;尖角角度与缝宽比的连续扫描缺同批次公开数据。

    待验证: ①尖角角度(约 30°–150°)连续扫描下 Cp,min 的曲线形状——单调、非单调还是带平台;②缝宽比从宽到窄的收窄过程中是否存在第二次换挡而非单次;③曲率半径 r/B 从 0 到 0.5 的分离—再附着临界值;④上述三参数与风水师凶级评级对齐后,断点位置是否重合。

    相关阅读

    证据等级: B
    参考文献: PMID 37845264, PMID 39953100, PMID 33551545, PMID 22352169, PMID 33304555, PMID 39612111, PMID 28587592


    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。

    How Many Degrees Does a Wall Corner Become Evil: Is the Breakpoint in the Data or Where You Slice It

    Evidence level: B

    Feng Shui speaks of “becoming bad beyond X degrees,” and engineering has indeed found critical coordinates regarding spacing. However, “finding a threshold” and “the existence of a threshold” are two distinct matters separated by a rigorous test.

    Main Text

    I. The concept of a critical value predates engineering in Feng Shui discourse.

    Terms like sharp-corner Sha, sky-cut Sha, and reverse-arch Sha all include degrees in older methodologies: sharper angles are worse, narrower gaps are worse, and tighter curves are worse, often accompanied by a specific line—crossing which counts as “bad.” While not entirely unreasonable in engineering terms, separation flow, reattachment, and vortex shedding do indeed shift under certain geometric conditions rather than progressing linearly. Empirical tests on staggered cut-cornered prisms provided a coordinate resembling a “critical” point: interference effects become significant when spacing is less than (4.0, 4.0), whereas above this coordinate, the drag coefficients of the main pillar and isolated pillar align with non-Gaussian distributions, and the interference factor approaches 1 (PMID 37845264). This represents a genuine shift line.

    The issue lies in the gap between “the existence of a shift” and “the existence of the specific threshold you require.”

    II. The Habit of Piecewise Regression: If You Don’t Check for a Threshold, It Will Provide One.

    Conducting gradient scans on angles, gap-width ratios, and curvature radii yields a series of wind pressure coefficients, which are then fed into piecewise regression. The design purpose of this model is explicitly to “find the inflection point and bend the line.” Consequently, even if the true relationship is a smooth curve, it will output an inflection point, a slope difference, and a p-value at the location minimizing residuals. Without adding a null hypothesis test regarding “whether a threshold exists,” so-called “critical angles of 47 degrees” are merely factory settings of the fitter, not natural steps.

    To draw a valid conclusion, the threshold must be treated as a parameter to estimate: assign it an interval rather than a single number. Bayesian hierarchical piecewise regression simultaneously outputs the point estimate and credible interval for the threshold (PMID 28587592). Without this step, there is no distinction between “critical 47 degrees” and “critical 43 degrees”—a difference sufficient to determine whether a building requires modification.

    III. The Most Compelling Counter-Evidence: Including Observation Error in the Model Leaves Only One of Four Thresholds.

    Ecology provided a clean demonstration. Researchers first used standard piecewise logistic regression to identify four “threshold relationships” in bird occupancy data; then, they explicitly incorporated the fact that “species were not seen” into the model and re-ran it. Three of the four thresholds had credible intervals containing 0, meaning they no longer held (PMID 22352169). Ignoring the observation process systematically reads smooth relationships as steps.

    This corresponds fully in the context of Form Sha: wind pressure coefficients are extrapolated from finite-length pressure time histories, requiring Gumbel distribution extrapolation to standard storm durations. Current ASCE 7-10 tables are proven to severely underestimate external pressure coefficients on low-rise building envelopes (PMID 33551545). In other words, every point on the vertical axis carries extrapolation error and underestimation bias. When such noise enters piecewise regression, the most typical consequence is that the threshold position is pulled away, the slope difference is amplified, and significance is artificially created.

    IV. Shapes Provided by Geometric Measurements Are Not Steps, But Turns.

    The same studies utilized three corner treatments (cut-corner, rounded, concave) with a fixed modification rate of 10% and a Reynolds number of 22,000, scanning the spacing ratio from 1.2 to 8. The result was not a monotonic relationship such as “smaller angles are worse”: rounded corners exhibited the strongest shielding effect, cut corners the strongest amplification effect, while concave corners showed a bimodal probability density of wind pressure on adjacent facades at critical spacing (PMID 39953100). The bimodality is critical—it indicates that at that geometry, pressure does not fluctuate around a mean but jumps between two states. Discussing “average wind pressure corresponding to which severity level” at this point is equivalent to describing two worlds with a single number.

    On the curvature radius side, the situation is more straightforward: searching PubMed reveals almost no directly citable quantitative reports linking rounded corner radii to continuous scans of separation/reattachment. This gap must be acknowledged; conclusions from sharp angles cannot substitute here.

    V. Major Hazards Are Tail Events and Should Use Quantile Regression, Not Mean Regression.

    Piecewise linear regression fits conditional means, while “major hazards” fall on the tail of the distribution. Comparisons between piecewise linear and piecewise quantile regression in ecology show that quantile regression yields smaller threshold biases, narrower confidence intervals, and narrower prediction bands, offering distinct advantages with small samples and high error variance (PMID 33304555). The logic is simple: the mean is dragged by a large mass of “indifferent” samples, diluting information about tail shifts. To find thresholds in the tail, one must fit in the tail.

    VI. There Is Also Noise on the Side of the Evaluators.

    Aligning thresholds with Feng Shui ratings requires calculating kappa. The reality is not pretty: two sleep endoscopy scoring systems, five raters using blind methods, yielded a Fleiss kappa of 0.32–0.59, with ordinal data using linear-weighted kappa at only 0.37 (PMID 39612111). Such magnitudes are termed “moderate to weak,” meaning that viewing the same building with a different person might shift the rating from “moderately bad” to “slightly bad”—exactly crossing the threshold you are testing. Rating noise does not smear data uniformly; it specifically flattens thresholds or pushes them toward the most densely sampled segment.

    Therefore, this paper does not establish a specific angle number, but rather three gates.

    First gate, Threshold Existence: Perform a null hypothesis test first, report the credible interval for the threshold, and do not report a single value. Second gate, Error Inclusion: Explicitly write extrapolation error, threshold selection, and rating reliability into the model to see if the threshold persists—leaving three thresholds versus leaving one represents two fundamentally different conclusions. Third gate, Tail Alignment: Use quantile regression to find major hazards, not mean regression.

    Only after passing these three gates can one discuss “how many degrees constitute bad.” If not, the phrase “critical angle of 47 degrees” should be rewritten into a more honest statement: Near 47 degrees, the wind pressure coefficient changes most rapidly. Rapid change does not equal crossing a step. These two matters have been conflated for many years.

    Evidence and Limitations

    Verified: Geometric parameters indeed alter interference effect patterns and exhibit “characteristic coordinate” shifts—interference is significant when spacing is less than (4.0, 4.0) and approaches an isolated state otherwise (PMID 37845264); corner treatments (cut/rounded/concave) produce opposite directional effects at a 10% modification rate, with concave corners showing bimodal pressure distributions at critical spacing (PMID 39953100); peak pressures require extrapolation from time histories to standard storm durations, and ASCE 7-10 is proven to severely underestimate external pressure coefficients on low-rise envelopes (PMID 33551545); ignoring incomplete observation misreads smooth relationships as threshold relationships—three of four thresholds cease to hold after modeling (PMID 22352169); piecewise quantile regression estimates thresholds better than piecewise linear regression under small samples and high error variance (PMID 33304555); inter-rater reliability can be as low as moderate to weak, with Fleiss kappa at 0.32–0.59 and weighted kappa at 0.37 (PMID 39612111); Bayesian piecewise regression can provide point estimates and credible intervals for thresholds (PMID 28587592).

    Partially Verified: The above geometric conclusions derive from scaled LES and wind tunnels at Reynolds number 22,000, which still differ from real urban boundary layers; “thresholds pulled by noise” is extrapolated from ecological conclusions, and Form Sha data has not yet been empirically measured; continuous scans of sharp corner angles and gap-width ratios lack concurrent public data.

    To Be Verified: ① The curve shape of $C_{p,min}$ under continuous scanning of sharp corner angles (approx. 30°–150°)—monotonic, non-monotonic, or with a plateau; ② Whether a second shift occurs during the narrowing process of gap-width ratios from wide to narrow rather than a single shift; ③ The critical value for separation-reattachment as curvature radius $r/B$ ranges from 0 to 0.5; ④ Whether the threshold positions coincide after aligning the above three parameters with Feng Shui severity ratings.

    Related Reading

    Evidence Level: B
    References: PMID 37845264, PMID 39953100, PMID 33551545, PMID 22352169, PMID 33304555, PMID 39612111, PMID 28587592


    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.

  • 把照片涂成灰白体块之后:风水师还认得出凶吗

    Evidence level: B

    前两篇在问”形煞点位和风压异常对不对得上”。这一篇问一个更前置的问题:风水师评级时,判的是那个形,还是那个”看起来就像凶宅”的整体印象?

    正文

    前面已经算过重合度、也讨论过三类异常谁在背锅。但那两步都建立在一个默认前提上:风水师的评级是对几何形状的判断。这个前提从来没被单独检验过。实验室里检验它的办法只有一个——把形以外的一切都拿走,看还剩下多少判准能力。

    先划清什么能遮、什么不能遮。 风水讲”形势”,形是体块轮廓,势是走向与气势。可遮的东西包括:材质颜色、招牌文字、周边可比对的楼、”某某小区”这类地名信息、拍摄者标注。不能遮的是轮廓本身——遮了就没题了。所以遮挡版的做法是:统一灰白无纹理渲染、统一天空、去掉邻栋标识、只留体块与开窗的几何。这不是削弱刺激,是还原风水师自己声称的判读对象。

    暗示效应有硬证据。 一项双臂设计(N=83 与 N=70)让职业安全监察员看同一处工地,唯一改动是一段与检查任务无关的”公司安全历史”背景信息,评级就出现系统性偏移,而监察员本人普遍否认受到影响(PMID 34092303)。另一项研究找 149 名持照心理师,发现一旦形成初步诊断假设,就倾向继续搜寻证实性信息,即使给了多次机会(PMID 39298184)。综述层面,这类偏差在法庭科学、情报分析、医学与司法判断中已成共识(PMID 35963213)。

    但”遮挡后掉分”不等于暗示。 这可能是真信息被拿掉了——材质、周边、尺度感本来就在风水判读里。所以关键不是绝对变化,而是差值中的差值:专家的降幅减去新手的降幅。专家掉得多、新手掉得少,说明掉的那部分正是”专业解读依赖语境”的部分,这才是暗示;两者掉一样多,只是信息量减少,与专业性无关。

    度量要选对。 AUC 只说排序能力,说不清”哪一档被挪了”。经典框架讲得清楚:AUC 是判别,净重分类指数 NRI 管重分类,IDI 管整体改善,三者回答不同问题(PMID 20010215)。这里该用 NRI 的逆向读法——遮挡后有多少”大凶”被降级、多少”小凶”被升级,两个方向分开报。只报 AUC,会把方向相反的挪动抵消成”无差异”的假象。

    一致性是第三条线。 一项让 11 位专家盲评单被试图谱的研究发现,专家之间的评读一致性并不理想,这也正是呼吁同时报告效应量的原因(PMID 39052399)。另一项用 Bradley-Terry 模型分解 1020 名学生的图像排序,发现用时、年龄、自评兴趣都显著改变排序,专家与新手的注视时长也不同(PMID 36118447)。含义很直接:评级本身带噪声。若遮挡后专家间 kappa 不降反升,说明原本的分歧来自对语境的解读分歧,而非对形的判读分歧——这比 AUC 变化更能说明问题。

    设计上还有两个坑。 一是顺序:同一人先看完整版再看遮挡版会锚定,必须随机分组、每人只评一次、两个版本由不同的人评。二是”只给轮廓”本身有损:枕颞皮层对物体形状的编码依赖轮廓、曲率、中轴三者的正交与共享成分(PMID 33052726),灰白体块若把曲率与中轴也抹平,降掉的就是形本身。所以遮挡版必须保留轮廓、曲率、中轴,只去纹理与颜色。

    文化特异性不能当常量。 中国被试对建筑的美学判断里,曲线轮廓的偏好与层高、开敞度存在交互,只在低层高、封闭空间中才偏好曲线(PMID 35381002)。同一几何在不同语境下评级不同,这本身就是语境效应的正面证据——也因此不能把某一群体的偏好当普适参数塞进模型。

    判决实验骨架。 ①双版本刺激:完整照片与灰白体块版,保留轮廓曲率中轴,去纹理颜色标识。②随机分组:专家、新手各半,每人只评一版,1-5 级凶应。③主分析:ΔAUC 与双向 NRI。④关键对比:ΔΔ 等于专家降幅减新手降幅,报 bootstrap 区间。⑤合规:报 kappa 与 QUADAS-C 式偏倚条目(PMID 34698503)。

    三条判据。 若 ΔΔ 区间不含 0 且专家降幅更大——存在语境暗示,须声明评级含语境成分,不可外推为纯几何判读。若 ΔΔ 区间含 0、且两版 AUC 都在 0.7 以上——形本身可判读,暗示不成立,这是最有利于传统的结论。若两版 AUC 都不超过 0.5——连完整信息都判不准,该议题应退出。

    落到风水上。 风水师最常讲的一句辩护是”我看的是形,不是照片”。这个实验正是检验这句话:遮掉一切只留形还能判准,这句话就成立,”峦头”确实有一套独立于语境的判读能力;判准度依赖语境,那风水判读就和刑侦、放射、安监一样,是会被地名、传闻、人情影响的专业判断。这不丢人,医学影像也这样,但必须承认、并且做盲法。真正该问的从来不是”灵不灵”,而是”在哪一层灵”。

    证据与局限

    已验证:与任务无关的背景信息可系统性偏移专业人员的评级,且当事人普遍无察觉(PMID 34092303);持照心理师在形成假设后表现稳定的证实性信息搜寻(PMID 39298184);跨领域专家判断偏差已成方法学共识(PMID 35963213);专家盲评一致性不理想、需并列报告效应量(PMID 39052399);评级受用时、年龄、兴趣与专业度影响(PMID 36118447);形状编码依赖轮廓、曲率、中轴的正交与共享成分(PMID 33052726);中国被试的建筑美学判断存在轮廓×层高×开敞度交互(PMID 35381002);AUC、NRI、IDI 各答不同问题(PMID 20010215);比较性诊断准确性研究的偏倚评估框架(PMID 34698503)。

    部分验证:暗示效应在安监、心理诊断、法庭科学中已证实,但在风水形煞评级上无任何直接研究;”灰白体块呈现能保留形煞判读所需信息”是基于形状编码文献的推断,未做过刺激等价性检验。

    待验证:①遮挡版的刺激等价性(专家是否认为该版本仍”能看出形”);②ΔΔ 的实际量级与方向;③遮挡后专家间 kappa 是升是降;④地名、传闻、业主口碑在真实判读中的权重。

    相关阅读

    证据等级: B(暗示效应、专家判断偏差、形状编码、重分类度量均有 PubMed 直接支撑;”在形煞评级上做遮挡对照”本身无先例,属研究设计提案)
    参考文献: PMID 34092303, PMID 39298184, PMID 35963213, PMID 39052399, PMID 36118447, PMID 33052726, PMID 35381002, PMID 20010215, PMID 34698503


    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。

    After Painting Photos into Gray-White Blocks: Can a Feng Shie Master Still Identify Misfortune?

    Evidence level: B

    The previous articles asked whether “form sha locations align with wind pressure anomalies.” This article addresses a more upstream question: When Feng Shui practitioners are rated, are they judging the specific form, or the overall impression that “looks like a凶宅” (凶宅)?

    Main Text

    We have already calculated overlap and discussed which of the three anomalies bears the blame. However, both steps rely on a default assumption: that practitioner ratings are judgments of geometric shape. This assumption has never been tested in isolation. The only way to test it in a lab is to strip away everything else and see how much discriminative power remains.

    First, clarify what can be obscured and what cannot. Feng Shui discusses “situation” (Ling), where Xing (form) is the mass outline and Shi (trend) is the direction and momentum. Obstructible elements include: material color, signage text, comparable surrounding buildings, place names like “XX Community,” and photographer annotations. What cannot be obscured is the outline itself—obscuring it renders the task moot. Therefore, the occlusion method involves uniform gray-white rendering without texture, a unified sky, removal of neighboring building identifiers, and retention only of mass geometry and window openings. This is not weakening the stimulus; it is restoring the object practitioners claim to read.

    Suggestion effects have hard evidence. One study using a between-subjects design (N=83 vs. N=70) had safety inspectors view the same construction site; the only change was a background segment unrelated to the inspection task regarding “company safety history.” Ratings showed systematic shifts, yet inspectors generally denied being affected (PMID 34092303). Another study with 149 licensed psychologists found that once a preliminary diagnostic hypothesis formed, they tended to continue seeking confirmatory information even given multiple opportunities (PMID 39298184). At the review level, such biases are established consensus in forensic science, intelligence analysis, medicine, and judicial judgment (PMID 35963213).

    However, “score drops after occlusion” does not equal suggestion. This could simply mean true information was removed—material, surroundings, and sense of scale are often part of Feng Shui interpretation. Therefore, the key is not absolute change, but the difference in differences: the expert’s drop minus the novice’s drop. If experts drop significantly more than novices, the lost portion reflects “professional interpretation dependent on context,” which constitutes suggestion. If both drop equally, it is merely reduced information volume, unrelated to expertise.

    Metrics must be selected correctly. AUC only indicates ranking ability; it does not clarify “which tier was shifted.” The classic framework clarifies this: AUC measures discrimination, Net Reclassification Index (NRI) handles reclassification, and IDI addresses overall improvement; they answer different questions (PMID 20010215). Here, we should read NRI inversely—reporting separately how many “major sha” were downgraded and how many “minor sha” were upgraded after occlusion. Reporting only AUC would cancel out opposing shifts, creating a false impression of “no difference.”

    Consistency is the third line. One study where 11 experts blind-rated single-subject maps found inter-expert consistency was not ideal, which is precisely why effect sizes should be reported concurrently (PMID 39052399). Another study using a Bradley-Terry model to decompose image sorting from 1,020 students found that time spent, age, and self-rated interest all significantly altered sorting; experts and novices also differed in fixation duration (PMID 36118447). The implication is direct: ratings inherently carry noise. If inter-expert kappa increases after occlusion, it suggests original divergence stemmed from disagreements on contextual interpretation rather than shape judgment—this speaks more to the issue than AUC changes.

    Two pitfalls remain in design. First, order: viewing the full version before the occluded version creates anchoring; randomization is required, each person rates only one version, and versions are rated by different people. Second, “giving only outlines” itself causes loss: the occipitotempal cortex encodes object shape relying on orthogonal and shared components of outline, curvature, and axis (PMID 33052726). If gray-white blocks flatten curvature and axis, what is lost is the form itself. Thus, occluded versions must retain outline, curvature, and axis, removing only texture and color.

    Cultural specificity cannot be treated as a constant. Chinese participants showed interaction between curve outline preference and floor height/openness in architectural aesthetic judgments, preferring curves only in low-height, enclosed spaces (PMID 35381002). The same geometry receives different ratings in different contexts; this is positive evidence of context effects—therefore, one group’s preference cannot be plugged into models as a universal parameter.

    Judgment experiment skeleton. ① Dual-stimulus versions: full photo and gray-white block version, retaining outline/curvature/axis while removing texture/color/identifiers. ② Randomized groups: half experts, half novices; each person rates only one version on a 1-5 severity scale. ③ Primary analysis: ΔAUC and bidirectional NRI. ④ Key comparison: ΔΔ equals expert drop minus novice drop; report bootstrap intervals. ⑤ Compliance: report kappa and QUADAS-C style bias items (PMID 34698503).

    Three criteria. If ΔΔ interval excludes 0 and expert drop is larger—contextual suggestion exists; rating must be declared as containing context, not extrapolated as pure geometric judgment. If ΔΔ interval includes 0 and both versions have AUC > 0.7—form itself is readable; suggestion does not hold; this is the most favorable conclusion for tradition. If both versions have AUC ≤ 0.5—neither full nor stripped information is readable; the topic should step back.

    Applying to Feng Shui. The most common defense practitioners offer is “I read the form, not the photo.” This experiment tests that statement: if accuracy remains after stripping everything to leave only form, the statement holds, and Luan Tou indeed possesses interpretive capability independent of context. If accuracy depends on context, Feng Shui judgment is a professional judgment susceptible to place names, rumors, and personal connections—like criminal investigation, radiology, or safety inspection. This is not shameful; medical imaging works similarly—but it must be acknowledged and conducted with blinding. The question that truly matters is never “is it effective,” but “at which level is it effective.”

    Evidence and Limitations

    Verified: Background information unrelated to the task can systematically shift professional ratings, with subjects generally unaware (PMID 34092303); licensed psychologists show stable confirmatory information seeking after hypothesis formation (PMID 39298184); cross-domain expert judgment biases are methodological consensus (PMID 35963213); expert blind-rating consistency is suboptimal, requiring concurrent effect size reporting (PMID 39052399); ratings are influenced by time, age, interest, and expertise (PMID 36118447); shape encoding relies on orthogonal and shared components of outline, curvature, and axis (PMID 33052726); Chinese participants show interaction between outline × floor height × openness in architectural aesthetic judgments (PMID 35381002); AUC, NRI, and IDI answer different questions (PMID 20010215); bias assessment framework for comparative diagnostic accuracy studies (PMID 34698503).

    Partially Verified: Suggestion effects are confirmed in safety inspection, psychological diagnosis, and forensic science, but there is no direct research on Feng Shui form sha ratings; the inference that “gray-white blocks retain information required for sha judgment” is based on shape encoding literature without stimulus equivalence testing.

    To Be Verified: ① Stimulus equivalence of occluded versions (do experts believe this version still “shows the form”); ② Actual magnitude and direction of ΔΔ; ③ Whether inter-expert kappa increases or decreases after occlusion; ④ Weight of place names, rumors, and owner reputation in real-world judgments.

    Related Reading

    Evidence Level: B (Suggestion effects, expert judgment bias, shape encoding, and reclassification metrics all have direct PubMed support; “occlusion control specifically for form sha ratings” has no precedent, constituting a research design proposal)
    References: PMID 34092303, PMID 39298184, PMID 35963213, PMID 39052399, PMID 36118447, PMID 33052726, PMID 35381002, PMID 20010215, PMID 34698503


    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.

  • 三类风压异常谁在背锅:共线栅格上的贡献分解为何做不出来

    Evidence level: B

    想给”极值区、加速区、脉动区”排名,先得接受一个事实:在建筑角部,这三样东西常常同时出现、同时消失,本来就分不开。

    正文

    上一篇把风水师的形煞点位与风压异常带叠在一张图上,问重合度是否高于随机。这一篇问下一步:既然重合了,是哪一类异常在起作用——Cp 极值区的强负压,加速比超过 1.5 的高速带,还是脉动峰值区的湍流冲击?很多研究到这里就顺手跑个回归,把三类异常当三个自变量,看谁系数大、谁 p 值小,把系数当贡献报出去。这个动作在多数情况下无效。

    三类异常物理同源。 它们不是三种独立现象,而是同一股分离流在三个通道上的读数。方形建筑角部发生流动分离,剪切层卷起成涡,涡脱落在侧壁造成强负压——这是 Cp 极值;同一股被挤压的流动绕过角部进入通道,形成加速比高值区;涡脱落的间歇性让压力信号变非高斯、尾部变厚——这是脉动峰值。一篇 2026 年风洞研究说得很直白:130 个工况下高斯区集中在仰角 60 度以下的中心,外围因涡脱落呈非高斯(PMID 42297878)。即脉动区的判据(非高斯)与极值区的成因(涡脱落)指向同一过程。同源,空间上就必然共线,共线到无法用常规回归拆开。

    共线让系数分解变成随机数。 相关预测变量下可用的相对重要性方法至少有四种:Shapley 值分解、Genizi 法、相对权重、优势分析(PMID 42375422)。2026 年那篇基于鲍鱼数据的比较研究给了关键细节:同一份数据、同一个模型,PMVD 显示各变量重要性重叠,Genizi 却给出清晰分离,区间宽度也不同。同一问题、同一份数据,结论随尺子变。这不是方法不成熟,而是变量相关时”独立贡献”本身没有唯一解。广义可加模型更明显:生态学把这种共线性叫 concurvity,通行做法是按”平均共享方差”把总 R² 摊给各变量(PMID 39280972)。”摊”字是关键——分配规则人为约定,换摊法就换排名。

    随机森林也救不了。 2010 年那篇专研置换重要性的模拟研究结论分情况:变量既相关又真有信号时,非条件置换会把更大份额分给相关的那一组,条件重要性会压低相关变量得分,缩放重要性在有信号与无信号下都明显有偏(PMID 20187966)。更早一项模拟发现基尼重要性在相关下偏倚、偏倚取决于相关强度与数量,置换型更稳(PMID 19460890)。合起来:重要性指标不中立,自带倾向;选非条件置换,等于默认”相关的一组一起加分”。

    空间数据还多一层。 上面讲的是变量之间相关,栅格数据另有位置之间相关。2018 年一篇方法学研究用模拟证明:存在空间自相关时,常规双置换检验会抬高一类错误率,本不显著的关联被误判为显著;作者用 Moran 谱随机化保留原有空间结构,才把错误率控住(PMID 30289571)。这条格外致命:形煞点位成片出现(一栋楼的角部、一条街的夹缝),异常带也成片,两者的”重合”很容易只是因为都成片。

    换三个能答的问法。 原问题无解,可换成:增量(加第三类带来多少额外解释力)、分层(先筛仅含一类的格点再比命中率)、可辨识性(量化共线,报最小可分辨的贡献差)。

    判决实验骨架。 ①取子集:筛出仅含一类异常带的格点,若某类剩余不足总量 15%,退出排名。②多方法并列:同份数据同跑 Shapley、Genizi、相对权重、优势分析、条件置换重要性,五种排名并列报告。③空间零模型:不用简单随机置换,改 Moran 谱随机化保留评级点位的聚集结构(PMID 30289571)。④增量检验:以两类为基线逐个加入第三类,报 ΔAUC 及其在空间零模型下的置信区间,用区间而非 p 值(PMID 29906269)。⑤透明化:报方差膨胀因子、两两空间相关、每种方法排名在 1000 次 bootstrap 下的翻转频率。

    三条判据。 若五种排名一致且增量在空间零模型下仍显著——可报”某一类为主”,须注明结论对方法选择不敏感,这是最强情形,实际很少见。若排名随方法变化(Genizi 与 PMVD 分道扬镳,已有先例)——结论只能是”三类在此数据上不可分辨”,改报合并指标。若某类筛后不足 15%——不作独立因子,只作描述。三种里中间那种最可能也最有用:它把”哪个最凶”换成了”这个尺度上问不出答案”。

    落到形煞上。 风水对形煞是分条目的:尖角煞、天斩煞、反弓煞,各有各的凶应,各有各的化解。这套分类隐含一个假设——不同形状制造不同性质的麻烦,所以要用不同方法对付。风工程给的图像不同:不同形状制造的是同一股分离流的不同强度与位置,Cp 极值、加速、脉动是这股流的三个读数,不是三种病。若判决落在”不可分辨”,形煞的条目化分类就更接近一套便于传承的形状索引,而非病因学分类——它告诉你去看这个角,不告诉你这个角以哪种方式伤人。这个区分不是贬低传统,是把传统里可靠的部分(哪里该看)与过度延伸的部分(会怎么伤人)分开。古人讲”峦头一诀,百形皆应”,说的是看形方法通用;后人读成”每一形各有各的应”,是从方法跨到了病因,物理上没有支撑。

    证据与局限

    已验证:相关预测变量下相对重要性有多种分解方法且结论不一致(PMID 42375422,PMVD 显示重叠、Genizi 显示清晰分离);GAM 中共曲性需按平均共享方差摊分(PMID 39280972);随机森林非条件置换在相关且真有信号时把份额分给相关组、条件重要性压低相关变量、缩放重要性有偏(PMID 20187966);基尼重要性在相关下偏倚、置换型更稳(PMID 19460890);空间自相关抬高双置换检验一类错误率、Moran 谱随机化可控制(PMID 30289571);风压脉动区因涡脱落呈非高斯、与极值区同源(PMID 42297878);建筑凹角涡系同时决定通风与风舒适(PMID 32287964);转角优化改变双塔干扰效应(PMID 39953100)。

    部分验证:”三类异常带在真实建筑栅格上达到何种共线水平”无直接测量;”贡献分解在此共线水平下是否可辨识”是方法学外推的判断,未在形煞数据上验证。

    待验证:①三类异常两两空间相关的实际量级(预期 r>0.6);②筛出单类子集后各类剩余样本量是否够支撑分层比较;③五种分解方法的排名在 bootstrap 下的翻转频率;④控制空间自相关后,评级-异常重合度是否仍显著。

    相关阅读

    证据等级: B(贡献分解方法学、空间自相关对检验的影响、风压异常带同源性均有 PubMed 直接支撑;”在形煞栅格上做贡献分解”本身无先例,属研究设计提案)
    参考文献: PMID 42375422, PMID 39280972, PMID 20187966, PMID 19460890, PMID 30289571, PMID 42297878, PMID 32287964, PMID 39953100, PMID 29906269


    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。

    Who Is Taking the Blame for Three Types of Wind Pressure Anomalies: Why Contribution Decomposition on Co-linear Grids Cannot Be Done

    Evidence level: B

    To rank “extreme zones, acceleration zones, and pulsation zones,” one must first accept a fact: at building corners, these three phenomena frequently co-occur and vanish simultaneously; they are inherently inseparable.

    Main Text

    The previous article overlaid Feng Shui practitioners’ Form Sha locations with wind pressure anomaly bands and asked whether the overlap exceeded random chance. This article asks the next step: given the overlap, which specific anomaly is driving the effect—the strong negative pressure in the Cp extreme zone, the high-speed band with an acceleration ratio exceeding 1.5, or the turbulent impact in the pulsation peak zone? Many studies stop here and casually run a regression, treating the three anomalies as independent predictors to see which has the largest coefficient or smallest p-value, then report the coefficients as “contributions.” This action is largely ineffective in most cases.

    Physical Homogeneity of the Three Anomalies. They are not three independent phenomena but rather readings from a single separated flow across three channels. At the corner of a square building, flow separation occurs; the shear layer rolls into vortices, and vortex shedding creates strong negative pressure against the side wall—this is the Cp extreme. The same compressed flow bypasses the corner and enters a channel, forming a high acceleration ratio zone; the intermittent nature of vortex shedding makes the pressure signal non-Gaussian with fat tails—this is the pulsation peak. A 2026 wind tunnel study states plainly: under 130 test cases, Gaussian regions concentrated below a 60-degree elevation angle at the center, while peripheries became non-Gaussian due to vortex shedding (PMID 42297878). Thus, the criterion for the pulsation zone (non-Gaussian) and the cause of the extreme zone (vortex shedding) point to the same process. Being homogenous in origin means they are inevitably co-linear in space, co-linear to the point where conventional regression cannot separate them.

    Co-linearity Turns Coefficient Decomposition into Random Numbers. For relative importance methods under correlated predictors, there are at least four approaches: Shapley value decomposition, Genizi method, relative weights, and advantage analysis (PMID 42375422). A 2026 comparative study based on abalone data provided key details: with the same dataset and model, PMVD showed overlapping variable importance, while Genizi yielded clear separation with different interval widths. The same problem, same dataset, yet conclusions vary based on the metric used. This is not a lack of method maturity; when variables are correlated, “independent contribution” itself has no unique solution. Generalized Additive Models (GAMs) are even more explicit: ecology refers to this co-linearity as concurvity, and the standard practice is to allocate total $R^2$ among variables based on “average shared variance” (PMID 39280972). The word “allocate” is critical—the distribution rule is a human convention; changing the allocation method changes the ranking.

    Random Forests Cannot Save It. A 2010 simulation study on permutation importance concluded differently based on conditions: when variables are correlated yet truly signal-bearing, unconditional permutation assigns a larger share to the correlated group, while conditional importance suppresses scores for correlated variables, and scaled importance is significantly biased whether signal exists or not (PMID 20187966). An earlier simulation found that Gini importance is biased under correlation, with bias depending on correlation strength and count, whereas permutation-based methods are more stable (PMID 19460890). Combined: importance metrics are not neutral and carry inherent biases; selecting unconditional permutation defaults to “adding points to the correlated group together.”

    Spatial Data Adds Another Layer. The above discusses correlation between variables; grid data also involves correlation between locations. A 2018 methodological study using simulations proved that when spatial autocorrelation exists, conventional double permutation tests inflate the Type I error rate, falsely flagging non-significant associations as significant; authors controlled the error rate only by using Moran spectral randomization to preserve the original spatial structure (PMID 30289571). This is particularly fatal: Form Sha points appear in clusters (corners of buildings, gaps between streets), and anomaly bands also appear in clusters; their “overlap” is easily just because both are clustered.

    Three Alternative Questions. The original question has no solution; it can be reframed as: incremental gain (how much extra explanatory power does adding the third type bring?), stratification (screen for grid cells containing only one type first, then compare hit rates), or identifiability (quantify co-linearity and report the minimum resolvable contribution difference).

    Judgment Experiment Skeleton. ① Subset selection: screen for grid cells containing only one type of anomaly band; if a category remains below 15% of the total, exit the ranking. ② Parallel methods: run Shapley, Genizi, relative weights, advantage analysis, and conditional permutation importance on the same dataset simultaneously, reporting all five rankings side-by-side. ③ Spatial null model: do not use simple random permutation; instead, use Moran spectral randomization to preserve the clustering structure of rating points (PMID 30289571). ④ Incremental testing: add the third category one by one using two categories as baselines, report $\Delta$AUC and its confidence interval under the spatial null model, using intervals rather than p-values (PMID 29906269). ⑤ Transparency: report variance inflation factors, pairwise spatial correlations, and the flip frequency of rankings across 1000 bootstrap iterations.

    Three Criteria. If all five rankings are consistent and the increment remains significant under the spatial null model—report “one category dominates,” noting that the conclusion is insensitive to method choice. This is the strongest scenario but rare in practice. If rankings vary by method (e.g., Genizi and PMVD diverge, as seen in prior cases)—the conclusion must be “these three categories are indistinguishable on this dataset,” and report a combined metric instead. If a category falls below 15% after screening—do not treat it as an independent factor; only describe it. Among the three, the middle case is likely and most useful: it transforms “which is the worst” into “the question cannot be answered at this scale.”

    Applying to Form Sha. Feng Shui categorizes Form Sha into specific items: sharp corner Sha, sky-cut Sha, reverse bow Sha, each with its own凶应 (adverse response) and remediation. This classification implies an assumption that different shapes create different natures of trouble, requiring different countermeasures. Wind engineering provides a different image: different shapes produce different intensities and positions of the same separated flow. Cp extremes, acceleration, and pulsation are merely three readings of this single flow, not three distinct diseases. If the judgment falls on “indistinguishable,” the itemized classification of Form Sha approaches a shape index for transmission rather than an etiological classification—it tells you to look at this corner, not how that corner injures. This distinction does not demean tradition but separates the reliable parts of tradition (where to look) from over-extended parts (how it injures). Ancient texts say “One principle of landscape, all forms respond,” referring to universal viewing methods. Later readers interpret this as “each form has its own response,” crossing from method to etiology, which lacks physical support.

    Evidence and Limitations

    Verified: Multiple decomposition methods exist for relative importance under correlated predictors with inconsistent conclusions (PMID 42375422; PMVD shows overlap, Genizi shows clear separation); concurvity in GAMs requires allocation by average shared variance (PMID 39280972); in Random Forests, unconditional permutation assigns shares to correlated groups when signal exists, conditional importance suppresses correlated variables, and scaled importance is biased (PMID 20187966); Gini importance is biased under correlation with bias dependent on strength and count, while permutation is more stable (PMID 19460890); spatial autocorrelation inflates Type I error in double permutation tests, controllable via Moran spectral randomization (PMID 30289571); wind pressure pulsation zones become non-Gaussian due to vortex shedding and share origin with extreme zones (PMID 42297878); building re-entrant corners vortex systems simultaneously determine ventilation and wind comfort (PMID 32287964); corner optimization alters double-tower interference effects (PMID 39953100).

    Partially Verified: The degree of co-linearity reached by “three anomaly bands on real building grids” lacks direct measurement; whether “contribution decomposition is identifiable at this level of co-linearity” is a methodological extrapolation not verified on Form Sha data.

    To Be Verified: ① Actual magnitude of pairwise spatial correlation between the three anomaly types (expected $r > 0.6$); ② Whether sample sizes per category after screening single-type subsets are sufficient to support stratified comparisons; ③ Flip frequency of rankings across the five decomposition methods under bootstrap; ④ Whether rating-anomaly overlap remains significant after controlling for spatial autocorrelation.

    Related Reading

    Evidence Level: B (Methodological support for contribution decomposition, impact of spatial autocorrelation on testing, and homogeneity of wind pressure anomaly bands all have direct PubMed backing; “performing contribution decomposition on Form Sha grids” itself has no precedent and constitutes a research design proposal)
    References: PMID 42375422, PMID 39280972, PMID 20187966, PMID 19460890, PMID 30289571, PMID 42297878, PMID 32287964, PMID 39953100, PMID 29906269


    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.

  • 同煞不同灾该怎么验:敏感的人,还是被挑中的人?

    Evidence level: B

    「敏感的人受煞更重」是一个交互项,不是一句常识;而在自选改造的设计里,这个交互项可以被完全伪造出来。

    正文

    第一层:一句话翻译成一个式子,再拆出三个前提。 「敏感个体对化解更敏感」在统计上就是 症状变化 = a + b·Δ + c·敏感度 + d·(Δ×敏感度),要 d 显著为正。但这要求三件事同时成立:Δ 是外生的、敏感度是同一个人身上稳定的属性、结局的尺子前后刻度一致。缺一条,d 就可能纯属人造。而通行设计——自评量表分层 + 前后双阶段问卷 + PSG/HRV——恰好把三条全踩了。

    第二层:Δ 根本不是外生的。 谁会给房子做形态改造?被吵得最难受的那批人。于是 Δ 与基线症状天然正相关;基线症状高的人,无论干预真假都会往下走一截,这是回归均值。改造又由本人自选:愿意花钱动房子的人本来就相信「有煞」、相信「改了会好」。所谓「敏感组」,在这套设计里同时是高基线组、高期望组、最可能改造组。三者叠加,一个显著的正交互可零生理机制地长出来。在这套设计里,d 不可解释。

    第三层:敏感度的尺子,恰在最需要的那一端最钝。 中文简化版 MSSQ 给了硬数字:3111 人,MSSQ-S 内部一致性 0.900、重测 0.895,「轻度 vs 中度」分得开(P=0.006);但 MSSQ-L 在「中度 vs 重度」上 P=0.081,分不开(PMID 38858214)。而「重度敏感」正是本假说押注的那一端——尺子在这一端失效;它对症状严重度的预测准确率也只有 62.2%。VIMSSQ 短版因子分析:抽出「恶心相关」与「眼动相关」两个潜变量,与 MSSQ 相关仅 r=0.55(PMID 40432616)——两把尺子量的未必是同一个东西。再叠加二分化本身的信息损失与功效折损(PMID 41802432),分层先亏一次。

    第四层:结局的尺子会在干预后自己变,这条最致命。 前后比较的前提是「同一个人对同一份问卷的标尺不变」,这个前提被反复证伪。脊柱术后,报告改善者各维度权重重排——活动目标在移动(PMID 23519975)。前列腺癌治疗队列更直接:情绪与认知分量表出现重新校准,判断随时间变得更宽松,抵消了真实的负向变化(PMID 26882096)。落到形煞化解上:一个人花几万块动了房子,他对「睡得好不好」的判据会自然放宽;这种放宽在敏感者身上更大——他们更留意身体、更频繁自评、也更容易被说服。于是「敏感度×Δ」的交互,可以被「敏感度×校准漂移」完美伪造。

    第五层:期望是共同原因,不是噪声。 莱姆病持续症状的随机试验里,治疗前期望稳定预测了 14 周与 52 周的功能改善,与分组无关;治疗后「以为用的是抗生素而非安慰剂」也预测了短期改善(PMID 34031759)。抑郁治疗的纵向数据同向:期望越高,下一时点症状越低(PMID 30420001)。期望不是要扣掉的噪声,它本身就是一条因果通路。于是问题变成——若期望能产生改善、又和敏感度相关,就永远分不清 d 是生理交互,还是「敏感的人更容易被自己的期待说动」。

    第六层:功效。 交互项功效约为同等大小主效应的四分之一,亚组分析是假阳性、低功效、选择性报告的重灾区(PMID 42753990)。发表过的次声长期暴露 RCT 只有 38 人(PMID 39427080),连主效应都勉强,遑论交互。而该文结论恰恰是「未来应以敏感度标记分层」——一个被明确写下、从未被满足的缺口:它需要的样本量不是 38,是几百。路径两条:适应性设计,期中判断预测效应是否已足,不足则补样本并控一类错误(PMID 42706943);或个体数据汇总后用递归划分识别亚组,切点交给数据而非研究者预设(PMID 29383818)。

    第七层:判决实验长什么样。 ① 别让「谁改造」由本人决定——用实测 Δ 设一个预先登记的断点,跨过者进改造组、未跨者进对照,做断点回归(PMID 16579820)。这一步同时解掉自选择与回归均值:断点两侧基线连续,唯一不连续的是「改没改」。② 敏感度保持连续不切分,加法与乘法两尺度同报。③ 结局加锚:常规问卷之外加 then-test(改造后回评改造前的自己)与内标锚题,估计重新校准量。④ 期望单独测,作中介与共同原因同时入模。⑤ 盲法:改造由第三方施工,住户只知「做了两套方案中的一套」。判据三条:断点处 Δ 跳跃显著而交互不显著 ⇒ 效应普遍,个体差异是自选择造的;两者都显著且校准量不随敏感度变 ⇒ 交互初步可采信;交互显著但校准量随敏感度同步变大 ⇒ 判为报告偏移。

    易学落点: 风水最硬的一句经验是「同煞不同灾」——同一个尖角,一家人里有头疼的有无事的。这话作为观察是对的,个体差异确实存在。但传统从没让当事人自己报名说「我敏感」,它用的是外部标记:生辰、宅命、形貌、气色——古人知道自报不可信,硬性引入了一个不听话的第三方判据。今天的量化研究反而退了一步:自评量表分层、住户自己决定改不改、改完自己报好没好——三段全靠同一个人。这在方法学叫「共同方法」,在风水里叫「自己给自己看风水」。所以结论不是「同煞不同灾」错了,而是:要检验它,就不能让被检验的人同时当裁判。 真正接得住古人的做法,是先固定「谁被改造」,再问「谁受益更大」——顺序一反,问的就不是煞,是人情。

    证据与局限

    已验证: 敏感度量表的心理测量性能与其区分度上限(MSSQ 中文简化版 3111 人,MSSQ-L 在中重度间 P=0.081 不可分,个体预测准确率 62.2%,PMID 38858214);敏感度存在至少两个潜变量维度且不同量表相关仅 r=0.55(PMID 40432616);二分化连续变量导致信息损失与功效下降(PMID 41802432);亚组分析普遍低功效、易假阳性、存在选择性报告(PMID 42753990);干预后自我报告存在重排与重新校准两类响应偏移(PMID 23519975、PMID 26882096);治疗前期望独立预测症状改善(PMID 34031759、PMID 30420001);前后对照设计受自选择与回归均值污染,断点回归可作替代(PMID 16579820)。

    部分验证: 交互项”功效约为主效应四分之一”是通用经验法则而非本场景实测;次声长期暴露 RCT 的样本量证据(PMID 39427080,n=38)来自 6 Hz、80-90 dB 夜间暴露场景,与本篇 4-16 Hz 建筑微振动场景存在频段与剂量差异;该文明确提出”应按敏感度分层”,是本篇假说未被检验的直接旁证。

    待验证: ① 自选改造人群里,Δ 与基线症状的实际相关强度及由此产生的回归均值幅度;② 形煞化解场景下响应偏移的实测大小,是否真随敏感度同步增大;③ 断点回归在住宅改造中的可行性边界(Δ 是否可稳定测得足以设断点,取决于前篇 MDC 结论);④ 4-16 Hz 段”敏感度×Δ”交互的真实效应量,决定所需样本究竟是几百还是上千。

    相关阅读

    证据等级: B
    参考文献: PMID 38858214, PMID 40432616, PMID 41802432, PMID 42753990, PMID 23519975, PMID 26882096, PMID 34031759, PMID 30420001, PMID 16579820, PMID 39427080, PMID 42706943, PMID 29383818


    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。

    How to Verify Same Curse Different Calamity: Are Sensitive People or the Selected Ones Affected?

    Evidence level: B

    “Sensitive individuals suffer more from Sha” is an interaction term, not a common sense fact; within self-selected remediation designs, this interaction can be entirely fabricated.

    Main Text

    Layer 1: Translate into an equation, then extract three prerequisites. “Sensitive individuals are more sensitive to mitigation” statistically translates to $\text{Symptom Change} = a + b \cdot \Delta + c \cdot \text{Sensitivity} + d \cdot (\Delta \times \text{Sensitivity})$, requiring $d$ to be significantly positive. However, this requires three conditions to hold simultaneously: $\Delta$ must be exogenous (unrelated to who performed the remediation), sensitivity must measure a stable trait within the same individual, and the outcome scale must remain consistent across time points. Missing any one allows $d$ to be purely artifactual. The prevailing design—stratification via self-rated scales, pre-post dual-phase questionnaires, and PSG/HRV—accidentally satisfies all three.

    Layer 2: $\Delta$ is not exogenous. Who performs the morphological remediation on a house? Those who are most bothered by the noise. Thus, $\Delta$ is naturally positively correlated with baseline symptoms; individuals with high baseline symptoms will decline regardless of intervention truth due to regression to the mean. Since remediation is self-selected, those willing to spend money on house modifications already believe “there is Sha” and “it will get better.” The so-called “sensitive group” in this design simultaneously represents high baseline, high expectation, and the most likely remediation group. When these three factors overlap, a significant positive interaction can emerge without any physiological mechanism. In this design, $d$ is uninterpretable.

    Layer 3: The sensitivity scale is blundest exactly where it is needed most. The simplified Chinese MSSQ provides hard numbers: 3,111 participants, MSSQ-S internal consistency 0.900, retest 0.895, distinguishing “mild vs. moderate” ($P=0.006$); however, MSSQ-L fails to distinguish “moderate vs. severe” ($P=0.081$) (PMID 38858214). “Severe sensitivity” is precisely the endpoint this hypothesis bets on—the scale fails at this endpoint. Its prediction accuracy for individual symptom severity is only 62.2%. Furthermore, factor analysis of the VIMSSQ short version indicates sensitivity is not unidimensional: extracting two latent variables (“nausea-related” and “eye-tracking-related”) yields a correlation with MSSQ of only $r=0.55$ (PMID 40432616)—two scales may not be measuring the same thing. Adding the information loss and power reduction inherent in dichotomization (PMID 41802432), this stratification step incurs an initial loss.

    Layer 4: The outcome scale shifts itself after intervention, the most fatal flaw. The premise of pre-post comparison assumes “the same individual uses the same scale for the same questionnaire,” a premise repeatedly falsified. Spinal surgery follow-ups show that individuals reporting significant improvement undergo a reallocation of weights across dimensions—activity goals shift (PMID 23519975). Prostate cancer treatment cohorts show even more directly: emotion and cognitive scales undergo recalibration, with judgments becoming more lenient over time, offsetting true negative changes (PMID 26882096). Applied to Sha remediation: spending tens of thousands to modify a house naturally relaxes one’s criterion for “sleep quality”; this relaxation is greater in sensitive individuals—who pay more attention to their bodies, self-assess more frequently, and are more easily persuaded. Thus, the interaction “Sensitivity $\times$ $\Delta$” can be perfectly fabricated by “Sensitivity $\times$ Calibration Drift.”

    Layer 5: Expectation is a common cause, not noise. In randomized trials for Lyme disease chronic symptoms, pre-treatment expectation independently predicted functional improvement at 14 and 52 weeks regardless of group assignment; post-treatment, believing “one is using antibiotics rather than placebo” also predicted short-term improvement (PMID 34031759). Longitudinal data from depression treatment aligns similarly: higher expectation correlates with lower symptoms at the next time point; and “the greater the gap between expectation and actual severity, the more symptoms decrease subsequently” (PMID 30420001). Expectation is not noise to be subtracted but a genuine causal pathway. Therefore, the question becomes: if expectation itself generates improvement and correlates with sensitivity, can one ever distinguish whether $d$ represents a physiological interaction or “sensitive individuals are more easily moved by their own expectations”?

    Layer 6: Power. The power of an interaction term is approximately one-quarter that of an equally sized main effect, while subgroup analysis is a hotspot for false positives, low power, and selective reporting (PMID 42753990). Published RCTs on infrasound long-term exposure have only 38 participants (PMID 39427080)—barely sufficient for main effects, let alone interactions. Yet the authors’ own conclusion explicitly states “future studies should stratify by sensitivity”—a clearly written gap that has never been filled, requiring hundreds rather than 38 participants. Feasible paths include adaptive designs, where interim judgments determine if the predictive effect is sufficient to proceed to an extension phase with sample replenishment and control of Type I error (PMID 42706943); or individual data pooling followed by recursive partitioning for subgroup identification, handing the cutoff to data rather than researcher preset (PMID 29383818).

    Layer 7: What a judgment experiment looks like. ① Prevent self-selection from determining “who modifies”—use measured $\Delta$ to set a pre-registered breakpoint; those crossing enter the remediation group, those not crossing enter control, performing regression discontinuity (PMID 16579820). This simultaneously resolves self-selection and regression to the mean: baseline continuity exists on both sides of the breakpoint, with only “whether modified” being discontinuous. ② Maintain sensitivity as continuous, do not dichotomize, report both additive and multiplicative scales. ③ Add anchors to outcomes: alongside standard pre-post questionnaires, add then-tests (re-evaluating oneself post-remediation) and internal anchor items specifically estimating calibration drift. ④ Measure expectation separately, entering it as both mediator and common cause in the model. ⑤ Blinding: remediation performed by third parties; residents know only “one of two schemes was implemented.” Three criteria for judgment: significant $\Delta$ jump at breakpoint but non-significant interaction ⇒ effect is universal, individual differences created by self-selection; both significant with then-test calibration not varying by sensitivity ⇒ interaction can be tentatively accepted; significant interaction but calibration scales up synchronously with sensitivity ⇒ classified as reporting bias.

    Yijing Studies Application: The hardest empirical rule in Feng Shui is “Same Sha, Different Disasters”—the same sharp corner causes headaches for some family members but nothing for others. As observation, this is correct; individual differences do exist. However, tradition never allows participants to self-report “I am sensitive”; it uses external markers: birth charts, house destiny, form, complexion. In modern terms, ancients knew self-report is unreliable, so they rigidly introduced an uncooperative third-party criterion. Modern quantitative research actually regresses a step: self-rated scale stratification, residents deciding whether to modify and reporting improvement themselves—three segments all relying on the same person. Methodologically, this is “common method bias”; in Feng Shui, it is “looking at one’s own Feng Shui.” Thus, the conclusion is not that “Same Sha, Different Disasters” is wrong, but that: to verify it, the person being tested cannot also be the judge. To truly uphold ancient practice, one must first fix “who is modified,” then ask “who benefits more”—reversing the order changes the question from Sha to human relations.

    Evidence and Limitations

    Verified: Psychometric performance and discrimination limits of sensitivity scales (MSSQ Chinese simplified version with 3,111 participants; MSSQ-L indistinguishable between moderate and severe ($P=0.081$); individual prediction accuracy 62.2%) (PMID 38858214); existence of at least two latent variable dimensions for sensitivity with cross-scale correlation only $r=0.55$ (PMID 40432616); information loss and power reduction caused by dichotomizing continuous variables (PMID 41802432); subgroup analysis generally exhibiting low power, high false positive risk, and selective reporting (PMID 42753990); two types of response bias (reallocation and recalibration) in self-reported outcomes post-intervention (PMID 23519975, PMID 26882096); pre-treatment expectation independently predicting symptom improvement (PMID 34031759, PMID 30420001); pre-post control designs contaminated by self-selection and regression to the mean, with regression discontinuity as an alternative (PMID 16579820).

    Partially Verified: The rule “interaction power is about one-quarter of main effect” is a general heuristic not empirically tested in this specific scenario; the sample size evidence for infrasound long-term exposure RCT (PMID 39427080, n=38) derives from a 6 Hz, 80-90 dB nighttime exposure scenario, differing in frequency band and dose from the 4-16 Hz building micro-vibration scenario in this article; the explicit proposal to “stratify by sensitivity” serves as direct corroboration of this article’s hypothesis remaining untested.

    To Be Verified: ① Actual correlation strength between $\Delta$ and baseline symptoms in self-selected remediation populations, and the resulting magnitude of regression to the mean; ② Measured magnitude of response bias in Sha remediation scenarios, whether it truly scales synchronously with sensitivity; ③ Feasibility boundaries of regression discontinuity in residential renovation (whether $\Delta$ can be stably measured sufficiently to set a breakpoint, dependent on conclusions from the preceding MDC article); ④ True effect size of the “Sensitivity $\times$ $\Delta$” interaction in the 4-16 Hz range, determining whether required sample size is hundreds or thousands.

    Related Reading

    Evidence Level: B
    References: PMID 38858214, PMID 40432616, PMID 41802432, PMID 42753990, PMID 23519975, PMID 26882096, PMID 34031759, PMID 30420001, PMID 16579820, PMID 39427080, PMID 42706943, PMID 29383818


    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.

  • 改造前后差多少分贝才算数:风水「化解」要跨过的那道噪声底线

    Evidence level: B

    在问”化解有没有效”之前,有个更靠前的问题没人问:我们用来判断”变好了”的那把尺,本身有多准?

    正文

    第一层:把问题翻个面。 形煞化解的标准说法是”改完就顺了”。但这句话要能被检验,先得回答一个更靠前的问题:我们用来判断”顺了”的那把尺,本身有多准? 一间房改造前后各测一次低频声级,差 3 dB——这是改造的功劳,还是换了个测量日子、换了个测点,或者仪器本来就有这么大抖动?测量学里这个门槛叫最小可检出变化(MDC),也叫最小真实差异(SRD):在现有测量误差下,前后两次读数至少要差多少,我们才有 95% 的把握说这不是噪声(PMID 11822790)。

    第二层:这道门槛怎么算。 标准式是 MDC95 = 1.96 × √2 × SEM,SEM 是测量标准误。系数 √2 来自一个容易漏掉的事实:前后两次测量,两个数各自带一份误差,相减时误差不是抵掉而是叠加;1.96 是 95% 分位。这套口径在康复评定里已用到很细——手功能五项测试的 SRD、手持测力计的 MDC、功能量表的测量误差与最小重要变化,都按同一算法报(PMID 19261767、PMID 36118657、PMID 37606246)。但要留意:MDC 只管”是不是噪声”,不管”有没有用”——后者是另一个量。

    第三层:MDC 有个从不明说的前提——噪声得是稳定的。 上面那套把前后两次当成”条件相同、只差一个处理”。可形煞改造的前后测量,中间隔着几个月甚至一个季节。室内环境的季节变异是实测过的:VOC 浓度的空间与季节变化可使同一房间不同季节显著不同(PMID 27163962),室内灰尘细菌成分与麦角甾醇的重复性与季节变异同样明确(PMID 24992650、PMID 24883434)。当噪声本身在漂,MDC 就不是安全垫,而是一张写着上季度数字的旧收据。 更麻烦的是,噪声评价这行自己算过账:测量不确定度会直接决定噪声制图验证能否通过(PMID 32310126)——同一条限值,换个不确定度口径,结论就翻。

    第四层:低频段的误差大头不是仪器,是”你站在哪儿测”。 中高频房间里声场近似扩散,挪个位置差别不大;低频不是。混响室里 315 Hz 以下扩散场假设本身就不成立,必须换测量方法(PMID 41711454);小尺度混响室的隔声量会被试样尺寸与开口几何直接改写(PMID 42451325)。翻成现场语言:4-20 Hz 的驻波图样在房间里是一张固定的峰谷网,测点挪二十公分,读数变化可能比整次改造的效应还大。 而 MDC 公式里没有”位置”这一项——它把测量当成一个点,低频测量却从来不是一个点。

    第五层:为什么必须有”没改造的对照楼”。 有 MDC 还不够,因为还有回归均值:被改造的楼通常是被投诉”特别吵”才选中的,极端值下次测量天然往回走。Bland-Altman 框架讲得清楚,用差值均值与标准差定界,前提是差值里没有系统性成分(PMID 10501650、PMID 26110027);要估这个界,样本量还得单独算(PMID 27838682)。所以骨架只能是:同期、同季节、同仪器,改造楼与对照楼各一批,双测——对照楼不涨不落的那部分,就是必须扣掉的漂移。噪声暴露与烦恼反应的研究也提示:人对”变了多少”的反应依赖于变化前后各自的水平,而不只是变化量(PMID 22902956)。

    第六层:MDC 得自己算,不能抄。 骨密度这一行的教训最直白:厂家推荐的精度评估样本量往往不足,算出的最小显著变化不稳定,临床”监测”里冒出一堆假变化(PMID 16900302);精度误差还随测量条件系统性变化(PMID 17937536);LSC 必须与监测时间间隔成对给出——变化慢的指标,间隔太短永远测不出真变化(PMID 23452871、PMID 26032212)。对应到形煞:MDC 必须用本楼、本房间、本仪器的重测数据算,还要和”等多久才可能跨过 MDC”一起报。 纵向面板的样本量估算方法已经成熟(PMID 28957035),改造研究没有理由不算。

    第七层:判决实验长什么样。 ① ≥10 栋改造加 ≥10 栋对照,同季节双测,前后各 ≥3 个时点、跨 ≥1 个季节;② 每时点每房间 ≥3 次重复读数,测点、高度、朝向固定并拍照留档,同步记温湿度、外窗状态、HVAC 档位、在场人数;③ 方差分解出仪器/位置/日内/季节/模态五项贡献率;④ 用本楼重测数据算 MDC95 与监测时间间隔;⑤ Δ 先用对照楼漂移校正,再与 MDC95 比。判据三条:Δ 校正后低于 MDC95 ⇒ 不可判,须写”本设计分辨率不足”而非”无效”;落在 MDC95 与两倍之间 ⇒ 只能报”可检出但不稳健”,需延长观测;MDC95 里”位置”项贡献过半 ⇒ 该房间低频测量不能用于个体前后比较,只能比组均值。 干预评估还有一条老教训:前后对照不做时间序列,很容易把自然波动读成效果(PMID 39181016、PMID 32829342)。

    易学落点: 风水讲”化解”,从来附带一个”应期”——三年、五年、一个运。过去这被当成玄虚;换成测量学的话说,“应期”就是一个朴素的 MDC 直觉:知道自己这杆秤不准,就不敢报小变化。 骨密度领域把这个直觉写成了公式:最小显著变化与监测时间间隔必须同时给出(PMID 23452871)。古人的”三年不改”和现代测量的”观测间隔须长过 MDC 所需时间”,是同一条纪律的两种说法。 反过来看,今天”化解”营销最大的毛病恰好是把噪声当信号:季节漂了、测点挪了、极端值回落了,统统说成”布局起效”。真正的传统不是”改完就灵”,是”改完等应期”——它承认:噪声底下,沉默比报喜更专业。

    证据与局限

    已验证: 最小可检出变化(MDC/SRD)的计算口径及其与测量标准误的关系,在康复评定与功能量表领域有大量实证支撑(PMID 11822790、PMID 19261767、PMID 36118657、PMID 37606246);室内环境测量存在明确的空间与季节变异,重复性有限(PMID 27163962、PMID 24992650、PMID 24883434);测量不确定度足以改变噪声评价的通过与否(PMID 32310126);低频段扩散场假设失效,测量几何直接改写结果(PMID 41711454、PMID 42451325);精度评估样本量不足会导致最小显著变化不稳定、监测频出假变化(PMID 16900302、PMID 17937536、PMID 23452871、PMID 26032212)。

    部分验证: 上述方法学结论均来自医学测量、职业卫生与建筑声学各自领域,尚无一项研究把它整体搬到”形煞改造前后”这一具体场景;本设计中的阈值(MDC95、两倍 MDC95、位置贡献过半)是引用通用惯例设定的待标定初值,不是实测结论。

    待验证: ① 单栋住宅 4-20 Hz 段重测噪声底的实际量级(SD 与 ICC),本楼与楼间差多少;② 五项方差贡献(仪器/位置/日内/季节/模态)在真实住宅里的排序,是否真由”位置”占主导;③ 改造后 Δ 的分布是否真能跨过 MDC95,跨过的比例有多大;④ “应期”式的长观测窗(≥1 年)是否显著抬高可检出率。

    相关阅读

    证据等级: B
    参考文献: PMID 11822790, PMID 19261767, PMID 36118657, PMID 37606246, PMID 27163962, PMID 24992650, PMID 24883434, PMID 32310126, PMID 41711454, PMID 42451325, PMID 10501650, PMID 26110027, PMID 27838682, PMID 22902956, PMID 16900302, PMID 17937536, PMID 23452871, PMID 26032212, PMID 28957035, PMID 39181016, PMID 32829342


    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。

    How Many Decibels Matter Before and After Renovation: The Noise Threshold Feng Shui \”Resolution\” Must Cross

    Evidence level: B

    Before asking whether “mitigation works,” there is a more upstream question rarely asked: How accurate is the ruler we use to judge that things have “improved”?

    Main Text

    Layer 1: Flip the Question. The standard claim for mitigating Form Sha is “it becomes harmonious immediately after renovation.” But this claim can only be verified if we first answer a more upstream question: How accurate is the ruler we use to judge that it has become “harmonious”? Measuring low-frequency sound levels before and after a renovation, with a difference of 3 dB—is this due to the renovation, or a different measurement day, a different measurement point, or inherent instrument jitter? In metrology, this threshold is called the Minimum Detectable Change (MDC), also known as the Minimum Significant Real Difference (SRD): Under existing measurement error, the difference between two readings must exceed a certain amount for us to have 95% confidence that it is not noise (PMID 11822790).

    Layer 2: How to Calculate This Threshold. The standard formula is MDC95 = 1.96 × √2 × SEM, where SEM is the standard error of measurement. The coefficient √2 stems from a fact often overlooked: in two successive measurements, each number carries its own error, which adds up rather than cancels out when subtracted; 1.96 is the 95th percentile. This framework is already applied in detail in rehabilitation assessment—SRDs for the five hand function tests, MDCs for handheld dynamometers, and measurement errors with minimum important changes for functional scales all follow the same algorithm (PMID 19261767, PMID 36118657, PMID 37606246). However, note: MDC only addresses “is it noise,” not “is it meaningful”—the latter is a separate metric.

    Layer 3: An Unstated Premise for MDC—Noise Must Be Stable. The framework above treats the two measurements as “same conditions, differing by only one intervention.” However, measurements before and after Form Sha renovation are separated by months or even a season. Seasonal variation in indoor environments is empirically documented: spatial and seasonal variations in VOC concentrations can cause significant differences in the same room across seasons (PMID 27163962); repeatability and seasonal variation in indoor dust and bacterial components, as well as ergosterol, are equally clear (PMID 24992650, PMID 24883434). When the noise itself is drifting, MDC is not a safety pad but an old receipt stamped with last quarter’s figures. More麻烦的是 (troublesome), the noise evaluation field itself has accounted for this: measurement uncertainty directly determines whether noise mapping validation passes (PMID 32310126)—the same limit can yield opposite conclusions depending on the uncertainty framework.

    Layer 4: The Main Source of Error in the Low-Frequency Band Is Not the Instrument, But “Where You Stand.” In mid-to-high frequency rooms, sound fields approximate diffusion; moving the position makes little difference. This is not true for low frequencies. The assumption of a diffuse field below 315 Hz in reverberation rooms is itself invalid and requires alternative measurement methods (PMID 41711454); sound transmission loss in small-scale reverberation rooms is directly rewritten by specimen size and opening geometry (PMID 42451325). Translated into field language: Standing wave patterns in the 4–20 Hz range form a fixed peak-valley grid in a room; moving the measurement point by twenty centimeters can cause reading changes larger than the entire renovation’s effect. Yet the MDC formula lacks a “position” term—it treats measurement as a point, whereas low-frequency measurement is never a single point.

    Layer 5: Why a “Unrenovated Control Building” Is Essential. MDC alone is insufficient because of regression to the mean: renovated buildings are typically selected because they were already complained about as “particularly noisy,” so extreme values naturally regress on the next measurement. The Bland-Altman framework clarifies this, using mean difference and standard deviation to set bounds, provided there is no systematic component in the difference (PMID 10501650, PMID 26110027); estimating this bound requires separate sample size calculations (PMID 27838682). Thus, the skeleton must be: Same season, same instrument, concurrent measurement of a renovation building and a control building in batches, with double testing—the portion in the control building that shows no rise or fall is what must be deducted as drift. Research on noise exposure and annoyance responses also suggests: Human reactions to “how much it changed” depend on the levels before and after the change, not just the magnitude of change (PMID 22902956).

    Layer 6: MDC Must Be Calculated Independently, Not Copied. The lesson from the bone density field is most direct: manufacturer-recommended precision assessment sample sizes are often insufficient, leading to unstable minimum significant changes and spurious changes appearing in clinical “monitoring” (PMID 16900302); precision errors also change systematically with measurement conditions (PMID 17937536); LSC must be paired with monitoring time intervals—slow-changing indicators will never show true changes if the interval is too short (PMID 23452871, PMID 26032212). Corresponding to Form Sha: MDC must be calculated using retest data from the same building, same room, and same instrument, and must be reported alongside “how long until it is possible to cross the MDC.” Sample size estimation methods for longitudinal panels are already mature (PMID 28957035); there is no reason renovation studies should not calculate them.

    Layer 7: What a Judgment Experiment Looks Like. ① ≥10 renovated buildings plus ≥10 control buildings, double testing in the same season with at least 3 time points before and after spanning ≥1 season; ② At each time point, ≥3 repeated readings per room, with fixed measurement points, heights, and orientations, photographed for records, while simultaneously recording temperature/humidity, external window status, HVAC settings, and number of occupants; ③ Variance decomposition to extract the contribution rates of five factors: instrument, position, intra-day, season, and mode; ④ Calculate MDC95 and monitoring time intervals using the building’s own retest data; ⑤ First correct Δ using control building drift, then compare with MDC95. The criteria are threefold: If Δ after correction is below MDC95 ⇒ Cannot judge; must write “this design lacks resolution” rather than “ineffective”; If Δ falls between MDC95 and twice that ⇒ Can only report “detectable but not robust,” requiring extended observation; If the “position” term contributes over half within MDC95 ⇒ Low-frequency measurement in that room cannot be used for individual pre-post comparison, only group means. A further old lesson in intervention assessment: without time series in pre-post controls, natural fluctuations are easily misread as effects (PMID 39181016, PMID 32829342).

    Yijing Application Point: Feng Shui discusses “mitigation” and always附带 (incorporates) an “applicable period”—three years, five years, or a specific cycle. In the past, this was regarded as esoteric; in measurement terms, “applicable period” is a simple MDC intuition: knowing your scale is inaccurate, you dare not report small changes. The bone density field codified this intuition into a formula: minimum significant change and monitoring time interval must both be provided (PMID 23452871). “The ancients’ ‘do not change for three years’ and modern measurement’s ‘observation interval must exceed the time required to cross MDC’ are two statements of the same discipline.” Conversely, today’s biggest flaw in “mitigation” marketing is treating noise as signal: seasonal drifts, shifted measurement points, and regression of extreme values are all claimed as “layout efficacy.” True tradition is not “effective immediately after renovation,” but “wait for the applicable period after renovation”—it acknowledges that silence is more professional than reporting good news beneath the noise.

    Evidence and Limitations

    Verified: The calculation framework for Minimum Detectable Change (MDC/SRD) and its relationship with the standard error of measurement is empirically supported in rehabilitation assessment and functional scales (PMID 11822790, PMID 19261767, PMID 36118657, PMID 37606246); indoor environmental measurements exhibit clear spatial and seasonal variations with limited repeatability (PMID 27163962, PMID 24992650, PMID 24883434); measurement uncertainty is sufficient to alter whether noise evaluation passes (PMID 32310126); the diffuse field assumption fails in low-frequency bands, with measurement geometry directly rewriting results (PMID 41711454, PMID 42451325); insufficient sample sizes for precision assessment lead to unstable minimum significant changes and frequent spurious changes in monitoring (PMID 16900302, PMID 17937536, PMID 23452871, PMID 26032212).

    Partially Verified: The above methodological conclusions derive from medical measurement, occupational hygiene, and architectural acoustics fields respectively; no single study has transferred this entire framework to the specific scenario of “before and after Form Sha renovation”; the thresholds in this design (MDC95, twice MDC95, position contribution exceeding half) are calibrated initial values based on general conventions, not empirical conclusions.

    To Be Verified: ① The actual magnitude of retest noise floors in the 4–20 Hz band for single residential units (SD and ICC), and the difference between within-building and between-building differences; ② The ranking of the five variance contributions (instrument, position, intra-day, season, mode) in real residential settings, and whether “position” truly dominates; ③ Whether the distribution of post-renovation Δ truly crosses MDC95 and what proportion does; ④ Whether “applicable period”-style long observation windows (≥1 year) significantly increase detection rates.

    Related Reading

    Evidence Level: B
    References: PMID 11822790, PMID 19261767, PMID 36118657, PMID 37606246, PMID 27163962, PMID 24992650, PMID 24883434, PMID 32310126, PMID 41711454, PMID 42451325, PMID 10501650, PMID 26110027, PMID 27838682, PMID 22902956, PMID 16900302, PMID 17937536, PMID 23452871, PMID 26032212, PMID 28957035, PMID 39181016, PMID 32829342


    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.

  • 形煞分类是一堵墙还是一条带:分层建模之前,先问「类」在哪里断

    Evidence level: B

    上一轮问”统一模型能不能跨五种楼形外推”,这一轮问反过来的问题:当统一模型失败时,”按截面类型分层”是不是答案?以及更扎心的——形煞的那些类,到底是不是物理上真实的分界?

    正文

    第一层:先把问题翻译对。 本课题的表面问题是”分层模型 vs 统一模型谁预测得更准”,但真正的问题是:风水师按形状分的那些类(方/圆/锥/切角/扭转),是否恰好落在 St 依赖方式的物理断点上? 分层建模只是手段,目的是检验”分类本身有没有物理依据”。若分类是真实的,分层该赢;若分类只是对连续形态空间的离散命名,分层赢就是假的——多出来的参数只是在拟合噪声。

    第二层:分层”赢”常常是统计幻觉。 类数检出有个残酷的实证:潜在剖面分析里,当真实类间距小(Cohen’s d = 0.2-0.5)时,七种模型选择方法没有一个能可靠挑出正确的类数,只有类间距极大(d = 1.5)时 LMR、bootstrap LRT、BIC 才靠谱,而 AIC 与熵几乎总是选错(PMID 24489457)。更糟的是伪类:增长混合建模里,结果变量只是轻微偏离正态,就会提取出本不存在的类,并把实质意义硬安上去(PMID 29795894)。“分成几类效果更好”本身,可能是非正态误差造出来的。 所以裁决的第一步不是比模型,是确认这个类数在更严的零假设下还存在。

    第三层:但”借力”本身有坚实的正面理由。 亚组分析里的经验贝叶斯收缩给出了经典结果:当存在 3 个以上亚组时,把每个亚组的估计向总体均值收缩,虽然单个估计可能被”拉偏”,但总预期误差下降——独立分析反而更差(PMID 20138396)。而且借用强度可量化:借用强度与借用指数用 Mallow 距离构造,能从 MCMC 样本算出每组之间实际借了多少信息(PMID 32341613)。分层不是”要不要借”,而是”借多少、该不该借”——这把一刀切的问题变成了一个可测的量。

    第四层:问题是”该借多少”不是数据自己能定的。 basket trial 的方法学给出了最直接的教训:完全贝叶斯的分层模型在 10 个以下亚组时,没有足够信息确定借用程度,会显著膨胀一类错误;校准版把收缩参数定义成”亚组相似度”的函数、用模拟标定,才把一类错误压下来(PMID 29499621)。对应到我们的五类截面:样本少到一定程度,”该不该借”根本测不出来,分层模型的漂亮结果可能是借错了方向。

    第五层:连裁判工具都不可靠。 用留一交叉验证做贝叶斯模型比较,有一个著名警示:PSIS-LOO 在多层回归+事后分层(MRP)的工作流里,恢复不了模型的正确排序——因为在聚合阶段,个体层面的预测误差互相抵消了(PMID 38146825)。本课题的外推误差正是”聚合到截面族层面”来比较的,恰好踩中这个抵消模式。所以裁决实验要先验证比较工具本身:用已知真结构的模拟数据测试 LOO 能否认出对的模型,再拿去判真实数据。

    第六层:物理侧的真分界,少而粗。 有两个并行的证据:一方面,St-Re 的统一关系 St=1/(A+B/Re) 对多种形状成立,Re→∞ 时极限 St≈0.21 与形状无关,形状差异全被压进参数 B(PMID 26565336)——函数形式没变,类与类之间只是参数连续变化;另一方面,锐缘形状确实有质的不同:尖角让分离点位置被固定住,而圆缘的分离点随 Re 漂移(PMID 39851685)——这才是真正的物理断点。合起来指向一个判断:真分界可能只有”尖/圆”这一刀,其余(切角 30° 还是 45°、锥率 1:3 还是 1:5)都是连续谱上的采样点。

    第七层:裁决设计与三种结局。 用五类截面 × 形态参数扫描做 CFD+风洞,并列三模型:统一共享参数、按截面族分层(组内共享)、全独立;先用模拟数据标定 LOO 工具的认回率,再用留一族交叉验证比外推误差,最后把组间参数的散度画进形态参数空间找断点。① 分层显著赢且断点与”尖/圆”重合:形煞的最大分类(尖圆)是真实物理分界,细分则不是——一堵粗墙,墙内是带;② 分层赢但断点落在风水分类完全不同的位置:形煞分类与物理分界错位,评级按形煞走会系统性误判;③ 统一模型不输:五种楼形之间根本没有墙,”形煞分类”只是命名习惯。

    回到易学。 风水把形分成尖圆方劈诸类,本质是给一个连续的形状空间起名字。起名字有大用——它让师徒能指物而谈、让吉凶有个粗分的框架,这不丢人。但古人没有仪器,分不出”这一刀是真的断,那几刀只是刻度”。现代测量说得很清楚:名字不等于边界。 若最终裁决是”尖圆一刀、其余连续”,那么形煞口诀里最该保留的是”尖者带煞、圆者中和”这一类极粗判断,而”切角几度、扭转几分”这类细分,必须降格为经验描述、不得当公式用。给连续的东西命名,是人类的天性;把命名当成发现了边界,是学科的陷阱。

    证据与局限

    已验证: 类间距小时没有任何常用统计量能可靠检出真实类数,AIC 与熵尤其差(PMID 24489457);结果变量非正态时混合建模会提取伪类并赋予实质意义(PMID 29795894);3 个以上亚组时经验贝叶斯收缩降低总预期误差(PMID 20138396);借用强度/借用指数可量化分层模型的实际信息借用(PMID 32341613);小样本下完全贝叶斯分层模型定不了借用程度、一类错误膨胀,需校准(PMID 29499621);PSIS-LOO 在多层模型聚合层面对模型排序不可靠(PMID 38146825);St-Re 统一关系跨形状成立、形状只进参数 B(PMID 26565336);锐缘固定分离点、圆缘分离点随 Re 漂移(PMID 39851685)。

    部分验证: “统一函数形式+参数连续变化”与”尖圆存在质的分界”两个主张各自有证据,但”风水形煞分类是否与 St 依赖方式的物理分界重合”从未被同一批算例直接检验过;形煞分类的类数与间距(切角/锥率/扭转如何取值)没有公开的标准化数据集可供裁决。

    待验证: ① 分层模型相对统一模型的增益在留一族外推下是否显著;② 若显著,组间参数散度在形态参数空间是否形成可定位的断点,断点是否与”尖/圆”重合;③ 风水形煞分类(方/圆/锥/切角/扭转)与物理分界的重合度;④ 用模拟数据标定后,LOO 工具在本课题的聚合结构下能否正确认出真实模型。

    相关阅读

    证据等级: B
    参考文献: PMID 24489457, PMID 29795894, PMID 20138396, PMID 32341613, PMID 29499621, PMID 38146825, PMID 26565336, PMID 39851685


    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。

    Formal Classification: Is It a Wall or a Strip? Asking Where the Class Boundary Lies Before Layered Modeling

    Evidence level: B

    The previous round asked whether a unified model can be extrapolated across five building shapes. This round asks the reverse question: When the unified model fails, is “layered modeling by cross-section type” the answer? And more piercingly—do the categories of Form Sha represent physically real boundaries?

    Main Text

    Layer 1: Translate the problem correctly. The surface issue of this topic is “which predicts more accurately: layered models vs. unified models,” but the real question is: Do the categories used by Feng Shui practitioners (square/circular/conical/angular/twisted) align with the physical breakpoints of St-dependence? Layered modeling is merely a means; the goal is to test whether “the classification itself has physical basis.” If the classification is real, layered modeling should win; if it is merely a discrete naming of a continuous morphological space, winning with layered modeling is false—the extra parameters are just fitting noise.

    Layer 2: Layered “wins” are often statistical illusions. There is a harsh empirical fact regarding class detection: in latent profile analysis, when the true inter-class distance is small (Cohen’s d = 0.2-0.5), none of the seven model selection methods can reliably identify the correct number of classes. Only when the inter-class distance is very large (d = 1.5) do LMR, bootstrap LRT, and BIC perform reliably, while AIC and entropy almost always select incorrectly (PMID 24489457). Worse still are pseudo-classes: in growth mixture modeling, if the outcome variable only slightly deviates from normality, the model extracts non-existent classes and imposes substantive meaning on them (PMID 29795894). “Dividing into more classes yields better results” itself may be generated by non-normal errors. Therefore, the first step in adjudication is not comparing models but confirming whether this class number exists under a stricter null hypothesis.

    Layer 3: But “borrowing strength” itself has solid positive justification. Empirical Bayesian shrinkage in subgroup analysis yields a classic result: when there are three or more subgroups, shrinking each subgroup’s estimate toward the overall mean reduces the total expected error, even if individual estimates may be biased—analysis without borrowing performs worse (PMID 20138396). Moreover, borrowing strength is quantifiable: borrowing strength and the borrowing index are constructed via Mallow’s distance, allowing calculation of actual information borrowed between groups from MCMC samples (PMID 32341613). Layered modeling is not a “yes/no” question of whether to borrow, but “how much to borrow and whether to borrow”—turning a blunt-cut problem into a measurable quantity.

    Layer 4: The question of “how much to borrow” cannot be determined by data alone. Methodological lessons from basket trials provide the most direct insight: fully Bayesian layered models with fewer than 10 subgroups lack sufficient information to determine the degree of borrowing, leading to significant inflation of Type I error; calibrated versions define the shrinkage parameter as a function of “subgroup similarity” and use simulation calibration to suppress Type I error (PMID 29499621). Corresponding to our five cross-section types: when sample sizes are too small, whether “to borrow” cannot be measured at all; the pretty results of layered models may stem from borrowing in the wrong direction.

    Layer 5: Even the adjudication tools are unreliable. Using leave-one-out cross-validation for Bayesian model comparison carries a famous warning: PSIS-LOO fails to recover the correct model ranking in workflows involving multivariate regression plus post-hoc stratification (MRP)—because at the aggregation stage, individual-level prediction errors cancel each other out (PMID 38146825). Our topic’s extrapolation error is compared exactly at the “aggregated cross-section family level,” coincidentally stepping on this cancellation mode. Therefore, adjudication experiments must first validate the comparison tool itself: test whether LOO can identify the correct model using simulated data with known true structures, then apply it to real data.

    Layer 6: True physical boundaries are few and coarse. There are two parallel lines of evidence: On one hand, the unified St-Re relationship St=1/(A+B/Re) holds for multiple shapes; as Re→∞, the limit St≈0.21 is shape-independent, with all shape differences compressed into parameter B (PMID 26565336)—the functional form does not change; classes differ only by continuous parameter variation; On the other hand, sharp-edged shapes indeed have qualitative differences: sharp corners fix the separation point location, while rounded edges cause the separation point to drift with Re (PMID 39851685)—this is the true physical breakpoint. Together, they point to a judgment: The true boundary may only be “sharp vs. rounded,” while the rest (30° vs. 45° cut angles, 1:3 vs. 1:5 taper ratios) are just sampling points on a continuous spectrum.

    Layer 7: Adjudication design and three outcomes. Using CFD and wind tunnel testing with five cross-section types and morphological parameter scanning, compare three models: unified shared parameters, layered by cross-section family (shared within groups), and fully independent. First, calibrate LOO tool recognition rates using simulated data; second, compare extrapolation errors using leave-one-group-out cross-validation; finally, plot the dispersion of inter-group parameters into morphological parameter space to find breakpoints. ① If layered modeling wins significantly and the breakpoint coincides with “sharp/rounded”: Form Sha’s maximum classification (sharp/rounded) represents a true physical boundary, while subdivision does not—a thick wall with a band inside; ② If layered modeling wins but the breakpoint falls at a location completely different from Feng Shui classifications: Form Sha classification misaligns with physical boundaries, leading to systematic misjudgment when rating follows Form Sha; ③ If the unified model does not lose: there is no wall between the five building shapes; “Form Sha classification” is merely a naming convention.

    Returning to Yijing Studies. Feng Shui divides forms into sharp, round, square, split, and other categories, essentially naming a continuous shape space. Naming is highly useful—it allows masters and apprentices to refer to objects and discuss them, providing a coarse framework for auspiciousness and inauspiciousness; this is not shameful. However, ancients lacked instruments to distinguish “whether this cut is a true break or those few cuts are just scales.” Modern measurement is very clear: names do not equal boundaries. If the final adjudication is “one cut for sharp/rounded, continuous elsewhere,” then the most valuable retention in Form Sha aphorisms is the extremely coarse judgment “sharp carries sha, round is neutral,” while subdivisions like “how many degrees of cut angle or how much twist” must be downgraded to empirical descriptions and should not be used in formulas. Naming continuous things is human nature; treating names as discovered boundaries is a disciplinary trap.

    Evidence and Limitations

    Verified: When inter-class distance is small, no common statistical metric can reliably detect the true number of classes; AIC and entropy perform particularly poorly (PMID 24489457); when the outcome variable is non-normal, mixture modeling extracts pseudo-classes and assigns them substantive meaning (PMID 29795894); empirical Bayesian shrinkage reduces total expected error when there are three or more subgroups (PMID 20138396); borrowing strength and borrowing index quantify actual information borrowing in layered models (PMID 32341613); fully Bayesian layered models cannot determine borrowing degrees with small samples, causing Type I error inflation and requiring calibration (PMID 29499621); PSIS-LOO is unreliable for model ranking at the aggregation level of multivariate models (PMID 38146825); the unified St-Re relationship holds across shapes with shape differences only entering parameter B (PMID 26565336); sharp edges fix separation points while rounded edges cause drift with Re (PMID 39851685).

    Partially Verified: Both claims—”unified functional form with continuous parameter variation” and “qualitative boundary exists between sharp and round”—have supporting evidence, but whether “Feng Shui Form Sha classifications align with physical breakpoints of St-dependence” has never been directly tested using the same set of cases; there is no publicly available standardized dataset for adjudicating the number and spacing of Form Sha classes (how cut angles/taper ratios/twist are defined).

    To Be Verified: ① Whether the gain of layered models over unified models is significant under leave-one-group-out extrapolation; ② If significant, whether the dispersion of inter-group parameters forms a locatable breakpoint in morphological parameter space and if it coincides with “sharp/rounded”; ③ The degree of overlap between Feng Shui Form Sha classifications (square/circular/conical/angular/twisted) and physical boundaries; ④ Whether, after calibration with simulated data, the LOO tool can correctly identify the true model under the aggregation structure of this topic.

    Related Reading

    Evidence Level: B
    References: PMID 24489457, PMID 29795894, PMID 20138396, PMID 32341613, PMID 29499621, PMID 38146825, PMID 26565336, PMID 39851685


    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.