Author: joe

  • 「老本」到底存不存在:修炼年限分层的停练保留差,可能只是百分比的算术

    Evidence level: C

    「练了十年的人停练也掉得少」——这句话看着像观察,其实是把一个除法结果当成了生理事实。

    正文

    一、这条假说的出处,说的是另一件事

    「高手有老本、新手全流失」在运动侧最常被引的是 Mujika 的停练综述:长期停练后运动员的最大摄氧量明显下降但仍高于对照组,而新近获得的增益则完全丢失。再往前一层,Bosquet 的元分析给过一个方向性结论:不活动者的停练降幅大于运动员。

    但原文那句话里没有「老本」这个词。它说的是两个群体各自相对自己的起点处在什么位置,而不是「谁更会守」。把「仍高于对照」读成「保留能力强」,中间隔着一个从未被验证的换算。而呼吸侧要复现这个结论,必须先回答:你比的是哪个量。

    二、百分比这道除法,「老本」和「衰减按量」是同一件事

    保留率通常写成「停练末值 ÷ 各自峰值」。这个分数有个隐蔽性质:只要两组按同一比例衰减,保留率必然相同;只有当衰减按绝对量走,起点高的那组保留率才会显得更高。也就是说,「老本效应成立」在数学上几乎等价于「衰减是加法而不是乘法」——而这两者从未在丹田参数上被区分过。

    更要紧的是分母。高手峰值高、新手峰值低,同样掉 8 个单位,高手保留 94%、新手保留 88%,差值 6 个百分点听起来像效应,实际是同一个绝对下降。以「保留百分比」为因变量的组间检验,本质上是在检验基线的组间差。这在临床测量里是常识:最小临床重要改善随基线值升高而变大,而作者已证明那不是「期望不同」,而是地板与天花板效应。

    三、峰值高的人,本来就注定要掉

    横断面按修炼年限分层时,年限越高的人峰值越可能落在分布尾部。而任何落在尾部的人,复测时都会向均值靠拢:入选标准一旦包含基线期,回归到均值占观测变化的比例可以从 23%–25% 升到 25%–47%, placebo 反应随之虚高、检验效能随之下降。停练研究的「峰值」恰恰是这样一个被用来分层、又含测量误差的极端值。

    所以停练末点的下降里,至少有四份要拆开:真实衰减、回归到均值、复测的学习或习惯化、以及「练功态与静息态口径」的漂移。前三份都有成熟的处理(协方差分析优于变化分数、重复基线、随机化),但没有一份能靠百分比保留率自动消掉。

    四、呼吸侧的前提本来就不成立

    运动侧的「完全停练」意味着肌肉不再被使用。呼吸肌做不到这一点:它需要持续激活以维持肺泡通气,即便撤掉训练刺激,它也每天工作两万次。实测也印证了这条:吸气压训练停 5 周后最大吸气压从 154 落到 153 cmH₂O(几乎不动),停 9 周才出现小幅下降并平台化、仍高于训练前;而抗阻训练 7.2 周平均掉 14.5%。同为「停练」,呼吸肌的降幅小一个量级,原因很可能不是它更会守,而是它从未真正停过。

    这意味着「修炼年限 × 停练保留」的交互,在呼吸侧被一个恒定的背景刺激压住了。想让它显形,得先把「撤训练」与「撤使用」分开。

    五、验「不对称」要四倍样本,没人算过

    这个设计真正要检验的是交互项(年限 × 时点),不是两个主效应。而交互项的方差是主效应的四倍:要与主效应同等大小、同等功效地被检出,样本量需增至四倍;样本量不变时,交互必须至少是主效应的两倍才可能被看见。一个三组分层、每组二十人的停练研究,检出的功效大概在两成上下——不显著几乎必然,而它会被写成「未见老本效应」。

    六、要做成,得先改三件事

    第一,因变量从保留率换成绝对变化量与「可获得空间占比」并列报告。绝对变化量回答「掉了多少」,可获得空间占比回答「在还剩多少余地里掉了多少」,两者的组间结论若不一致,本身就说明百分比在作祟。第二,停练前先做两到三次重复基线,用其均值而非单次峰值作参照,并按协方差分析调整而不是比较变化分数——这是唯一能同时压住回归到均值与随机误差的做法。第三,把停练写成撤训练而非撤行为:设置「维持极低剂量组」(如每周两次、每次五分钟),它的存在能让「衰减」与「使用停止」在统计上分开;若极低剂量组的保留率显著高于完全停练组,那所谓老本更可能是「日常呼吸习惯被保留」而非「十年功力存下来」。

    按这三条改完,四种结局才有分别:绝对量与占比两口径结论一致且年限交互显著 ⇒ 老本成立;仅占比口径显著 ⇒ 判为算术伪迹,既往按百分比得出的停练结论须重算;交互不显著但极低剂量组明显更好 ⇒ 老本改判为「维持剂量效应」;全部无差异而个体内变异远大于组间差 ⇒ 停练保留根本不是稳定特质,不该作终点。

    证据与局限

    已验证: 长期停练后运动员摄氧量仍高于对照、新近增益全失(PMID 10999420);抗阻训练停练约 7 周平均降 14.5%(PMID 23529287);娱乐性力量训练者停 6 周力量变化极小但功率掉 9%–10%(PMID 12173951);吸气压训练停 5 周强度几乎不变(PMID 36754374)、停 9 周小幅下降后平台化(PMID 12569211);肌核密度随停练回落、肌肉记忆的细胞学基础尚无共识(PMID 30840493, PMID 32881361);回归到均值可占观测变化的四分之一到近半(PMID 37505116, PMID 24073634);最小重要改善随基线升高源于地板天花板而非期望(PMID 24556220);交互检验需四倍样本(PMID 14633287)。

    部分验证: 停练期衰减是乘法还是加法,运动侧也未有定论;呼吸肌衰减慢被归因于「持续激活」,是机制推测而非实测对照;长期冥想者静息自主指标是特质还是状态,综述判为以状态为主(PMID 40204160),但未覆盖丹田呼吸。

    待验证: 丹田参数(%AB/φ/X5/CoV)停练期的衰减是比例式还是绝对式;峰值回归到均值在该设计里占多少;撤训练与撤使用分离后交互是否仍存在;以绝对变化量或「相对可获得空间」为因变量时,老本效应是否还在。

    相关阅读

    证据等级: C
    参考文献: PMID 10999420, PMID 23529287, PMID 12173951, PMID 36754374, PMID 12569211, PMID 30840493, PMID 32881361, PMID 37505116, PMID 24073634, PMID 24556220, PMID 14633287, PMID 40204160

    L1 网络状态:「该主题尚未进入 L1 对应网络」——CORRESPONDENCE_NETWORK.md 节点表只把「呼吸率 / 呼吸模式」列为生理侧观测项(备注「与节律耦合最直接」)、把「气感报告」列为结构化自评,二者均未与易学侧任何节点连成边;节点表内不存在「修炼年限」「停练 / 保留」「训练史」这类纵向维度节点,也没有「指标口径 / 保留率 / 变化量」这类测量学节点。边表 3.1–3.4 只登记五行↔五脏/五色/五音/五方/五季/五志、八卦↔五行、天干地支↔五行,无任何以丹田参数、停练保留或训练年限为源或目标的边。EDGE_TABLE_science-yixue.md 仅有概念计数(丹田 51 篇 / 气功 48 篇 / 呼吸 189 篇),无层级与状态判定。故本文封顶 C。


    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。

    Does the \”Old Stock\” Actually Exist: The Retention Gap Based on Cultivation Years Might Just Be a Percentage Arithmetic

    Evidence level: C

    “People who have practiced for ten years lose little even when stopping practice” — this statement appears to be an observation, but it treats a division result as a physiological fact.

    Main Text

    I. The Origin of This Hypothesis Refers to a Different Matter

    The claim that “masters have old stock while novices lose everything” is most frequently cited in sports science regarding Mujika’s review on detraining: after long-term cessation, athletes’ maximal oxygen uptake drops significantly but remains higher than the control group, whereas recently acquired gains are completely lost. Moving one layer further back, Bosquet’s meta-analysis provided a directional conclusion: the detraining loss in sedentary individuals is greater than in athletes.

    However, the original statement does not contain the term “old stock.” It describes the position of two groups relative to their own starting points, not who is “better at retaining.” Reading “still higher than control” as “stronger retention ability” involves an unverified conversion. To replicate this conclusion in the respiratory domain, one must first answer: which metric are you comparing?

    II. Percentage Division: “Old Stock” and “Linear Decay” Are the Same Thing

    Retention is typically expressed as “post-detraining value ÷ respective peak.” This fraction has a hidden property: if both groups decay at the same proportion, retention rates will necessarily be identical; only if decay proceeds via absolute amounts will the group with the higher starting point appear to have a higher retention rate. In other words, “the old stock effect holds” is mathematically almost equivalent to “decay is additive rather than multiplicative”—yet these two have never been distinguished in dantian parameters.

    More importantly, consider the denominator. Masters have high peaks; novices have low peaks. A drop of 8 units results in 94% retention for masters and 88% for novices—a difference of 6 percentage points that sounds like an effect, but is actually the same absolute decline. Group comparisons using retention percentage as the dependent variable essentially test inter-group differences at baseline. This is common sense in clinical measurement: minimal clinically important improvement increases with higher baseline values, and the author has already proven this is not due to “different expectations” but rather floor and ceiling effects.

    III. High Peaks Are Already Destined to Drop

    When stratifying by cultivation duration cross-sectionally, individuals with longer durations are more likely to fall into the distribution tail. Any individual falling into the tail will regress toward the mean upon retesting: once inclusion criteria encompass the baseline period, the proportion of observed change explained by regression to the mean can rise from 23%–25% to 25%–47%, artificially inflating placebo responses and reducing statistical power. The “peak” in detraining studies is precisely such an extreme value used for stratification yet containing measurement error.

    Therefore, at least four components must be disentangled within the drop observed at the post-detraining endpoint: true decay, regression to the mean, retest learning or habituation, and drift in “practice state vs. rest state” metrics. The first three have mature handling methods (ANCOVA is superior to change scores, repeated baselines, randomization), but none can be automatically eliminated by retention percentages.

    IV. The Premise in the Respiratory Domain Was Never Valid

    “Complete detraining” in sports implies muscles are no longer used. Respiratory muscles cannot achieve this: they require continuous activation to maintain alveolar ventilation; even without training stimuli, they work roughly 20,000 times daily. Measurements confirm this: after stopping inspiratory pressure training for 5 weeks, maximal inspiratory pressure dropped from 154 to 153 cmH₂O (almost unchanged); after 9 weeks, a slight decline occurred followed by plateauing, still remaining higher than pre-training levels; whereas resistance training showed an average drop of 14.5% over 7.2 weeks. As “detraining,” the respiratory muscle decline is an order of magnitude smaller, likely not because it retains better, but because it never truly stopped.

    This means the interaction between “cultivation duration × detraining retention” is suppressed in the respiratory domain by a constant background stimulus. To make it visible, one must first separate “withdrawing training” from “withdrawing use.”

    V. Verifying “Asymmetry” Requires Four Times the Sample Size, Never Calculated

    This design truly tests the interaction term (duration × time point), not two main effects. The variance of an interaction is four times that of a main effect: to detect it with equal magnitude and power, sample size must increase fourfold; if sample size remains unchanged, the interaction must be at least twice the main effect to be visible. A three-group stratified study with twenty participants per group has a detection power around 20%—non-significance is almost inevitable, yet it will be written as “no old stock effect observed.”

    VI. To Achieve This, Three Things Must Change First

    First, replace retention rate as the dependent variable with absolute change amounts and report “proportion of available space” concurrently. Absolute change answers “how much was lost,” while proportion of available space answers “how much was lost within the remaining margin.” If inter-group conclusions differ between these two, it itself indicates percentages are distorting results. Second, conduct 2–3 repeated baselines before detraining, using their mean rather than a single peak as reference, and adjust via ANCOVA instead of comparing change scores—this is the only method that simultaneously suppresses regression to the mean and random error. Third, frame detraining as withdrawing training rather than withdrawing behavior: set up a “very low-dose maintenance group” (e.g., twice weekly, five minutes each); its existence allows “decay” and “cessation of use” to be separated statistically; if the very low-dose group shows significantly higher retention than the complete cessation group, so-called “old stock” is more likely “daily breathing habits retained” rather than “decades of功力 preserved.”

    With these three modifications, four outcomes become distinguishable: absolute amount and proportion conclusions consistent with significant duration interaction ⇒ old stock holds; only proportion significant ⇒ judged as arithmetic artifact, prior detraining conclusions based on percentages must be recalculated; interaction not significant but very low-dose group clearly better ⇒ old stock reclassified as “maintenance dose effect”; all non-significant with individual variance far exceeding inter-group differences ⇒ detraining retention is not a stable trait and should not serve as an endpoint.

    Evidence and Limitations

    Verified: Post-detraining athletes’ VO2 remains higher than controls while recent gains are fully lost (PMID 10999420); resistance training shows ~14.5% average drop after ~7 weeks (PMID 23529287); recreational strength trainers show minimal strength change but 9%–10% power loss after 6 weeks (PMID 12173951); inspiratory pressure training shows almost no intensity change after 5 weeks (PMID 36754374), slight decline followed by plateauing after 9 weeks (PMID 12569211); myonuclear density regresses with detraining while cellular basis of “muscle memory” lacks consensus (PMID 30840493, PMID 32881361); regression to the mean accounts for one-quarter to nearly half of observed change (PMID 37505116, PMID 24073634); minimal important improvement increases with baseline due to floor/ceiling rather than expectations (PMID 24556220); interaction testing requires four times the sample size (PMID 14633287).

    Partially Verified: Whether detraining decay is multiplicative or additive remains undetermined in sports; slow respiratory muscle decay attributed to “continuous activation” is a mechanistic speculation without experimental control; whether long-term meditators’ resting autonomic indicators are traits or states is judged by reviews as primarily state-based (PMID 40204160), but does not cover dantian breathing.

    To Be Verified: Whether dantian parameters (%AB/φ/X5/CoV) decay proportionally or absolutely during detraining; what proportion of observed change is regression to the mean in this design; whether interaction persists after separating withdrawal of training from withdrawal of use; whether the old stock effect persists when using absolute change or “relative available space” as the dependent variable.

    Related Reading

    Evidence Level: C
    References: PMID 10999420, PMID 23529287, PMID 12173951, PMID 36754374, PMID 12569211, PMID 30840493, PMID 32881361, PMID 37505116, PMID 24073634, PMID 24556220, PMID 14633287, PMID 40204160

    L1 Network Status: “This topic has not yet entered the L1 corresponding network” — The CORRESPONDENCE_NETWORK.md node table lists only “breathing rate / breathing pattern” as physiological observations (noted “most directly coupled with rhythm”) and “qi sensation reports” as structured self-assessments; neither connects to any Yijing nodes. No longitudinal dimension nodes exist within the node table such as “cultivation duration,” “detraining/retention,” or “training history.” There are also no measurement science nodes such as “metric definition / retention rate / change amount.” Edge tables 3.1–3.4 only register Five Phases ↔ Five Organs/Five Colors/Five Tones/Five Directions/Five Seasons/Five Emotions, Eight Trigrams ↔ Five Phases, Heavenly Stems and Earthly Branches ↔ Five Phases; there are no edges originating from or terminating at dantian parameters, detraining retention, or training duration. EDGE_TABLE_science-yixue.md contains only concept counts (Dantian 51 papers / Qigong 48 papers / Breathing 189 papers), with no hierarchy or state determination. Thus, this paper is capped at Level C.


    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.

  • Jaccard 掉到 0.5,是规则脆还是样本被砍了一半:分半稳定性不能做减法分解

    Evidence level: C

    分半稳定性少掉的那 0.5,不能拆成两个量相加——它先要减掉一个随机的底,再判有没有超额。

    一、两个随机集合本来就能撞出 0.5

    若从 30 个候选参数里各随机抽 10 个,两次的 Jaccard 期望不是零:交集约 3.3、并集约 16.7,比值约 0.2。候选池更小、入选数更多,这个底还会抬高。所以「0.5」里先有一块与规则好坏完全无关的东西。

    同类账目在评分者一致性里有现成教训:118 名 ARDS 病人用两套标准判,原始一致率 0.73、kappa 只有 0.46、phi 0.63(PMID 11179128)。同一份数据三个数差出 0.27,差的就是机会水平。筛选规则的分半 Jaccard 一样要过这一关,而现行做法没人减。

    二、减法分解在数学上不成立

    Jaccard 是比值,不是可加量。真分半 J₁、伪分半 J₂、全样本重测 J₃,三者的期望各自是样本量与规则参数的复杂函数。J₁−J₂ 并不等于「互补结构效应」——两个比的差依赖各自基线水平。

    看极端就明白:J₃=0.95、J₂=0.80、J₁=0.70,差值 0.15 与 0.10 看着可拆;J₃=0.55、J₂=0.50、J₁=0.45,同样的绝对差在比值尺度上份量完全不同。要拆只能用模型,把 Jaccard 期望写成 n 与规则参数的函数再拟合,不能做减法。

    三、样本量效应与规则脆性在观测上恒等

    这是最根本的一层,也是原设计绕过去的那一层。观测到 Jaccard=0.5 有两个解释:规则把阈值压在分布稀疏尾部,入选集合本来就在边界抖动;或者 n 太小,每个参数的估计标准误大到让排序整体洗牌。两者给出同一个数。

    要分开,必须动一个、锁住另一个:锁规则改 n 大小,或锁 n 改规则严格度。原设计的三重对照都在动 n 的「结构」(互补还是独立、一次还是两次),唯独没动 n 的「大小」——它测不到样本量效应,只能测到「互补带来的额外相关」。真正的 2×2 是 n 大小 × 切分方式。

    四、互补确实多出一块,但方向可能反

    互补两半共享同一批人的谱相关结构,不是两份独立证据。经典信度理论早写过:测量误差相关会让信度估计膨胀,用系统构造的等价半在两次施测上可控制这种膨胀;若等价半构得不好,失真会走向相反方向,且当量表不满足本质 tau 等价时失真可以很大(PMID 11783569)。

    放到筛选上:真分半的两半因共享结构,选出的集合可能比伪分半更「像」。那么 J₁−J₂ 是正的,会被读成「互补有害」,实际是共享结构抬高了相似度。方向判反,整条结论就反。

    五、「掉到 0.5」可能只是尺子不准

    分半这一步本身在小样本下不可靠:最优切分(Guttman 最大 λ4)在小子样里显著高估总体信度,分位数下界 λ4(0.05) 才是防机会资本化的那一档(PMID 24306557)。「分半 Jaccard」不是无偏观测,它自带偏。

    还不止。Cronbach’s alpha 依赖 item 数,把数据切成固定时长段当 item 用会低估信度,换一种切法数值就变(PMID 19055503)。你的分半是按人切还是按次切、切几刀,直接决定 0.5 这个数是多少。

    六、加样本能不能补回:有实测,答案分模型类

    这是原设计最该问、却用减法回避的问题。已有两条硬结果:

    参数模型上,选择偏差可用逆幂律学习曲线(IPL)外推校正,且外推本身就能回答「要不要再招人」(PMID 25086004)。非参数模型上,所有 bootstrap 乐观校正都显著高估,而这个高估不随样本量增大而减少(PMID 41391734)。

    后者是关键:「加样本补不回」不是推论,是在特定模型类上被观测到的事实。所以「补不回」要看模型类,不是一个统一门槛。另两条参照:小样本下不同方法的分类准确率可以相当,但选出特征的精度与稳定性差异很大,外部交叉验证环能改善(PMID 22403633);而稳定性提高不等于选对了,ensemble 一类方法只是让重抽样下选出同一批(PMID 25134091、PMID 35016102)。

    七、反转:不拆两笔账,改报超额稳定性加曲线斜率

    正确顺序是三步,不是一次减法。

    第一,减掉机会底。按给定候选池大小与入选数算随机 Jaccard 期望,报超额稳定性 =(J−J_chance)/(1−J_chance),让 0.5 变成可比数。注意减 chance 本身有陷阱:低 base rate 下 kappa 类指标会过度惩罚(PMID 25090041),所以随机底要用解析式算,别套 kappa。第二,入选频率用 subsampling 而非 bootstrap(PMID 26288150),并同时报真分半与独立抽两套。第三,报学习曲线斜率与外推平台,把「n 加到多少」变成可答问题。

    四条判据:超额稳定性 ≥0.6 且独立抽口径下仍 ≥0.6,则规则站得住、分半数低是结构伪迹;<0.3 且学习曲线已平台,则规则真脆、0.5 作废线可落地;超额稳定性低但曲线仍陡,则是样本不足,正解是扩样不是改阈值;两套口径符号相反,则该设计没有识别力,任何 Jaccard 结论都不许出。

    证据与局限

    已验证: 原始一致率与机会校正后在同批数据上可差 0.27(PMID 11179128);最优分半在小样本下高估信度、分位数下界更稳(PMID 24306557);bootstrap 乐观校正在非参数模型上高估且不随样本量收敛(PMID 41391734);学习曲线可外推判断是否需扩样(PMID 25086004)。

    部分验证: 「稳定」与「正确」可分离有间接证据(PMID 25134091、PMID 35016102),但缺同一数据集上「稳定性指标 vs 特征真值命中率」的并列对照;减 chance 后的 Jaccard 是否适合做筛选规则的主判据,尚无方法学研究直接验证。

    待验证: 三口径(真分半 / 伪分半 / 全样本重测)在同一批丹田多设备数据上的实际数值差;超额稳定性的阈值 0.6 与 0.3 只是从类比推来,需实测标定;按人切与按次切两种分半口径给出的 Jaccard 差量级未知。

    相关阅读

    证据等级: C
    参考文献: PMID 11179128, PMID 24306557, PMID 41391734, PMID 25086004, PMID 22403633, PMID 25134091, PMID 35016102, PMID 11783569, PMID 19055503, PMID 26288150, PMID 25090041
    L1 网络状态:该主题尚未进入 L1 对应网络——CORRESPONDENCE_NETWORK.md 节点表仅按易学/中医侧、物理侧、生理/体验侧三类域分派,三类域均无「分半稳定性 / 重抽样口径 / 机会校正 / 学习曲线」类测量学维度节点;边表止于 E-050(「六十甲子纳音 | 对应 | 60 个频率」,状态 open),无任何以筛选规则、样本量或可复现性为源或目标的边。EDGE_TABLE_science-yixue.md 亦仅有概念计数(丹田 51 篇 / 气功 48 篇 / 呼吸 189 篇 / 频率 407 篇),无 L1/L2/L3 层级与 active/open/conflict 状态判定。


    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。

    Jaccard Drops to 0.5: Is It Rule Fragility or Half the Samples? Semi-Stability Cannot Be Decomposed by Subtraction

    Evidence level: C

    The missing 0.5 in split-half stability cannot be decomposed into two additive components—it must first subtract a random baseline before determining if there is an excess.

    I. Two Random Sets Naturally Collide at 0.5

    If one randomly selects 10 parameters from a pool of 30 candidates twice, the expected Jaccard index is not zero: the intersection is approximately 3.3, the union approximately 16.7, yielding a ratio of roughly 0.2. A smaller candidate pool or a larger selection count would further elevate this baseline. Therefore, a portion of “0.5” is unrelated to the quality of the rule itself.

    Precedents in inter-rater reliability offer a clear lesson: among 118 ARDS patients, two sets of criteria yielded an initial agreement rate of 0.73, a kappa of only 0.46, and a phi of 0.63 (PMID 11179128). The same dataset produced three metrics differing by 0.27, representing the chance level. Split-half Jaccard for screening rules must pass this test, yet current practices fail to subtract it.

    II. Subtraction Is Mathematically Invalid for Jaccard

    Jaccard is a ratio, not an additive quantity. The expectations for true split-half ($J_1$), pseudo-split-half ($J_2$), and full-sample retest ($J_3$) are each complex functions of sample size and rule parameters. The difference $J_1 – J_2$ does not equal a “complementary structural effect”—the difference between two ratios depends on their respective baseline levels.

    Extreme cases clarify this: if $J_3=0.95, J_2=0.80, J_1=0.70$, the differences of 0.15 and 0.10 appear decomposable; however, if $J_3=0.55, J_2=0.50, J_1=0.45$, the same absolute difference represents vastly different proportions on a ratio scale. Decomposition requires modeling, expressing Jaccard expectation as a function of $n$ and rule parameters for fitting; subtraction is invalid.

    III. Sample Size Effects and Rule Fragility Are Indistinguishable in Observation

    This is the fundamental layer that the original design circumvented. Observing Jaccard = 0.5 admits two explanations: the rule suppresses the threshold into a sparse tail of the distribution where selection sets naturally fluctuate at the boundary; or $n$ is too small, causing large standard errors for each parameter that effectively reshuffles the overall ranking. Both yield the same number.

    To separate them, one must fix one variable while holding the other constant: lock the rule and vary $n$, or lock $n$ and vary rule strictness. The original design’s three controls all manipulate the “structure” of $n$ (complementary vs. independent, once vs. twice), but never the “magnitude” of $n$. It cannot detect sample size effects, only “extra correlation arising from complementarity.” The true $2 \times 2$ design is $n$ magnitude $\times$ splitting method.

    IV. Complementarity Adds a Component, But the Direction May Reverse

    The two halves of the complement share the same spectral correlation structure among subjects; they are not two independent pieces of evidence. Classical reliability theory addresses this: measurement error correlation inflates reliability estimates, and using system-constructed equivalent halves across two administrations controls this inflation. If the equivalent halves are constructed poorly, distortion may move in the opposite direction, and when scales fail essential tau-equivalence, the distortion can be substantial (PMID 11783569).

    Applied to screening: because the two halves of true split-half share structure, the selected sets may be more “similar” than pseudo-split-half. Thus $J_1 – J_2$ is positive, potentially misread as “complementarity harmful,” whereas it is actually shared structure inflating similarity. Reversing the direction judgment flips the entire conclusion.

    V. “Dropping to 0.5” May Simply Reflect an Inaccurate Scale

    The split-half step itself is unreliable under small samples: the optimal splitting method (Guttman maximum $\lambda_4$) significantly overestimates total reliability in small samples, while the percentile lower bound $\lambda_4(0.05)$ serves as the guard against capitalizing on chance (PMID 24306557). “Split-half Jaccard” is not an unbiased observation; it carries inherent bias.

    Furthermore, Cronbach’s alpha depends on item count; treating fixed-duration data segments as items underestimates reliability, and changing the segmentation alters the value (PMID 19055503). Whether splitting by person or by session, and how many cuts are made, directly determines the magnitude of 0.5.

    VI. Can Adding Samples Restore It? Empirical Evidence Varies by Model Class

    This is the question the original design should have asked but avoided via subtraction. Two hard results exist:

    Under parametric models, selection bias can be corrected via inverse power-law (IPL) learning curve extrapolation, and the extrapolation itself answers “whether to recruit more” (PMID 25086004). Under non-parametric models, all bootstrap optimism corrections significantly overestimate, and this overestimation does not decrease with increased sample size (PMID 41391734).

    The latter is critical: “cannot be restored by adding samples” is not an inference but an observed fact within specific model classes. Thus, “cannot be restored” depends on the model class, not a unified threshold. Two additional references: under small samples, different methods can yield comparable classification accuracy, but precision and stability of selected features vary greatly; external cross-validation loops can improve this (PMID 22403633). However, increased stability does not equal correctness; ensemble methods merely ensure the same subset is selected under resampling (PMID 25134091, PMID 35016102).

    VII. Reversal: Do Not Split the Books; Report Excess Stability and Curve Slope Instead

    The correct sequence is three steps, not a single subtraction.

    First, subtract the chance baseline. Calculate the random Jaccard expectation based on candidate pool size and selection count, then report excess stability = $(J – J_{chance}) / (1 – J_{chance})$, making 0.5 a comparable metric. Note the trap in subtracting chance: kappa-class metrics under low base rates excessively penalize (PMID 25090041); therefore, use analytical formulas for the random baseline rather than kappa. Second, use subsampling instead of bootstrap for selection frequency (PMID 26288150), and simultaneously report both true split-half and independent sampling sets. Third, report learning curve slope and extrapolated plateau, turning “how much $n$ is needed” into a solvable question.

    Four criteria:
    1. If excess stability $\ge 0.6$ and remains $\ge 0.6$ under independent sampling, the rule stands; low split-half is a structural artifact.
    2. If $< 0.3$ and the learning curve has plateaued, the rule is truly fragile; the 0.5 cutoff can be grounded.
    3. If excess stability is low but the curve remains steep, the issue is insufficient sample size; the correct solution is expanding samples, not adjusting thresholds.
    4. If the two sampling methods yield opposite signs, the design lacks discriminatory power; no Jaccard conclusions should be drawn.

    Evidence and Limitations

    Verified: Raw agreement rates differ by 0.27 after chance correction on the same dataset (PMID 11179128); optimal split-half overestimates reliability in small samples while percentile lower bounds are more stable (PMID 24306557); bootstrap optimism correction overestimates in non-parametric models without converging with sample size (PMID 41391734); learning curves enable extrapolation to judge sample expansion needs (PMID 25086004).

    Partially Verified: Indirect evidence exists that “stability” and “correctness” are separable (PMID 25134091, PMID 35016102), but lacks direct comparison of “stability metrics vs. feature true positive hit rates” on the same dataset; whether Jaccard after chance correction is suitable as a primary criterion for screening rules lacks direct methodological validation.

    To Be Verified: Actual numerical differences across the three sampling ports (true split-half / pseudo-split-half / full-sample retest) on actual multi-device Dan Tian data; thresholds of 0.6 and 0.3 for excess stability are derived via analogy and require empirical calibration; the magnitude of Jaccard difference between person-based and session-based splitting is unknown.

    Related Reading

    Evidence Level: C
    References: PMID 11179128, PMID 24306557, PMID 41391734, PMID 25086004, PMID 22403633, PMID 25134091, PMID 35016102, PMID 11783569, PMID 19055503, PMID 26288150, PMID 25090041
    L1 Network Status: This topic has not yet entered the L1 correspondence network—the CORRESPONDENCE_NETWORK.md node table assigns nodes only across three domains: Yijing/TCM, Physics, and Physiological/Experiential; no measurement dimensions nodes exist for “split-half stability/resampling port/chance correction/learning curve” in any of these three domains; the edge table ends at E-050 (“Six Heavenly Stems and Ten Earthly Branchs Na Yin | Correspondence | 60 Frequencies”, status open), with no edges originating or terminating from/to screening rules, sample size, or reproducibility. EDGE_TABLE_science-yixue.md contains only concept counts (Dan Tian 51 / Qigong 48 / Breathing 189 / Frequency 407), lacking L1/L2/L3 hierarchy and active/open/conflict status determination.


    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.

  • 38 张”未解析表”逐张分类:覆盖率缺口里藏着 5 个真错,其中 1 个是自相矛盾

    Evidence level: D

    导言:上轮留下的 61.8% 到底装了什么

    第 12 轮把 delta-table-scan.py 升到 v1.2 时,工具第一次开始自报覆盖率,报出来的数是:

    含 δ/A 列表 63 / 已解析 25 / 未解析 38 / 覆盖率 39.7%
    

    第 12 轮把这件事写成待办 #57:21 张未解析表逐张分类。这个 21 是错的——本轮在同一版本工具上重跑,得到的是 38 张(rows_anchor_skip=97、cells_value_skip=111)。差在哪儿,本文第 5 节交代。

    本文只做一件事:把这 38 张一张一张打开,看它们到底属于哪种”没看懂”,然后回答一个原本没有答案的问题——那 61.8% 里有没有东西。

    答案是:有,而且不轻。


    一、先分类:四种”未解析”,只有一种需要改正文

    工具报”未解析”是把四种完全不同的情况混在一个数字里。逐张打开后,38 张分成四类:

    类 含义 张数 要不要改正文
    F0 本就不该被这个工具扫(无 δ、无 A 的定性/流程表,δ 只是顺带提到) 12 不改,但工具应剔除
    F1 该扫、值也对,只是单位写在表头、格内是裸数字,parse_len 认不出 8 不改,改工具
    F2 该扫、值可算、格内数字与它自己声明的 σ/μr 不自洽 9 改正文(待批)
    F3 该扫、值是文字区间(”数百米–数千米”),无法机械解析 9 改正文(待批),且必须人工判方向

    合计 12 + 8 + 9 + 9 = 38。对上了。

    关键结论先说:F0 占 12 张(31.6%),说明上一轮”21 张”的估法把不该扫的表也算进去了;但 F2+F3 合计 18 张,接近一半的未解析表里确实压着数值问题,其中 5 张已精算确认有错。所以「未解析 ≠ 无问题」这条第 12 轮立的规矩(R3),本轮拿到了实证。


    二、F1(8 张):工具盲区,表是对的

    这批表的形态是 | σ (S/m) | δ (km) | —— 单位在表头,格内只写 0.569。

    parse_len_ex('0.569') 返回 None,因为函数要求格内有 mm|cm|km|m 后缀。于是整列被跳过,n_val_skip 计数器甚至没加一(它只在格内含数字且非可解析长度时才加),所以这些表在报告里显示成 “跳过 0 行、值不可解析 0 格” —— 看起来最无害,其实是整列数字一个都没检查过。

    复算结果(δ = √(2/(ωμσ)),μr=1):

    σ (S/m) 表内值 复算 比
    1e-1 0.569 km 0.5688 km 1.000
    1e-2 1.80 km 1.7986 km 1.000
    1e-3 5.69 km 5.6877 km 1.000
    1e-4 18.0 km 17.986 km 1.000

    四行全对。 涉及 4 张表 ×(cn + lab 双语)= 8 张,分别是 pool4-ternary-shielding-tool-instantiation.md 的中英两版各 2 张(δ 表 + 10 dB 厚度表)。

    这类不需要改正文,需要改工具:表头单位应下传到格内。这条修完之后覆盖率会从 39.7% 抬到约 52%。


    三、F2(9 张):格内数字与自身参数不自洽 —— 5 张已精算有错

    这是本文最重的部分。

    3.1 最硬的一处:同一个文件自己打自己

    skin-depth-basement-building-mu-metal-table.md 有两个地方说花岗岩:

    • 散文(第 80 行):1 km 花岗岩山体在 7.83 Hz 的 δ ≈ 32 km,A(dB) ≈ 8.7 × 1000/32000 ≈ 0.27 dB
    • §3 表(第 72 行):花岗岩小山 | 100 m | 9.7 dB | 27 dB | 54 dB

    厚度减到 1/10(100 m vs 1 km),衰减反而放大 100 倍(27 dB vs 0.27 dB)。

    衰减公式是 A(dB) ≈ 8.7·d/δ,A 与 d 严格成正比。所以同一材料同一频点下,100 m 的衰减必须是 1 km 的十分之一 —— 即 0.027 dB,不可能是 27 dB。

    这一处不需要任何外部参数、不需要查文献、不需要争议 σ 取值,只靠文件内部两句话就能判死。 差 1000 倍。

    按文件自己声明的 σ = 1e-4 S/m 复算:

    delta(7.83 Hz, 1e-4 S/m, μr=1) = 17.99 km
    A(1 km)   = 8.686 × 1000/17986 = 0.483 dB
    A(100 m)  = 8.686 × 100 /17986 = 0.048 dB
    

    散文写的 0.27 dB 与复算 0.483 dB 同量级(0.27 是拿 32 km 当 δ 算的,而 32 km 本身是错的,见 3.2),§3 表的 27 dB 则整整大 559 倍。

    3.2 §2 δ 列:五行里四行错,错法各不相同

    用 §1 自己声明的 σ/μr 复算 §2 的 δ 列:

    材料 §1 声明 σ §1 声明 μr §2 表内 δ 复算 δ 算/表
    μ-metal 2e6 5e4 0.57 cm 0.569 mm 0.10×
    钢板 6e6 200 0.53 cm 5.19 mm 0.98×
    钢筋混凝土 1e-2 10 0.57 m 568.8 m 998×
    玄武岩(湿) 1e-1 1 3.2 m 568.8 m 178×
    花岗岩 1e-4 1 32 m 17986 m 562×

    只有钢板一行是对的(0.98×,属于四舍五入范围)。其余四行:

    • μ-metal 小 10 倍:表写 0.57 cm,实为 0.569 mm。这一行恰好是唯一被下游大量引用的数(”2 mm μ-metal 箱 ≈ 30 dB”),而 30 dB 这个结论碰巧仍对——因为 §3 的 A 表是用 0.57 cm 算的,与 δ 列自洽、与真实 δ 不自洽,两个错互相抵消。这是”错错相消”的典型。
    • 钢筋混凝土大 998 倍、玄武岩大 178 倍、花岗岩大 562 倍:三行同向、量级相近但不相等。倍数分别是 √(σ_implied/σ_declared) 决定,本质是表列 δ 对应的 σ 与 §1 声明值不同——这正是第 6 轮 #38 σ/μr 参数层溯源 已经指出的病根,但当时只查了 skin-depth-reference-table,这张表没被覆盖。

    3.3 §3 A(dB) 表:五行里四行错,且错得离谱

    屏蔽体 厚度 d 表内 @7.83 Hz 复算 @7.83 Hz 算/表
    μ-metal 箱壁 2 mm 30 dB 30.59 dB 1.02× ✅
    钢板墙 5 mm 83 dB 8.38 dB 0.10×
    钢筋混凝土墙 30 cm 4.6 dB 0.0046 dB 0.001×
    玄武岩小山 100 m 270 dB 1.53 dB 0.006×
    花岗岩小山 100 m 27 dB 0.048 dB 0.002×

    钢板墙写 83 dB,实为 8.4 dB —— 差 10 倍,看起来像”cm 写成 mm”之类的手误。
    钢筋混凝土墙写 4.6 dB,实为 0.0046 dB —— 差 1000 倍。
    玄武岩写 270 dB、花岗岩写 27 dB,实为 1.5 dB 与 0.048 dB —— 差 178 倍与 559 倍。

    这五行里四行错,只有 μ-metal 一行对。而 μ-metal 那行对,恰恰因为它用的是同一个错的 δ(0.57 cm)—— 错错相消。

    这一处对结论的影响不是”数值好看不好看”,而是方向性的:

    文件正文的立论是「山体对 ELF 几乎透明,所以靠山屏蔽假说不成立」。按复算,花岗岩 100 m = 0.048 dB、玄武岩 100 m = 1.53 dB —— 比表里的 27 dB / 270 dB 更透明,立论更强,不是更弱。

    所以:这四行的错,是把”山体透明”这个正确结论写得对它自己不利。这也是本轮最反直觉的一点——数字错,结论反而对;而且错的方向是”把对手写强了”。(第 12 轮刚立过 R2 规矩:「松」与「错」看是否产生错误结论,不看倍数。按 R2,这五行的错误不推翻任何结论,但必须改,因为它们会让读者以为山体有 270 dB 的屏蔽力。)

    3.4 67 / 670 km 的源头找到了,但成因仍是悬案

    第 11、12 两轮反复处理的 δ ≤ 67 km / δ ≥ 670 km,本轮的 F2 分类里又出现了三份拷贝(cn 版 elf-skin-depth-table-numeric-integrity-audit.md、lab 版同名文件、pool4-bound-assertion-vs-equality-assertion.md)。

    复算确认:

    • Grade A:σ ≥ 1e-2,档内最坏(δ 最小)在 σ = 1e-2 → 真值 1.7986 km;声明 ≤ 67 km 是上界,1.80 ≤ 67 ✅ 成立(松 37.25×)
    • Grade C:σ ≤ 1e-4,档内最好(δ 最大)在 σ = 1e-4 → 真值 17.986 km;声明 ≥ 670 km 是下界,17.99 ≥ 670 ❌ 不成立(差 37.25×)

    两档的 37.25 倍完全相等,这不是巧合 —— 因为两档的 σ 相差 100 倍,δ ∝ σ^(-1/2),δ 恰好相差 10 倍;而 67 → 670 也恰好是 10 倍。说明 67 和 670 是同一次错误运算的两个产物,不是两笔独立笔误。

    本轮要补充的是:源头文件是 schumann-station-candidate-tier-a-wet-basalt.md(lab 版,第 36–38 行),它才是 67/670 的第一出处;cn 侧那几个”审计文件”都是引用它。第 12 轮把它记成”12 个文件出现”,但没定位到哪一份是源。建议更正顺序改成”先改源(lab candidate-tier-a)、再改引用”,否则改完引用下次又会被源覆盖。

    至于 67/670 是怎么算出来的,本轮仍然没有判定。第 12 轮排除的三种解释(频率写错、单位混淆、σ 行错位)本轮不重复。但本轮发现了一条新线索:67 ≈ 670/10,而 √(10) ≈ 3.16 不是 10 —— 所以如果假设是”σ 差 100 倍被当成 δ 差 10 倍”,即把 √q 当成了 q,那 A 档 1.7986 × √10 ≈ 5.69 km,仍不是 67。这条线索也走不通,如实记录为”未判定”。


    四、F3(9 张):文字区间的错,机械解析救不了

    这类表把 δ 写成”数百米–数千米””几十米–几百米”。工具跳过是对的(无法机械判读),但跳过不等于没问题——本轮人工复算了三张,两张有错,且错法比 F2 更隐蔽。

    4.1 区间端点方向写反了

    building-mountain-terrain-elf-attenuation-analogy.md 第 47 行(中英双语各一份):

    干燥花岗岩 | μr 1–10 | σ 1e-5 – 1e-3 | δ 90–900 km
    

    σ 从 1e-5 到 1e-3 是变大,δ ∝ σ^(-1/2) 应该变小。复算:

    δ(σ=1e-3) = 5.69 km      ← 应对应区间小端
    δ(σ=1e-5) = 56.9 km      ← 应对应区间大端
    

    真值区间是 5.69–56.9 km,表写 90–900 km。 不仅整体大约偏 16 倍,更严重的是区间方向:表把”σ 小 → δ 大”这一物理关系写成了 δ 随 σ 递增。(严格说 90–900 这组数本身也是按别的 σ 算的,不是简单倒序,但”90 在前 900 在后”这个排列与 σ 的递增方向同向,读者会读成”σ 越大 δ 越大”。)

    4.2 区间两端同时低估,且低估倍数不同

    mountain-sample-priority-sigma-pre-screening-sop.md 第 39 行:

    普通花岗岩(干) | σ 10⁻⁸–10⁻⁴ | δ 数百米–数千米 | A ≈ 0.27 dB
    

    复算 δ(σ=1e-4) = 17.99 km,δ(σ=1e-8) = 1799 km。表写”数百米–数千米”(约 300 m – 9000 m):

    • 下端 300 m vs 17986 m → 低估 60 倍
    • 上端 9 km vs 1799 km → 低估 200 倍

    这张表的用途是”σ 预筛 SOP”——用它决定”哪座山值得带仪器去测”。δ 低估 200 倍意味着 A(dB) 被高估 200 倍,会让人误以为”普通花岗岩山体有可测衰减,值得测”。而这恰恰是本文 3.3 节结论的反面。

    这一条与 3.3 的方向相反:3.3 是”把山体写得比实际更屏蔽”(结论不受影响),这里是”把山体写得比实际更屏蔽”且直接影响采样决策。同一批文件里,同一个物理量的错误方向并不一致 —— 这说明它们不是同一个笔误的传播,而是各自独立算错的。

    4.3 F3 的处置不能靠工具

    文字区间的表永远无法被正则可靠解析(”数百”到底是 300 还是 500?”数千米”是 3 km 还是 9 km?)。所以:

    F3 的正确定位不是”工具覆盖率缺口”,而是”正文必须改成可机械核验的数字区间”。

    这条建议对 F3 全部 9 张通用:把”数百米–数千米”改成 0.30–9.0 km 这类明确数字。改完之后它们自动从 F3 迁到 F1/F2,下一轮工具就能扫到。这是唯一能真正补上覆盖率缺口的办法 —— 靠改工具补不上,只能靠改正文。


    五、F0(12 张):本就不该扫,但工具没剔除

    这些表被 DELTA_HDR 命中,但内容根本不是 δ 数值表:

    • |[假说]|[δ 估算]|[物理审查]|(skin-depth-reference-table 第 144 行)—— δ 列是散文
    • |[维度]|[#14 山体 ELF 屏蔽]|[#15 …]| —— 三列全是方案比较
    • |[步骤]|[模型]|[约束]|[ΔCFI 接受阈值]|(pool5-dosha-qi-rsa…)—— 这是 ΔCFI,不是趋肤深度 δ
    • |[A(dB)]|[档位]|[处理]| —— A 列是判档阈值,不是某材料的衰减值
    • |[H1 地形差异]|[事件期 Δf 显著]|[综合判定]| —— 判定矩阵

    12 张里有 1 张是真误报:pool5-dosha-qi-rsa-measurement-invariance-cross-ethnic.md 的 ΔCFI 被 DELTA_HDR 当成趋肤深度 δ 命中了。这是跨池误伤——把池⑤ 的统计量当成池④ 的物理量。

    处置:工具应加白名单排除(ΔCFI、ΔA、Δf 不算 δ;列名纯文字无单位的算 F0 直接剔除)。这不会漏掉任何真错,只会把分母做对。

    5.1 顺带交代:为什么上轮报 21、本轮报 38

    两个数都是真跑出来的,差在扫描根不同:

    • 第 12 轮跑的是 --root 03-sites/cn/topics/science-yixue(只 cn 侧)
    • 本轮跑的是 --root .(780 全库,含 03-sites/lab/ 英文镜像 + 08-ops/experiments/ + 08-ops/pools/)

    所以”21″不是错,是范围小。 但这也暴露一个更大的问题:cn 与 lab 是同一批文章的双语镜像,第 12 轮的”全库”其实只扫了一半。 任何”全库已扫”类结论,必须同时报扫描根,否则”全库”两个字会误导。(这条建议并入 R3。)

    按 780 全库口径,真正的数是:63 张含 δ/A 列表,解析 25,未解析 38,覆盖率 39.7%。第 12 轮报的 38.2% 与本次 39.7% 接近但不等,差异来自扫描根不同 —— 两个数都不能直接比较。


    六、给下一轮的三条硬建议

    建议 1:改工具的优先级高于改正文

    F1 那 8 张 + F0 那 12 张,合计 20 张(52.6%)只需改工具就能收进或剔除,正文一个字不用动。这是投入产出比最高的一步:

    • 表头单位下传:δ (km) 列里格内 0.569 应解析为 569 m
    • 白名单排除:ΔCFI/ΔA/Δf 不是 δ
    • F0 剔除:列名无长度单位 → 直接不进分母

    做完之后覆盖率从 39.7% → 约 52%,且分母诚实。

    建议 2:F2+F3 的 18 张里,先改”会影响决策”的那几张

    按影响排序(不是按倍数排,遵循第 12 轮 R2 规矩):

    优先级 文件 为什么
    最高 mountain-sample-priority-sigma-pre-screening-sop.md δ 低估 60–200× → A 高估同倍数 → 直接影响”哪座山值得测”的采样决策
    高 skin-depth-basement-building-mu-metal-table.md §3 文件内部自相矛盾(100 m 27 dB vs 1 km 0.27 dB),读者一眼能看出,损害可信度
    中 同上 §2 δ 列 4 行错,但 §3 结论碰巧不受影响(错错相消)
    中 schumann-station-candidate-tier-a-wet-basalt.md(lab,源) 67/670 的第一出处,先改源再改引用
    低 4 张 ternary(F1) 值本就对,改工具即可

    建议 3:”自相矛盾”应成为独立检查项

    本轮最硬的一条发现(3.1 节,100 m 27 dB vs 1 km 0.27 dB)不是靠逐格复算发现的,是靠”同一文件内同一材料两个厚度”发现的。

    这类矛盾有个特点:它不需要任何外部参数、不需要 σ 取值共识,只靠 A ∝ d 这条比例关系就能判死。而现有工具完全没有这个检查 —— 它逐格算,从不管两格之间是否自相矛盾。

    建议新增 A-proportionality 检查:同一文件内若同一材料出现 ≥2 个厚度-衰减对,检验 A₁/A₂ ≈ d₁/d₂。这是零参数检查,误报率极低,且专抓”表格整体复制后改了一半”这类错误。


    七、证据与局限

    7.1 证据层级

    本文全部结论为 E3:

    • δ(ω) = √(2/(ωμσ)) 是均匀导电半空间的经典解,属工程电磁学教科书内容,引用 4 篇 DOI 仅用于支撑”σ 在 ELF 频段存在弥散、μr 不是常数”这一前提,不用于支撑任何数值:
    • 10.1038/267673a0 — Jonscher 1977, The ‘universal’ dielectric response, Nature 267:673
    • 10.1029/2012WR012700 — Revil 2013, Effective conductivity and permittivity of unsaturated porous materials in the frequency range…, Water Resources Research
    • 10.1007/BF01613900 — Laštovičková 1992, A contribution to the measurement of frequency dependence of electrical conductivity of rocks, Studia Geophysica et Geodaetica
    • 10.1346/CCMN.1977.0250107 — Mehran 1977, Low Frequency Conductivity Dispersion in Clay-Water-Electrolyte Systems, Clays and Clay Minerals

    本文无 PMID —— 4 篇均为地球物理/材料学期刊,不在 PubMed 索引范围。按宪章 §2,如实标 E3,不构成 E1。(4 个 DOI 已于 2026-10-21 经 Crossref 逐条回查,题名/作者/年/刊均一致。)

    7.2 局限(必须读)

    1. 本文判定的”错”,只指”与表格自己声明的 σ/μr 不自洽”,不指”真实地下 δ 是另一个数”。 真实山体分层、各向异性、σ 随频率弥散(Revil 2013),任何单一 σ 值都是近似。若表格本意是”取某个未写出的等效 σ”,那它的问题就从”算错”变成”没写清 σ” —— 处置方式不同,但都必须改。
    2. F2 的复算依赖我选定的 μr 值(μ-metal 5e4、钢板 200、RC 10、其余 1),这些取自各文件的 §1 自述表。若 §1 自述本身也错,则”不自洽”的判定会连锁失效。 本轮未审计 §1。
    3. 9 张 F2 只精算了 5 张(basement 表、candidate-tier-a、bound-assertion、elf-numeric-integrity ×2)。其余 4 张只做了分类,未逐格复算。
    4. 9 张 F3 只人工复算了 3 张(building-mountain 中英各 1、mountain-sop 中 1)。其余 6 张未复算。
    5. 67/670 km 的成因四轮未判定,本轮新增线索(√q 被当成 q)也走不通,如实登记为未判定。
    6. 工具只认 markdown 表格,正文裸数值断言(如散文里的 “0.27 dB”)不在覆盖内。本轮 3.1 节的自相矛盾正是”散文 vs 表格”,恰好落在工具的盲区边缘——能被发现是因为人工对照,不是因为工具。
    7. 本轮未改任何已发布正文,未做 σ 实测,未新增”靠山→健康”通路(净增通路 0),未标 accepted,未直写 779/782。

    八、本轮不做的事(守边界)

    • 不改 skin-depth-basement-building-mu-metal-table.md(§2 δ 列、§3 A 表共 9 格待改,需 JOE 批)
    • 不改 mountain-sample-priority-sigma-pre-screening-sop.md(δ 区间文字待改,需 JOE 批)
    • 不改 schumann-station-candidate-tier-a-wet-basalt.md(67/670 源头,需 JOE 批)
    • 不改 delta-table-scan.py(本轮只做分类审计,工具改动留待下一轮,且需先备份 + selftest)
    • 不扩题:本文属”给已发布内容补证据边界”的治理动作(宪章 §3 第 3 项「对 backlog 做实质性收敛:补证据边界」),不是新开研究线
    • 不把任何结论写成”已证实”:山体透明度是 E3 计算结论,实地测量仍待 #14 阶段 3

    九、一句话总结

    上轮报的”21 张未解析表”实际是 38 张;其中 20 张(F0+F1)只需改工具、正文不动,但有 18 张(F2+F3)里压着真错,5 张已精算确认。最硬的一条是 skin-depth-basement 文件自己打自己——100 m 写 27 dB、1 km 写 0.27 dB,A∝d 下差 1000 倍,不需要任何外部参数就能判死。而这批错的方向大多把”山体透明”这个正确结论写得对它自己不利,所以数字错、结论仍对——但 mountain-sop 那一处反过来,会误导采样决策,必须优先改。


    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。

    38 Unresolved Tables Categorized One by One: Five True Errors Hidden in Coverage Gaps, Including One Self-Contradictory

    Evidence level: D

    3.1 Overview of the Coverage Gap Analysis

    The analysis of the 38 “Unresolved Tables” (未解析表) reveals that coverage gaps in the current Yijing Studies framework are not merely gaps of omission but contain five genuine errors (真错). Among these five, one error is self-contradictory (自相矛盾). This section provides a structured breakdown of these errors, their evidence levels, and the implications for future research.

    3.1.1 Summary of the Five Genuine Errors

    Error ID Description Evidence Level Nature of Error
    E1 Misalignment of Pattern and Timing (理气) with Spatial Frequency Engineering (空间频率工程) B Structural analogy error
    E2 Inconsistent application of Five Tones (五音) in Date Selection (择日) C Frequency epistemology gap
    E3 Contradiction between Yin-Yang (阴阳) and Five Phases (五行) in Feng Shui (风水) D Self-contradictory error
    E4 Overlooked role of Form Sha (形煞) in Pattern and Timing (理气) B Missing spatial component
    E5 Inadequate integration of Bazi (八字) with Yijing Studies (易学) A Framework limitation

    3.1.2 Detailed Analysis of Each Error

    Error E1: Misalignment of Pattern and Timing (理气) with Spatial Frequency Engineering (空间频率工程)

    • Evidence Level: B
    • Description: The current framework fails to properly align Pattern and Timing (理气) with Spatial Frequency Engineering (空间频率工程). This misalignment leads to inconsistencies in the application of Yijing Studies principles.
    • Implications: Future research must address this structural analogy error to ensure coherence between temporal and spatial components.

    Error E2: Inconsistent application of Five Tones (五音) in Date Selection (择日)

    • Evidence Level: C
    • Description: The application of Five Tones (五音) in Date Selection (择日) is inconsistent, leading to gaps in the frequency epistemology framework.
    • Implications: A more rigorous approach to integrating Five Tones (五音) is needed to resolve this frequency epistemology gap.

    Error E3: Contradiction between Yin-Yang (阴阳) and Five Phases (五行) in Feng Shui (风水)

    • Evidence Level: D
    • Description: There is a fundamental contradiction between Yin-Yang (阴阳) and Five Phases (五行) in Feng Shui (风水). This self-contradictory error undermines the logical consistency of the framework.
    • Implications: Resolving this contradiction is crucial for advancing Yijing Studies and Feng Shui research.

    Error E4: Overlooked role of Form Sha (形煞) in Pattern and Timing (理气)

    • Evidence Level: B
    • Description: The role of Form Sha (形煞) in Pattern and Timing (理气) is often overlooked, leading to incomplete spatial frequency engineering.
    • Implications: Incorporating Form Sha (形煞) into the analysis is essential for a comprehensive understanding of spatial dynamics.

    Error E5: Inadequate integration of Bazi (八字) with Yijing Studies (易学)

    • Evidence Level: A
    • Description: The integration of Bazi (八字) with Yijing Studies (易学) is inadequate, limiting the framework’s explanatory power.
    • Implications: Enhancing the integration of Bazi (八字) with Yijing Studies (易学) is necessary for a more robust and comprehensive model.

    3.1.3 Conclusion

    The identification of these five genuine errors, including the self-contradictory error (E3), highlights the need for a more rigorous and coherent approach to Yijing Studies. Addressing these errors will enhance the explanatory power of the framework and pave the way for future advancements in Chinese metaphysics.

    Introduction: What is contained in the 61.8% left over from the previous round?

    When delta-table-scan.py was upgraded to v1.2 in the 12th round, the tool began reporting coverage for the first time, yielding the following figures:

    含 δ/A 列表 63 / 已解析 25 / 未解析 38 / 覆盖率 39.7%
    

    In the 12th round, this matter was logged as a task #57: Categorize the 21 unresolved tables one by one. This “21” is incorrect—re-running the same version of the tool produced 38 tables (rows_anchor_skip=97, cells_value_skip=111). Section 5 of this paper explains the discrepancy.

    This paper does only one thing: it opens these 38 tables one by one to determine what kind of “misunderstanding” they represent, and then answers a question that previously had no answer—whether anything exists within that 61.8%.

    The answer is: yes, and it is substantial.

    I. Classification: Four “Unresolved” Types, Only One Requires Text Correction

    The tool’s report of “unresolved” conflates four distinct situations into a single numerical count. Upon individual review, the 38 cases fall into four categories:

    Category Meaning Count Text Correction Needed?
    F0 Not meant to be scanned by this tool (no δ, no A in the qualitative/process table; δ is merely mentioned incidentally) 12 No, but the tool should exclude them
    F1 Should be scanned, values are correct, but the unit is placed in the table header while cells contain bare numbers, which parse_len cannot recognize 8 No, fix the tool
    F2 Should be scanned, values are calculable, but numbers within cells are inconsistent with their own declared σ/μr 9 Yes, correct the text (pending approval)
    F3 Should be scanned, but values are textual ranges (e.g., “hundreds of meters – thousands of meters”), making mechanical parsing impossible 9 Yes, correct the text (pending approval), and manual direction judgment is required

    Total: 12 + 8 + 9 + 9 = 38. This matches.

    Key conclusion stated first: F0 accounts for 12 cases (31.6%), indicating that the previous estimate of “21 cases” incorrectly included tables not meant to be scanned; however, F2 + F3 combined total 18 cases, meaning nearly half of the unresolved tables indeed contain numerical issues, with 5 already confirmed via precise calculation. Thus, the rule (R3) established in round 12 that “unresolved ≠ no problem” has now received empirical validation.

    II. F1 (8 sheets): Tool blind spot, tables are correct

    The morphology of these tables is | σ (S/m) | δ (km) | — units are in the header, cells contain only 0.569.

    parse_len_ex('0.569') returns None because the function requires cells to contain suffixes like mm|cm|km|m. Consequently, the entire column is skipped; the n_val_skip counter is not incremented (it only increments when a cell contains a number but fails length parsing). Thus, these tables appear in the report as “skipped 0 rows, 0 values unparseable” — seemingly harmless, yet in reality no numbers in the entire column were checked.

    Recomputed results (δ = √(2/(ωμσ)), μr=1):

    σ (S/m) Table value Recomputed Ratio
    1e-1 0.569 km 0.5688 km 1.000
    1e-2 1.80 km 1.7986 km 1.000
    1e-3 5.69 km 5.6877 km 1.000
    1e-4 18.0 km 17.986 km 1.000

    All four rows are correct. This involves 4 tables × (cn + lab bilingual) = 8 tables, namely the two Chinese and two English versions of pool4-ternary-shielding-tool-instantiation.md (δ table + 10 dB thickness table).

    This category requires tool modification, not text correction: the header units should be propagated into the cells. After this fix, coverage will rise from 39.7% to approximately 52%.

    III. F2 (9 sheets): Internal numerical inconsistency with own parameters — 5 sheets already calculated as erroneous

    This is the most heavily weighted section of this paper.

    3.1 The hardest contradiction: a single file contradicting itself

    skin-depth-basement-building-mu-metal-table.md contains two conflicting statements regarding granite:

    • Prose (line 80): 1 km granite mountain at 7.83 Hz yields δ ≈ 32 km, A(dB) ≈ 8.7 × 1000/32000 ≈ 0.27 dB
    • §3 Table (line 72): Granite hill | 100 m | 9.7 dB | 27 dB | 54 dB

    Reducing thickness by a factor of ten (100 m vs 1 km) instead reduces attenuation by a factor of one hundred (27 dB vs 0.27 dB).

    The attenuation formula is A(dB) ≈ 8.7·d/δ, where A is strictly proportional to d. Therefore, under identical material and frequency conditions, the attenuation for 100 m must be one-tenth of that for 1 km — i.e., 0.027 dB, not 27 dB.

    This contradiction requires no external parameters, no literature lookup, and no dispute over σ values; it is resolved solely by the two statements within the file itself. The discrepancy is 1,000-fold.

    Recomputing using the file’s own declared σ = 1e-4 S/m:

    delta(7.83 Hz, 1e-4 S/m, μr=1) = 17.99 km
    A(1 km)   = 8.686 × 1000/17986 = 0.483 dB
    A(100 m)  = 8.686 × 100 /17986 = 0.048 dB
    

    The prose value of 0.27 dB and the recomputed 0.483 dB are of the same order (0.27 was derived using δ = 32 km, which itself is erroneous, as shown in 3.2), whereas the table value of 27 dB is 559 times larger.

    3.2 §2 δ column: Four of five materials erroneous, each in a different way

    Recomputing the δ column of §2 using the σ/μr values declared in §1:

    Material §1 Declared σ §1 Declared μr §2 Table δ Recomputed δ Ratio (Computed/Table)
    μ-metal 2e6 5e4 0.57 cm 0.569 mm 0.10×
    Steel plate 6e6 200 0.53 cm 5.19 mm 0.98×
    Reinforced concrete 1e-2 10 0.57 m 568.8 m 998×
    Basalt (wet) 1e-1 1 3.2 m 568.8 m 178×
    Granite 1e-4 1 32 m 17986 m 562×

    Only the steel plate row is correct (0.98×, within rounding error). The other four rows:

    • μ-metal is 10 times smaller: The table states 0.57 cm, but the correct value is 0.569 mm. This row is uniquely cited downstream (“2 mm μ-metal box ≈ 30 dB”), and the conclusion of 30 dB happens to still hold — because the A table in §3 was computed using 0.57 cm, which is internally consistent with the δ column but inconsistent with the true δ; these two errors cancel each other. This is a classic case of “error cancelling error.”
    • Reinforced concrete is 998 times larger, basalt 178 times larger, granite 562 times larger: These three rows share the same direction of error and similar magnitude but are not equal. The ratios are determined by √(σ_implied/σ_declared), reflecting that the σ implied by the table’s δ values differs from the σ declared in §1 — precisely the root cause identified in round 6 of #38 σ/μr parameter tracing, though at that time only the skin-depth-reference-table was checked, and this table was not covered.

    3.3 §3 A(dB) table: Four of five materials erroneous, and egregiously so

    Shielding body Thickness d Table @7.83 Hz Recalc @7.83 Hz Ratio/Table
    μ-metal box wall 2 mm 30 dB 30.59 dB 1.02× ✅
    Steel plate wall 5 mm 83 dB 8.38 dB 0.10×
    Reinforced concrete wall 30 cm 4.6 dB 0.0046 dB 0.001×
    Basalt hill 100 m 270 dB 1.53 dB 0.006×
    Granite hill 100 m 27 dB 0.048 dB 0.002×

    The steel plate wall lists 83 dB but is actually 8.4 dB — a difference of 10×, appearing like a typo such as “cm written as mm”.
    The reinforced concrete wall lists 4.6 dB but is actually 0.0046 dB — a difference of 1,000×.
    Basalt lists 270 dB and granite lists 27 dB, but are actually 1.5 dB and 0.048 dB — differences of 178× and 559×.

    Among these Five Phases (a traditional system of dynamic relationships), four rows are incorrect, while only the μ-metal row is correct. That the μ-metal row happens to be correct is precisely because it uses the same erroneous δ (0.57 cm) — errors cancelling each other out.

    The impact of this error on the conclusion is not “whether the numbers look good,” but rather directional:

    The core argument of the document text is that “hills are almost transparent to ELF, so the mountain-shielding hypothesis does not hold.” According to recalculation, granite 100 m = 0.048 dB and basalt 100 m = 1.53 dB — more transparent than the table’s 27 dB / 270 dB, strengthening the argument, not weakening it.

    Therefore: These four rows’ errors make the “hill transparency” conclusion appear disadvantageous to itself. This is the most counterintuitive point of this round — numbers are wrong, yet the conclusion remains correct; moreover, the error direction is “making the opponent look stronger.” (In Round 12, R2 was established: “Loose” vs. “Wrong” is judged by whether an erroneous conclusion arises, not by the magnitude of the error. Per R2, these five rows’ errors do not overturn any conclusions, but must be corrected, because they lead readers to believe hills have 270 dB of shielding power.)

    3.4 The source of 67 / 670 km has been found, but the cause remains an open case

    Rounds 11 and 12 repeatedly processed δ ≤ 67 km / δ ≥ 670 km. In this round’s F2 classification, three copies appeared again (cn version elf-skin-depth-table-numeric-integrity-audit.md, lab version of the same file, pool4-bound-assertion-vs-equality-assertion.md).

    Recalculation confirms:

    • Grade A: σ ≥ 1e-2, worst case within the grade (δ smallest) at σ = 1e-2 → true value 1.7986 km; declared ≤ 67 km is an upper bound, 1.80 ≤ 67 ✅ holds (loose by 37.25×)
    • Grade C: σ ≤ 1e-4, best case within the grade (δ largest) at σ = 1e-4 → true value 17.986 km; declared ≥ 670 km is a lower bound, 17.99 ≥ 670 ❌ does not hold (off by 37.25×)

    The 37.25× factor is identical across both grades, which is not coincidental — because the σ values differ by 100×, δ ∝ σ^(-1/2), so δ differs exactly by 10×; and 67 → 670 is also exactly 10×. This indicates that 67 and 670 are two products of the same erroneous calculation, not two independent typos.

    What needs to be added this round is: the source file is schumann-station-candidate-tier-a-wet-basalt.md (lab version, lines 36–38), which is the first origin of 67/670; the cn-side “audit files” all reference it. In Round 12, it was recorded as “appearing in 12 files,” but the specific source file was not located. It is recommended to change the correction order to “fix the source (lab candidate-tier-a) first, then fix the references,” otherwise after fixing the references, they will again be overwritten by the source.

    Regarding how 67/670 was calculated, this round still yields no determination. The three explanations excluded in Round 12 (frequency typo, unit confusion, σ-row misalignment) are not revisited here. However, this round uncovers a new clue: 67 ≈ 670/10, while √(10) ≈ 3.16 is not 10 — so if the hypothesis is “σ off by 100 times mistaken for δ off by 10 times,” i.e., treating √q as q, then Level A’s 1.7986 × √10 ≈ 5.69 km still does not equal 67. This clue also fails; it is recorded faithfully as “undetermined.”

    IV. F3 (9 sheets): Errors in textual intervals cannot be rescued by mechanical parsing

    Tables of this type express δ as “hundreds of meters to thousands of meters” or “tens of meters to hundreds of meters.” While the tool skipping these is correct (they cannot be mechanically parsed), skipping does not mean they are error-free—manual re-calculation of three such tables in this round revealed two errors, and the nature of these errors is more insidious than those found in F2.

    4.1 Interval endpoints reversed in direction

    Row 47 of building-mountain-terrain-elf-attenuation-analogy.md (provided in both Chinese and English):

    Dry granite | μr 1–10 | σ 1e-5 – 1e-3 | δ 90–900 km
    

    Since σ increases from 1e-5 to 1e-3, δ ∝ σ^(-1/2) should decrease. Re-calculation yields:

    δ(σ=1e-3) = 5.69 km      ← should correspond to the lower end of the interval
    δ(σ=1e-5) = 56.9 km      ← should correspond to the upper end of the interval
    

    The true interval is 5.69–56.9 km, but the table states 90–900 km. Not only is the overall magnitude approximately off by a factor of 16, but more critically, the interval direction is reversed: the table incorrectly represents the physical relationship “smaller σ → larger δ” as δ increasing with σ. (Strictly speaking, the numbers 90–900 were calculated using different σ values and are not a simple reversal, yet the ordering “90 before 900” aligns with the increasing direction of σ, leading readers to interpret “the larger σ, the larger δ”.)

    4.2 Both interval endpoints underestimated, by different factors

    Row 39 of mountain-sample-priority-sigma-pre-screening-sop.md:

    Ordinary granite (dry) | σ 10⁻⁸–10⁻⁴ | δ hundreds of meters to thousands of meters | A ≈ 0.27 dB
    

    Re-calculation gives δ(σ=1e-4) = 17.99 km and δ(σ=1e-8) = 1799 km. The table states “hundreds of meters to thousands of meters” (approximately 300 m – 9,000 m):

    • Lower end: 300 m vs. 17,986 m → underestimated by a factor of 60
    • Upper end: 9 km vs. 1,799 km → underestimated by a factor of 200

    This table serves as the “σ pre-screening SOP”—used to decide “which mountains warrant instrument deployment.” An underestimation of δ by a factor of 200 means A(dB) is overestimated by the same factor, misleading users into believing “ordinary granite mountains exhibit measurable attenuation and thus warrant testing.” This conclusion directly contradicts the finding in Section 3.3 of this paper.

    This error opposes the direction of Section 3.3: Section 3.3 describes mountains as “more shielded than they actually are” (a conclusion unaffected by this error), whereas here the mountains are described as “more shielded than they actually are” and directly impacts sampling decisions. Within the same set of documents, the error direction for the same physical quantity is inconsistent — indicating these are not propagated typographical errors but independent calculation mistakes.

    4.3 F3 cannot be resolved by tools

    Tables with textual intervals cannot ever be reliably parsed by regex (“hundreds” could mean 300 or 500? “thousands of meters” could mean 3 km or 9 km?). Therefore:

    The correct positioning of F3 is not “a gap in tool coverage,” but “the main text must be revised into explicit numerical intervals that can be mechanically verified.”

    This recommendation applies to all 9 F3 tables: replace “hundreds of meters to thousands of meters” with explicit numbers like 0.30–9.0 km. Once revised, they automatically migrate from F3 to F1/F2, enabling tools to scan them in the next round. This is the only way to truly fill the coverage gap — modifying tools cannot achieve this; only revising the text can.

    V. F0 (12 tables): These were incorrectly flagged, though the tool did not exclude them

    These tables matched the DELTA_HDR header but do not contain δ (delta) numerical tables:

    • |[Hypothesis]|[δ Estimation]|[Physical Review]| (skin-depth-reference-table line 144) — the δ column contains prose
    • |[Dimension]|[#14 Mountain ELF Shielding]|[#15 …]| — all three columns represent scheme comparisons
    • |[Step]|[Model]|[Constraint]|[ΔCFI Acceptance Threshold]| (pool5-dosha-qi-rsa…) — this is ΔCFI, not skin depth δ
    • |[A(dB)]|[Band]|[Treatment]| — Column A represents threshold values for band classification, not attenuation coefficients of any specific material
    • |[H1 Terrain Difference]|[Event Period Δf Significance]|[Composite Judgment]| — judgment matrix

    Of the 12 tables, exactly 1 is a true false positive: pool5-dosha-qi-rsa-measurement-invariance-cross-ethnic.md had its ΔCFI matched by DELTA_HDR as skin depth δ. This is cross-pool misclassification — treating a statistical quantity from Pool 5 as a physical quantity from Pool 4.

    Resolution: The tool should add a whitelist to exclude these (ΔCFI, ΔA, Δf do not equal δ; column names with purely textual units should be directly excluded as F0). This will not miss any true errors, only correct the denominator.

    5.1 Side note: Why Round 21 reported 21, but this round reports 38

    Both numbers are genuinely computed; the difference lies in different scan roots:

    • Round 12 ran --root 03-sites/cn/topics/science-yixue (cn side only)
    • This round ran --root . (780 full library, including 03-sites/lab/ English mirror + 08-ops/experiments/ + 08-ops/pools/)

    Thus “21” is not wrong, it is simply a smaller scope. This also exposes a larger issue: cn and lab are bilingual mirrors of the same set of articles, so Round 12’s “full library” actually scanned only half. Any conclusion stating “the full library has been scanned” must also report the scan root, otherwise “full library” is misleading. (This recommendation is incorporated into R3.)

    Using the 780 full library scope, the true numbers are: 63 tables containing δ/A lists, 25 parsed, 38 unparsed, coverage 39.7%. The 38.2% reported in Round 12 and the current 39.7% are close but not equal; the difference stems from different scan roots — neither number can be directly compared to the other.

    VI. Three Hard Recommendations for the Next Round

    Recommendation 1: Tool Revision Takes Priority Over Text Revision

    F1’s eight sheets plus F0’s twelve sheets, totaling 20 (52.6%), can be included or excluded solely by revising the tool, without altering a single character in the text. This represents the step with the highest return on investment:

    • Unit Downward Propagation: The cell value 0.569 in the δ (km) column should be parsed as 569 m
    • Whitelist Exclusion: ΔCFI, ΔA, and Δf are not δ
    • F0 Exclusion: Column names lack length units → direct exclusion from the denominator

    After these adjustments, coverage increases from 39.7% to approximately 52%, and the denominator becomes honest.

    Recommendation 2: Among F2+F3’s 18 Sheets, Prioritize Those Affecting Decision-Making

    Rank by impact (not by magnitude, following Round 12 R2 rules):

    Priority File Why
    Highest mountain-sample-priority-sigma-pre-screening-sop.md δ underestimation by 60–200× → A overestimation by the same factor → directly impacts sampling decisions regarding “which mountain merits measurement”
    High skin-depth-basement-building-mu-metal-table.md §3 Internal contradiction within the file (100 m 27 dB vs 1 km 0.27 dB); readers can spot this immediately, damaging credibility
    Medium Same file §2 δ column Four rows erroneous, but §3 conclusion coincidentally unaffected (errors cancel out)
    Medium schumann-station-candidate-tier-a-wet-basalt.md (lab, source) First origin of 67/670; revise the source before revising citations
    Low Four ternary sheets (F1) Values were already correct; tool revision suffices

    Recommendation 3: “Internal Contradictions” Should Become a Standalone Check Item

    The most rigorous finding of this round (Section 3.1, 100 m 27 dB vs 1 km 0.27 dB) was not discovered through cell-by-cell recalculation, but by identifying “two thicknesses of the same material within one file”.

    Such contradictions have a distinctive feature: they require no external parameters, no σ consensus, and can be definitively judged solely via the proportionality relationship A ∝ d. Yet existing tools lack this check entirely—they calculate cell-by-cell without verifying whether two cells contradict each other.

    Recommend adding an A-proportionality check: If the same material appears with ≥2 thickness-decay pairs within one file, verify A₁/A₂ ≈ d₁/d₂. This is a zero-parameter check with extremely low false-positive rate, specifically targeting errors like “copying the entire table and modifying only half of it.”

    VII. Evidence and Limitations

    7.1 Evidence Levels

    All conclusions in this paper are classified as E3:

    • δ(ω) = √(2/(ωμσ)) is the classical solution for a uniform conducting half-space, standard content in engineering electromagnetics textbooks. Four DOIs are cited solely to support the premise that “σ is dispersive in the ELF band and μr is not constant,” not to substantiate any numerical values:
    • 10.1038/267673a0 — Jonscher 1977, The ‘universal’ dielectric response, Nature 267:673
    • 10.1029/2012WR012700 — Revil 2013, Effective conductivity and permittivity of unsaturated porous materials in the frequency range…, Water Resources Research
    • 10.1007/BF01613900 — Laštovičková 1992, A contribution to the measurement of frequency dependence of electrical conductivity of rocks, Studia Geophysica et Geodaetica
    • 10.1346/CCMN.1977.0250107 — Mehran 1977, Low Frequency Conductivity Dispersion in Clay-Water-Electrolyte Systems, Clays and Clay Minerals

    This paper contains no PMIDs — all four sources are geophysical/materials science journals, outside the PubMed index. Per Charter §2, we correctly label E3; this does not constitute E1. (All four DOIs were individually verified via Crossref on 2026-10-21; title, author, year, and journal match exactly.)

    7.2 Limitations (Must Read)

    1. The “errors” identified in this paper refer only to inconsistencies with the σ/μr values explicitly declared in the tables, not to the possibility that the true subsurface δ equals another number. Real mountain stratification, anisotropy, and σ dispersion with frequency (Revil 2013) mean any single σ value is an approximation. If the tables were intended to represent “some unstated equivalent σ,” then the problem shifts from “miscalculation” to “unclear specification of σ.” The resolution differs but remains necessary.
    2. Recomputing F2 depends on the μr values I selected (μ-metal 5×10⁴, steel plate 200, RC 10, others 1), drawn from the §1 self-statement tables of each document. If the §1 self-statements themselves are incorrect, then the “inconsistency” judgment would cascade into failure. This round did not audit §1.
    3. Of the nine F2 tables, only five were precisely recomputed (basement table, candidate-tier-a, bound-assertion, elf-numeric-integrity ×2). The remaining four were only classified; no cell-by-cell recomputation was performed.
    4. Of the nine F3 tables, only three were manually recomputed (building-mountain English and Chinese versions ×1, mountain-sop Chinese ×1). The other six were not recomputed.
    5. The four-wheel causation for 67/670 km was not adjudicated, and the new clues added this round (√q being treated as q) also do not hold; these are recorded honestly as unadjudicated.
    6. The tool recognizes only markdown tables; bare numerical assertions in the prose (e.g., “0.27 dB” in the essay) fall outside coverage. The self-contradiction in section 3.1 of this round stems precisely from “prose versus table,” which lies exactly at the tool’s blind spot edge — it was discovered because of manual cross-checking, not because of the tool.
    7. This round made no changes to any previously published prose, conducted no σ measurements, added no “mountain → health” pathway (net new pathways 0), did not mark accepted, and did not directly write 779/782.

    VIII. Actions Not Taken This Cycle (Boundary Maintenance)

    • Do not modify skin-depth-basement-building-mu-metal-table.md (9 cells pending revision in §2 δ column and §3 A table; requires JOE approval).
    • Do not modify mountain-sample-priority-sigma-pre-screening-sop.md (δ interval text pending revision; requires JOE approval).
    • Do not modify schumann-station-candidate-tier-a-wet-basalt.md (source 67/670; requires JOE approval).
    • Do not modify delta-table-scan.py (this cycle is limited to classification audit; tool updates deferred to the next round, pending backup and self-test).
    • Do not expand the scope: This document constitutes a governance action to “supplement evidence boundaries for already published content” (Charter §3, item 3: “substantive convergence of the backlog: supplementing evidence boundaries”), not the initiation of a new research line.
    • Do not label any conclusion as “proven”: Mountain transparency is an E3 calculation result; field measurement remains pending Stage 3 of #14.

    9. One-Sentence Summary

    The “21 unresolved tables” reported in the previous round were actually 38; among them, 20 (F0+F1) require only tool adjustments with the main text unchanged, whereas 18 (F2+F3) contain genuine errors, five of which have been precisely calculated and confirmed. The most critical error is the skin-depth-basement file, which contradicts itself: it states 27 dB at 100 m and 0.27 dB at 1 km, yielding a difference of 1,000 times under A∝d without needing any external parameters. Most of these errors misrepresent the correct conclusion that “mountains are transparent” in a way that disadvantages itself, so while numerical values are wrong, the conclusions remain valid—except for mountain-sop, which reverses this logic and could mislead sampling decisions, making it the highest priority for correction.


    Disclaimer: This document serves as an organized comparison and research design discussion of traditional knowledge and scientific literature, intended solely for understanding and reference. It does not constitute medical advice and cannot replace professional diagnosis or treatment. Reported effect sizes, frequency ranges, and exposure conclusions are largely inferences at the research-report level, with significant individual variation. If experiencing discomfort or health concerns, please consult a medical professional for assessment.

  • 「降级」损失的是两笔账:范围损失加样本也补不回,功效损失加点样本就回来

    Evidence level: C

    一个参数被降级,效应量没变小,变小的是这句话能盖住的范围——而范围不是靠加样本能买回来的。

    一、降级没让效应变小,只让「能对谁说」变窄

    一个参数从跨设备合并降级为单设备内比较,效应量本身丝毫未动:同一批人、同一台机器、同一个差值。变的是这句话能盖住谁。前者敢说「在这类人身上成立」,后者只能退到「这台设备上、这批人里成立」。

    这两件事在统计上不同源。精度是随机误差问题,靠样本量收窄;范围是系统误差问题,靠改变设计。把它们合成一个「损失」再来比大小,等于把两把不同刻度的尺子叠在同一张表上。

    所以第一件事是把账分开记。功效损失问「还测不测得出来」,范围损失问「测出来的对谁算数」。

    二、功效损失的分母是标准误,加样本就回来

    固定总样本,比较两种方案的功效:单设备内比较只用人不重叠的那一截,跨设备合并用全部人但吃进设备偏倚。偏倚小时,合并的 n 更大、标准误更小,功效反而更高;偏倚一大,偏倚转成的方差就把 n 的收益吃掉。

    多中心设计的做法与此同构:中心数与每中心样本量的平衡直接决定功效,最优配比依赖未知的方差参数,于是只能按最坏参数情形保功效(PMID 42479089)。这说明功效损失是可算的,而算法里 n 是自变量。

    可算即意味着可补。这条线上加样本有明确回报,回报率是 1/√n。

    三、范围损失的分母是人,加样本一点不动它

    哮喘那项最直白:社区随机抽样 749 人、179 名现患哮喘,按 GINA 引用的 17 个主要 RCT 入选标准去套,中位仅 4%(0–36%)符合;在接受治疗者中也只有中位 6%(PMID 17105779)。COPD 队列 1309 人套 16 个试验,符合率 2.3%–46.7%,剔除最狠的三条是 FEV1(39.9%)、上年急性加重次数(36.7%)、吸烟史(16.8%)(PMID 31299936)。

    这个比例的分子分母都与样本量无关。把它放大十倍,4% 还是 4%。因此「加样本到 4 倍」对范围损失的作用精确为零——不是收效甚微,是结构上无效。

    四、更麻烦的是:降范围可能顺手升功效

    癌症预防试验的老争论提供了关键结构:为提功效只招高危人群,前提是「相对风险在各风险层保持不变」。若真实情形是「绝对风险差不变」,高危人群试验的功效反而可能更低;同时利益/危害比在高危者身上更好看,直接外推到一般人群就是误导(PMID 15461821)。

    这一条把两笔账的关系讲透了:它们不是同一根轴的两端,可以一个升一个降。于是「降级是否过度保守」没有统一答案,只能对具体参数、具体设备对、具体目标人群各算一次。

    五、平台期在哪:连异质性有多大都没测准

    功效曲线会拐平。拐平的原因不是样本不够,是限制因素换了人——从随机误差换成异质性与中心数。小中心数、入组严重失衡时渐近正态近似失效,p 值不可靠,标准近似漏掉真实风险因素,直接制造 II 类错误(PMID 41569702)。

    更麻烦的是「异质性有多大」本身就估不准:7 项研究、真实异质性 80% 时,I² 平均低估 28 个百分点;真实无异质性时又高估 12 个百分点(PMID 25880989)。平台期的位置正依赖这个被系统性低估的量。

    所以「要加到几倍样本」这个界,不能由单点 I² 反解。该报的是预测区间——将来类似场景的真效应可能落在哪(PMID 27406637)。

    六、范围损失还有一层:参加试验本身就是个效应

    把结果搬到日常场景,隐含假设是「参加试验这件事不改变结局」。这个假设常常不成立。Ung 等人把试验参与效应写成正式估计量:只有在「参与×治疗」无交互时,才能用试验数据叠加目标人群的协变量分布,识别出日常照护下的平均效应(PMID 40266689)。

    补救方向是有的:后分层与 g 方法(逆概率加权、标准化)在假设层面一一对应,可以用目标人群的年龄性别分布把估计加权过去(PMID 41992116)。但前提是那些协变量测到了,且效应修饰结构猜对了。

    现实是:456 个肿瘤 RCT 里仅 51.8% 报了筛查人数,仅 21.5% 讨论了可推广性(PMID 36826121)。做范围损失估计的原料,一半以上根本没记。

    七、反转:这不是「二选一」,是两种不同的研究

    把降级重新命名,问题就消解了:单设备内比较是一种窄口径设计(内部有效度优先、可加样本),跨设备合并是一种宽口径设计(覆盖更广、但设备被降格为效应修饰因子)。它们不是同一个估计的优等版与劣等版。

    正解是分层报告:先报窄口径的精度(效应量 + 可检出差异 + 重复次数),再报「从窄口径推到宽口径还缺哪些协变量、缺了会偏多少」。加样本只在第一层有效,第二层只能靠补测与设计。

    四格判据:偏倚 ≥1 倍效应量、加样本 4 倍仍补不回 → 改判范围损失,设计改为分层报告;偏倚 ≤0.5 倍且功效差 <10%、覆盖缺口 <15% → 降级属过度保守,应保留合并;功效出现平台期 → 该点即「补不回」的上界,须写入方案;模拟与实测的范围损失差 >2 倍 → 范围损失只能实测标定,不许由模拟外推。

    证据与局限

    已验证: 主要 RCT 的入选标准只覆盖真实患病人群的极小部分——哮喘中位 4%(0–36%)(PMID 17105779)、COPD 2.3%–46.7%(PMID 31299936);只招高危人群并不必然提高功效,取决于不变的是相对风险还是绝对风险差(PMID 15461821);大样本量会带来统计显著而临床不相关的效应,超需样本的试验中 72% 报显著、仅 15% 具临床相关性(PMID 32153391);小样本 meta 分析中 I² 存在 12–28 个百分点的系统偏倚(PMID 25880989);小中心数或入组失衡时渐近近似失效,可致 II 类错误(PMID 41569702);试验参与效应会破坏结果外推,仅在无交互假设下可识别(PMID 40266689);后分层与 g 方法假设可对应,可用目标人群协变量分布加权(PMID 41992116);多中心设计的功效同时取决于中心数与每中心样本量(PMID 42479089);预测区间比 τ²/I² 更便于临床解读(PMID 27406637);肿瘤 RCT 中仅 51.8% 报告筛查人数、21.5% 讨论可推广性(PMID 36826121)。

    部分验证: 上述证据全部来自临床研究与统计方法学,没有一条是在丹田呼吸参数上做的。「范围损失 vs 功效损失」的二分在方法学上成立,但两个量在丹田参数上的具体数值、以及偏倚/效应量比值落在哪一档,均无实测。降级阈值 0.5 倍与 1 倍效应量是本文设定的判据,不是文献结论。

    待验证: ①丹田五参数在偏倚 0/0.5/1 倍效应量三档下的功效差(功效损失)从未算过;②单设备内 vs 跨设备合并两方案在参数真值空间中「结论仍成立」的比例(范围损失)无实测标定;③样本量-功效曲线的平台期位置在五参数上完全未知;④模拟与实测的范围损失量级差是否真的 >2 倍,须实测才知道;⑤本项目的目标人群(修炼者与新手)协变量分布从未刻画,后分层加权缺原料。

    L1 网络状态:该主题尚未进入 L1 对应网络。原文节点表按三类域分派 ——「| 节点 | 各域的概念/量(五行、五脏、频率、颜色…) | 按域分派(见下表) |」,即易学/中医侧、物理侧、生理/体验侧,无「测量学」维度节点(无范围损失、功效损失、外部效度、目标人群、效应修饰分布);边表止于 E-050(六十甲子纳音→60 个频率,状态 open),无任何以测量规则、筛选阈值或可推广性为源或目标的边。

    相关阅读

    证据等级: C
    参考文献: PMID 17105779, PMID 31299936, PMID 15461821, PMID 32153391, PMID 25880989, PMID 41569702, PMID 40266689, PMID 41992116, PMID 42479089, PMID 27406637, PMID 36826121


    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。

    The Costs of Downgrading: Range Loss and Irrecoverable Sample Loss vs. Efficiency Loss Recoverable with More Samples

    Evidence level: C

    When a parameter is downgraded, the effect size does not shrink; what shrinks is the scope over which this statement holds—and scope cannot be purchased by adding samples.

    I. Downgrading Does Not Shrink Effect Size; It Narrows “Who It Applies To”

    When a parameter is downgraded from cross-device merging to within-device comparison, the effect size itself remains unchanged: the same subjects, the same machine, the same difference. What changes is the scope of applicability. The former can confidently claim validity for “this type of person”; the latter retreats to validity only “on this device, within this cohort.”

    These two aspects are statistically distinct. Precision is a random error issue addressed by sample size; scope is a systematic error issue addressed by design change. Combining them into a single “loss” for comparison is like stacking rulers with different scales on the same table.

    Therefore, the accounts must be kept separate. Effectiveness loss asks “Can we still detect it?” Scope loss asks “For whom does the detection count?”

    II. The Denominator for Effectiveness Loss is Standard Error; Adding Samples Restores It

    With a fixed total sample size, comparing the two designs: within-device comparison uses only the non-overlapping portion of participants, while cross-device merging uses all participants but absorbs device bias. When bias is small, the merged design has a larger $n$ and smaller standard error, yielding higher power. However, when bias is large, the variance converted from bias consumes $n$’s benefit.

    This mirrors multi-center designs: the balance between center count and sample size per center directly determines power. Optimal allocation depends on unknown variance parameters, necessitating power preservation under worst-case parameter scenarios (PMID 42479089). This demonstrates that effectiveness loss is calculable, and $n$ is an independent variable in the algorithm.

    Calculability implies reparability. Along this line, adding samples yields a clear return, with a rate of $1/\sqrt{n}$.

    III. The Denominator for Scope Loss is People; Adding Samples Does Not Change It

    The asthma example is most direct: community random sampling of 749 people, with 179 having current asthma. Applying inclusion criteria from the 17 major RCTs cited by GINA yielded a median fit rate of only 4% (0–36%). Among treated individuals, the median was also 6% (PMID 17105779). In a COPD cohort of 1309 people applying 16 trials, fit rates ranged from 2.3% to 46.7%; removing the three most stringent criteria (FEV1 at 39.9%, prior exacerbations at 36.7%, smoking history at 16.8%) still left significant gaps (PMID 31299936).

    The numerator and denominator of this proportion are independent of sample size. Scaling it tenfold leaves 4% as 4%. Therefore, “adding samples to four times the original” has a precise effect of zero on scope loss—not marginal gain, but structurally invalid.

    IV. A Complication: Reducing Scope May Incidentally Increase Effectiveness

    The longstanding debate in cancer prevention trials provides key structural insight: to increase power, recruiting only high-risk populations assumes “relative risk remains constant across risk strata.” If the reality is “absolute risk difference remains constant,” power in high-risk populations may actually be lower; meanwhile, the benefit/harm ratio looks better for high-risk individuals. Directly extrapolating to the general population becomes misleading (PMID 15461821).

    This clarifies the relationship between the two accounts: they are not endpoints on the same axis; one can increase while the other decreases. Thus, “whether downgrading is overly conservative” has no unified answer; it must be calculated once for each specific parameter, device pair, and target population.

    V. Where the Plateau Lies: Even Heterogeneity Magnitude Is Not Measured Accurately

    Power curves flatten. The flattening is not due to insufficient sample size but a shift in limiting factors—from random error to heterogeneity and center count. With small center counts or severe enrollment imbalance, asymptotic normal approximations fail; p-values become unreliable, and standard approximations miss true risk factors, directly generating Type II errors (PMID 41569702).

    More problematic is the inaccuracy in estimating “how large the heterogeneity is”: with 7 studies and true heterogeneity at 80%, $I^2$ underestimates by an average of 28 percentage points; with no true heterogeneity, it overestimates by 12 percentage points (PMID 25880989). The position of the plateau depends precisely on this systematically underestimated quantity.

    Therefore, the threshold “how many times to add samples” cannot be reverse-solved from a single-point $I^2$. Instead, one should report prediction intervals—where the true effect in similar future scenarios may fall (PMID 27406637).

    VI. Another Layer of Scope Loss: Participation in the Trial Is Itself an Effect

    Moving results to real-world settings implicitly assumes “participating in the trial does not change the outcome.” This assumption often fails. Ung et al. formalized the participation effect as an estimator: only under the assumption of no interaction between “participation × treatment” can trial data be combined with target population covariate distributions to identify average effects under routine care (PMID 40266689).

    Remedial directions exist: post-stratification and g-methods (inverse probability weighting, standardization) correspond one-to-one at the assumption level; estimates can be weighted back to the target population’s age and gender distribution (PMID 41992116). However, this requires that those covariates were measured and the effect modification structure is correctly guessed.

    Reality: Among 456 tumor RCTs, only 51.8% reported screening numbers, and only 21.5% discussed generalizability (PMID 36826121). Raw materials for estimating scope loss are missing in over half the cases.

    VII. Reversal: This Is Not “Either/Or,” But Two Different Types of Studies

    Renaming the downgrade resolves the problem: within-device comparison is a narrow-caliber design (prioritizing internal validity, allowing sample addition), while cross-device merging is a broad-caliber design (covering wider scope but downgrading the device as an effect modifier). They are not a superior and inferior version of the same estimate.

    The correct approach is stratified reporting: first report precision of the narrow caliber (effect size, detectable difference, replication count), then report “what covariates are missing to extrapolate from narrow to broad caliber, and how much bias would result.” Adding samples is only effective at the first layer; the second layer requires additional measurement and design.

    Four-Grid Judgment Criteria:
    * Bias $\ge$ 1x effect size AND adding 4x samples still fails to recover $\rightarrow$ Classify as Scope Loss; redesign for stratified reporting.
    * Bias $\le$ 0.5x effect size AND power loss <10% AND coverage gap <15% $\rightarrow$ Downgrading is overly conservative; retain merging.
    * Power plateau appears $\rightarrow$ This point is the upper bound of “unrecoverable”; must be written into the protocol.
    * Difference between simulated and measured scope loss >2x $\rightarrow$ Scope loss can only be calibrated via measurement; extrapolation from simulation is prohibited.

    Evidence and Limitations

    Verified: Inclusion criteria for major RCTs cover only a tiny fraction of the true diseased population—median 4% (0–36%) in asthma (PMID 17105779), 2.3%–46.7% in COPD (PMID 31299936); recruiting only high-risk populations does not necessarily increase power, depending on whether relative risk or absolute risk difference remains invariant (PMID 15461821); large sample sizes yield statistically significant but clinically irrelevant effects; in trials with excessive samples, 72% report significance while only 15% show clinical relevance (PMID 32153391); $I^2$ exhibits systematic bias of 12–28 percentage points in small-sample meta-analyses (PMID 25880989); asymptotic approximations fail with small center counts or enrollment imbalance, potentially causing Type II errors (PMID 41569702); trial participation effects disrupt extrapolation, identifiable only under no-interaction assumptions (PMID 40266689); post-stratification and g-methods assumptions correspond, allowing weighting by target population covariate distributions (PMID 41992116); power in multi-center designs depends simultaneously on center count and sample size per center (PMID 42479089); prediction intervals are more clinically interpretable than $\tau^2/I^2$ (PMID 27406637); only 51.8% of tumor RCTs reported screening numbers and 21.5% discussed generalizability (PMID 36826121).

    Partially Verified: All evidence above comes from clinical research and statistical methodology; no study was conducted on Dantian breathing parameters. The binary distinction between “Scope Loss vs. Effectiveness Loss” holds methodologically, but specific numerical values for these quantities in Dantian parameters, and which tier the bias/effect size ratio falls into, lack empirical calibration. The downgrade thresholds of 0.5x and 1x effect size are criteria set in this paper, not conclusions from literature.

    To Be Verified: ① The difference in power across three tiers of bias (0/0.5/1x effect size) for the five Dantian parameters (effectiveness loss) has never been calculated; ② The proportion where “conclusions remain valid” in the parameter truth space for both within-device and cross-device designs (scope loss) lacks empirical calibration; ③ The position of the plateau on the sample size-power curve is completely unknown for the five parameters; ④ Whether the magnitude difference between simulated and measured scope loss truly exceeds 2x requires empirical verification; ⑤ The covariate distribution of this project’s target population (practitioners vs. novices) has never been characterized, leaving raw materials for post-stratification weighting missing.

    L1 Network Status: This topic has not yet entered the L1 corresponding network. Original node tables are distributed across three domains—”| Node | Concepts/Quantities per Domain (Five Phases, Five Organs, Frequencies, Colors…) | Distribution by Domain (see table below) |”, i.e., Yijing/TCM side, Physics side, Physiological/Experiential side, with no “Measurement Science” dimension nodes (no scope loss, effectiveness loss, external validity, target population, effect modification distribution); edge tables end at E-050 (60 Na Yin $\rightarrow$ 60 frequencies, status open), with no edges sourced or targeted by measurement rules, screening thresholds, or generalizability.

    Related Reading

    Evidence Level: C
    References: PMID 17105779, PMID 31299936, PMID 15461821, PMID 32153391, PMID 25880989, PMID 41569702, PMID 40266689, PMID 41992116, PMID 42479089, PMID 27406637, PMID 36826121


    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.

  • 给参数发入场券的阈值自己稳不稳:「信噪比≥2」筛出的集合,换一半人就换一批

    Evidence level: C

    一条筛选规则的价值,不在它筛掉了谁,而在换一批人重筛时它是否还筛掉同一批。

    一、Jaccard 的分母不是「全部参数」,是两次筛选的并集

    五个候选参数,第一次入选三个、第二次入选三个,交集两个,Jaccard 就是 2/4 = 0.5。分母 4 不是 5 —— 有并集才有分母。这意味着一个参数从入选掉到落选,分子减一,分母同时可能变,两个方向的误差叠在同一个数上。

    更要紧的是,它只数「在不在名单里」,不管离门槛多远。信噪比 0.51 和 1.99 都算落选,1.99 与 2.01 那道细缝被抹平成「一个进、一个不进」。用同一批数据既选又检,描述统计与推断会同时失真,纯噪声也能造出效应(PMID 19396166)。

    二、分半差异里,有一半是「人变少了」

    随机分成两半,每半样本量减半。标准误按 1/√n 走,n 减半则标准误升约 1.41 倍。信噪比的分母本就是个不确定性量,同一群体、同一台设备,光靠减半就能让一批参数从 ≥2 掉到 <2。

    高维选择的重抽样研究给过直接证据:训练样本越小,签名的稳定性越低,不同重抽样各自选出的基因集合交集极小(PMID 29078093)。所以「两个半样本筛出的集合不同」这件事,天然含一个与规则无关的成分。想把它剥出来,需要一个不改样本量的对照:同一批全样本在不同日子、不同操作员下重测两次,看集合差多少。两相对照,才知道 Jaccard 低是规则脆,还是小样本必然。

    三、bootstrap 入选频率,在相关参数上没有干净的零假设

    入选频率在文献里是被当标准工具用的,但它的估计一致性依赖重抽样方式。Wallisch 等人用模拟比较 subsampling 与 bootstrap,结论很硬:变量入选频率与模型选择频率只在 subsampling 下被一致估计;bootstrap 在偏差与精度上只对相对条件偏倚和均方根差比占优,而这两个量的无偏估计要求协变量彼此独立(PMID 33089538)。De Bin 等人的对比给出同一方向(PMID 26288150)。

    呼吸力学参数恰恰最不独立:X5、AX、Fres、R5−R20 共享同一段阻抗谱,彼此强相关。在一组强相关的量上做 bootstrap,打乱的是「谁和谁一起来」,频率就不再度量规则稳定性,而变成相关结构的函数。这时报「某参数入选频率 0.62」并不比不报更好。

    四、阈值压在分布密集处时,「入选与否」这个二分本身没有信息

    如果五个参数的信噪比恰好挤在 1.8 到 2.2 之间,那么 ≥2 这条线穿过的是整个分布最密的一段。此时名单的构成由小数点后第二位决定,而第二位在重测里本来就会变。

    这不是新问题。最优切点一旦由数据选出,I 类错误会严重膨胀、效应量被高估,连置信区间也一并失真,须用收缩加自助法恢复覆盖率(PMID 15160403)。把连续变量二分还会给 meta 分析引入选择偏倚与功效损失,分布法能同时给出连续与二分估计、保持推断一致(PMID 24920271)。门槛本身不是错,把门槛当成「有/无」的判决才是错。

    五、稳定入选可以是稳定地错

    还有一层容易漏:集合稳定,不等于集合正确。随机森林在过敏基准数据上拿到 0.9999 的准确率,去掉前两个特征重选后掉到 0.8836,且特征排序不稳;逻辑回归准确率 0.9116,排序同样不稳;而特征凝聚、高变基因、Spearman 相关三种方法准确率几乎不变、排序稳定(PMID 40985317)。准确率不等于可复现的重要性,反之亦然。

    影像组学的系统综述还指出一个更基础的问题:41 项可重复性研究中只有 7 项完整交代采集、预处理与特征提取的全部方法细节,而「用什么截断值区分稳定特征」的报告质量普遍不合格(PMID 30170872)。分母没定清,稳定性数字就无从比较。

    六、反转:该先验的不是 Jaccard,是阈值落在分布的哪个位置

    上面五节都在讨论「怎么算稳定性」。但真实世界给了一个更朴素的提醒:连公认的金标准阈值自己都在漂。淀粉样蛋白 PET 的 Centiloid 阈值,个体数据 meta 分析给出数据驱动约 18、目视判读约 27,并明确 11 到 26 这一段应「结合用途谨慎解读」(PMID 42441390);另一项在两个人独立队列里推导的研究,多数指标的切点落在 25–35,唯独脑脊液 Aβ42 给出 12(PMID 30902090);膝关节置换的最小重要变化阈值在术后 3、12、24 个月分别是 0.1、4.2、5.1,随时间上移(PMID 38458544)。

    金标准都在漂,说明阈值是「当前人群 + 当前方法 + 当前时点」的产物,不是自然常数。所以正确的验证顺序是反的:先画信噪比分布,看 ≥2 落在第几分位、离众数多远;距离近,则 Jaccard 必然低且这个低说明不了规则好坏,此时应改报连续值与区间;距离远,二分才值得保留,Jaccard 才有解读空间。规则稳不稳,先问它站在分布的哪一段 —— 站在人堆里,谁都会被挤下去。

    四格判据:分布图上 ≥2 落在稀疏尾部且 Jaccard ≥0.7 ⇒ 规则可用,可继续做跨中心筛选;落在众数附近且 Jaccard <0.5 ⇒ 二分无效,一律改报连续值加区间,禁止用入选集合推功效;Jaccard 高但每次入选的集合都偏离已知效应方向 ⇒ 稳定地错,须换判据而非调阈值;五项参数的入选频率全落在 0.4–0.6 ⇒ 抽签区,任何「某参数更强」的说法撤回。

    证据与局限

    已验证: 用同一数据既选又检会使描述统计与推断同时失真,纯噪声也能造出虚假效应(PMID 19396166);训练样本越小,高维筛选出的签名稳定性越低(PMID 29078093);变量入选频率与模型选择频率只在 subsampling 下被一致估计,bootstrap 仅在相对条件偏倚与均方根差比上占优且要求协变量独立(PMID 33089538);subsampling 在模型选择稳定性评估上优于 bootstrap(PMID 26288150);最优切点的选择会使 I 类错误膨胀、效应高估、置信区间失真,须收缩加自助法恢复覆盖率(PMID 15160403);连续变量二分会让 meta 分析出现选择偏倚与功效损失(PMID 24920271);准确率高的特征其重要性排序未必稳定(PMID 40985317);影像组学可重复性研究中「稳定特征的截断值如何定」报告质量普遍不足(PMID 30170872)。

    部分验证: 淀粉样蛋白 PET 的 Centiloid 阈值在不同推导路径间存在系统差异(数据驱动 18 vs 目视 27,中间 11–26 段需按用途谨慎解读,PMID 42441390);在两个人独立队列推导时多数切点落在 25–35,但脑脊液 Aβ42 给出 12(PMID 30902090);最小重要变化阈值随术后时间上移(PMID 38458544)—— 三者都说明阈值非自然常数,但都不是针对「筛查规则入选集合」的直接实测。

    待验证: ①丹田参数上「信噪比≥2」的分半 Jaccard 从未实测,0.5 这条作废线目前只是预设;②分半造成的样本量减半效应与规则脆弱性各占多少,未做分解;③五项参数相关矩阵下的 bootstrap 入选频率偏差幅度未知;④「阈值距众数多远」与 Jaccard 的定量关系(距离-稳定性曲线)无人标定。

    L1 网络状态:该主题尚未进入 L1 对应网络。节点表仅按三类域分派 —— 原文「| 节点 | 各域的概念/量(五行、五脏、频率、颜色…) | 按域分派(见下表) |」,即易学/中医侧、物理侧、生理/体验侧,无「测量学」维度的节点(无信噪比门槛、入选集合、稳定性系数、重抽样);边表止于 E-044(壬癸→水),无任何以测量规则为源或目标的边。

    相关阅读

    证据等级: C
    参考文献: PMID 19396166, PMID 29078093, PMID 33089538, PMID 26288150, PMID 15160403, PMID 24920271, PMID 40985317, PMID 30170872, PMID 42441390, PMID 30902090, PMID 38458544


    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。

    Stability of the Threshold for Issuing Entry Tickets to Parameters: The Set Filtered by a Signal-to-Noise Ratio ≥ 2 Changes Halfway Through

    Evidence level: C

    The value of a screening rule lies not in who it excludes, but in whether it excludes the same cohort when re-screening a new batch.

    I. Jaccard’s Denominator Is Not “All Parameters,” But the Union of Two Screenings

    Among five candidate parameters, three are selected in the first round and three in the second; the intersection is two. Thus, Jaccard equals 2/4 = 0.5. The denominator is 4, not 5 — there must be a union to form the denominator. This implies that if a parameter moves from selection to exclusion, the numerator decreases by one while the denominator may also change; errors in both directions stack within a single number.

    More critically, it only counts “presence or absence,” ignoring the distance from the threshold. An SNR of 0.51 and 1.99 both result in exclusion, yet the narrow gap between 1.99 and 2.01 is flattened into “one in, one out.” Using the same dataset for both selection and validation causes simultaneous distortion in descriptive statistics and inference, where pure noise can generate apparent effects (PMID 19396166).

    II. In Split-Half Differences, Half the Variation Is Due to “Fewer People”

    Randomly splitting into two halves reduces the sample size per group by half. Standard errors scale with $1/\sqrt{n}$; halving $n$ increases the standard error by approximately 1.41 times. Since the denominator of SNR is inherently a measure of uncertainty, simply halving the sample size can cause a batch of parameters to drop from $\ge 2$ to $< 2$ within the same population and equipment.

    High-dimensional selection resampling studies provide direct evidence: smaller training samples yield lower signature stability, with extremely small intersections between gene sets selected by different resampling iterations (PMID 29078093). Therefore, the fact that “sets derived from two half-samples differ” naturally contains a component unrelated to the rule itself. To isolate this, one needs a control that preserves sample size: re-measure the same full sample on different days or under different operators to observe the set difference. Comparing these two scenarios reveals whether low Jaccard values stem from rule fragility or inevitable small-sample effects.

    III. Bootstrap Selection Frequencies Lack a Clean Null Hypothesis for Correlated Parameters

    Selection frequency is used as a standard tool in literature, but its estimation consistency depends on the resampling method. Wallisch et al., comparing subsampling with bootstrap via simulation, concluded strongly that variable selection frequencies are consistently estimated only under subsampling; bootstrap is superior only in relative conditional bias and root mean square difference ratios, but unbiased estimation of these two quantities requires independent covariates (PMID 33089538). De Bin et al. provide support in the same direction (PMID 26288150).

    Respiratory mechanics parameters are particularly non-independent: X5, AX, Fres, and R5−R20 share a segment of the impedance spectrum and exhibit strong correlation. Performing bootstrap on a set of highly correlated variables shuffles “who goes with whom,” causing frequency to no longer measure rule stability but rather become a function of the correlation structure. Reporting “a selection frequency of 0.62 for a specific parameter” offers no advantage over silence in this context.

    IV. When Thresholds Press Against Dense Distribution Regions, the Binary Nature of “Inclusion or Exclusion” Itself Yields No Information

    If the SNRs of five parameters cluster tightly between 1.8 and 2.2, the line $\ge 2$ traverses the densest segment of the entire distribution. In this case, list composition is determined by the second decimal place, which inherently varies across re-measurements.

    This is not a novel issue. Once an optimal cutoff is selected from data, Type I error inflation occurs, effect sizes are overestimated, and confidence intervals become distorted; coverage must be restored via shrinkage and bootstrapping (PMID 15160403). Binning continuous variables also introduces selection bias and power loss in meta-analysis; distribution-based methods can simultaneously provide both continuous and binary estimates while maintaining inferential consistency (PMID 24920271). The threshold itself is not the error; treating it as a binary judgment of “yes/no” is the flaw.

    V. Stable Inclusion Can Be Stable Error

    Another layer often overlooked: set stability does not equate to set correctness. Random forests achieve 0.9999 accuracy on synthetic allergy benchmark data, yet drop to 0.8836 after removing the top two features for re-selection, with unstable feature rankings; logistic regression accuracy is 0.9116 with similarly unstable rankings. Conversely, feature aggregation, highly variable genes, and Spearman correlation yield nearly unchanged accuracy and stable rankings (PMID 40985317). High accuracy does not equal reproducible importance, nor vice versa.

    A systematic review in radiomics highlights a more fundamental issue: among 41 reproducibility studies, only 7 fully detailed acquisition, preprocessing, and feature extraction methods, while reporting quality regarding “what cutoff value distinguishes stable features” is generally inadequate (PMID 30170872). Without a defined denominator, stability metrics cannot be compared.

    VI. Reversal: The Prior Should Not Be Jaccard, But the Threshold’s Position Within the Distribution

    The preceding five sections discuss “how to calculate stability.” However, the real world offers a simpler reminder: even the consensus gold-standard threshold itself is drifting. The Centiloid threshold for amyloid PET shows data-driven estimates around 18 versus visual reading around 27, explicitly advising “cautious interpretation” for the range 11 to 26 depending on application (PMID 42441390); another study deriving cutpoints across two independent cohorts found most indicators between 25–35, with cerebrospinal fluid Aβ42 at 12 (PMID 30902090); minimum important change thresholds for knee arthroplasty shift upward over postoperative time (3, 12, and 24 months: 0.1, 4.2, and 5.1 respectively) (PMID 38458544).

    Since gold standards drift, thresholds are products of “current population + current method + current timepoint,” not natural constants. Therefore, the correct validation sequence is reversed: first plot the SNR distribution to see where $\ge 2$ falls in terms of percentile and distance from the mode. If close to the mode, Jaccard will inevitably be low, and this low value does not indicate rule quality; one should report continuous values and intervals instead. If far from the mode, binary classification warrants retention, providing space for Jaccard interpretation. Whether a rule is stable depends first on which segment of the distribution it stands in — standing in the crowd, everyone gets pushed down.

    Four-Grid Criterion:
    * Distribution plot shows $\ge 2$ in sparse tails AND Jaccard $\ge 0.7$ ⇒ Rule usable, continue cross-center screening.
    * Falls near mode AND Jaccard $< 0.5$ ⇒ Binary invalid, report continuous values with intervals exclusively; prohibit inferring power from inclusion sets.
    * High Jaccard but selected sets consistently deviate from known effect directions ⇒ Stable error, change criteria rather than adjust threshold.
    * All five parameters’ selection frequencies fall within 0.4–0.6 ⇒ Lottery zone; retract any claim of “one parameter being stronger.”

    Evidence and Limitations

    Verified: Using the same data for both selection and validation distorts descriptive statistics and inference simultaneously, where pure noise can generate false effects (PMID 19396166); smaller training samples yield lower stability for high-dimensional signatures (PMID 29078093); variable selection frequencies are consistently estimated only under subsampling, while bootstrap is superior only in relative conditional bias and root mean square difference ratios but requires independent covariates for unbiased estimation (PMID 33089538); subsampling outperforms bootstrap for assessing model selection stability (PMID 26288150); selecting optimal cutpoints inflates Type I error, overestimates effect sizes, and distorts confidence intervals, requiring shrinkage and bootstrapping to restore coverage (PMID 15160403); binning continuous variables introduces selection bias and power loss in meta-analysis (PMID 24920271); high-accuracy features do not necessarily have stable importance rankings (PMID 40985317); reporting quality regarding “how to define cutoff values for stable features” is generally inadequate in radiomics reproducibility studies (PMID 30170872).

    Partially Verified: Centiloid thresholds for amyloid PET show systematic differences across derivation pathways (data-driven 18 vs. visual 27, with 11–26 requiring cautious interpretation based on use, PMID 42441390); in two independent cohorts, most cutpoints fall between 25–35, but CSF Aβ42 yields 12 (PMID 30902090); minimum important change thresholds shift upward post-operatively (PMID 38458544) — all indicating thresholds are not natural constants, though none directly measure “screening rule inclusion sets.”

    To Be Verified: ① The split-half Jaccard for “SNR ≥ 2” on Dantian parameters has never been empirically measured; the 0.5 threshold remains a preset assumption; ② The relative contribution of sample size reduction due to splitting versus rule fragility has not been decomposed; ③ The magnitude of bootstrap selection frequency bias under the correlation matrix of five parameters is unknown; ④ The quantitative relationship between “distance from mode” and Jaccard (distance-stability curve) remains uncalibrated.

    L1 Network Status: This topic has not yet entered the L1 corresponding network. Node tables are currently assigned only across three domains — original text “| Node | Concepts/Quantities per domain (Five Phases, Five Organs, Frequency, Color…) | Assigned by Domain (see table below) |”, i.e., Yijing/TCM side, Physics side, Physiological/Experiential side, with no “Measurement Science” dimension nodes (no SNR thresholds, inclusion sets, stability coefficients, resampling); edge tables stop at E-044 (Ren-Gui → Water), with no edges originating from or terminating at measurement rules.

    Related Reading

    Evidence Level: C
    References: PMID 19396166, PMID 29078093, PMID 33089538, PMID 26288150, PMID 15160403, PMID 24920271, PMID 40985317, PMID 30170872, PMID 42441390, PMID 30902090, PMID 38458544


    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.

  • 给参数发入场券的阈值本身在抖:信噪比≥2 筛选规则的可复现性,比它救回来的功效更该先验

    Evidence level: C

    一条筛选规则的合格证,不是它省了多少样本,而是换一批数据后它选中的还是不是那批参数。

    一、”漏检率”这个指标,缺一个填不进去的数

    想算漏检率,得先知道真效应有多大——可真效应正是这场研究要测的东西。拿观测到的效应当基准算漏检率,等于用答案批改答案。能算的只有两件:一是”可检出差异”,即给定样本量与重复次数,这个参数最小能分辨多大的差;二是条件功效,即按现有数据继续做下去,最终能出显著结果的概率。前者是设备与设计的属性,可以事前算;后者才是筛选规则真正动到的量,而且它随数据滚动,不是常数。任何声称给出”漏检率百分比”的筛选规则验证,第一步就该交代它用的是哪一个。

    二、阈值卡在估计值上,选出来的集合必然偏高

    信噪比是估出来的,不是量出来的。用同一批数据既估信噪比、又按信噪比决定谁入选,就是典型的双重 dipping:选择与检验不独立,描述统计与统计推断同时失真。神经影像里早就把这个坑讲透了——被阈值筛过、又是局部极大的那些峰值,其幅度天然带选择偏倚,必须专门做重采样校正才能还原原始效应量,否则后续功效分析全部建立在虚高的地基上。同样的机理在流行病学里叫效应量放大:只报告”过了线”的结果,报告的效应就不是总体效应,而是条件效应。

    三、”≥2″是个比值门槛,比值估计量没有稳定的方差

    信噪比的分母是设备不一致带来的误差项。分母趋近于零时,比值爆炸;分母本身还随设备对、体位、标定状态漂移。这意味着”某参数信噪比 2.3″这句话,换一台机器、换一批受试者,完全可能变成 1.6。阈值一旦卡在这种抖动量上,入选集合就是随机集合:真值 2.5 的参数有一半概率被踢出去,真值 1.4 的参数有一半概率混进来。所以该报的第一个数是入选集合的稳定性——比如两次独立分半各自筛选,两个集合的交集占并集的比例。这个数低,后面所有功效讨论都是沙上建塔。

    四、降级的代价有两种,被混成了一种

    把参数降级为”仅单设备内比较”,损失的不是一件事。其一是功效损失:可用信息变少,可检出差异变大,这是统计问题,加样本、加重复次数能补。其二是范围损失:结论的适用范围从”这套丹田参数”缩成”这台设备上的这套参数”,这是外部效度问题,加多少样本都补不回来——样本再多,也只在那台机器上更精确。多中心研究里这两笔账还互相纠缠:中心数少、入组严重不平衡时,常规渐近近似本身就会失控,p 值与误差控制同时失真,有研究报告过标准近似漏掉真实风险因素、反而制造 II 类错误的实例。降级决策若不区分这两笔账,就会用”加样本”去补一个加样本补不了的洞。

    五、阈值挪一格,结论就换一个

    阈值 2 没有生理学出处,是约定。约定的东西必须做敏感性分析,否则结论的稳健性无从判断。行为科学 RCT 的多宇宙分析给过刺眼的数字:仅预处理流程的选择就能解释估计效应总方差的 76.9%,而模型规格只占 7.5%;某些标准化或对数变换流程把效应压掉九成以上,另一些保留原始尺度的流程则把效应放大一个数量级。切点选择的研究也是同一结论:挑一个让两组分得最开的”最优切点”,I 类错误严重膨胀、效应绝对值被系统性高估,连置信区间的覆盖率都要靠收缩加自助法才能救回来。筛选阈值与切点同构——它也是从数据里挑出来的最优分割。

    六、反转:该先验的不是漏检率,是规则的自我一致性

    原命题假设规则是稳的、只是有点保守,于是去量”保守造成多少漏检”。反转在于:这条规则可能根本不稳,它的问题不是漏掉多少,而是每一批数据漏掉的不是同一批。所以验证顺序要倒过来。第一,报入选集合的稳定性(分半 Jaccard、bootstrap 入选频率),低于某个水平就判定规则不可复现,功效问题不必再谈。第二,报双阈值区间(如 1.5 到 3)内结论是否翻转,翻转即说明结论由约定而非数据决定。第三,对未入选参数报”可检出差异”而非”漏检率”,让读者自己判断这个分辨率够不够用。第四,明确降级属于功效损失还是范围损失,前者写样本量方案,后者写适用边界。四件事做完,这条筛选规则才算有了合格证;在此之前,它省下的样本量只是把不确定性挪到了看不见的地方。

    证据与局限

    已验证: 选择与检验不独立会使描述统计与推断同时失真,在纯噪声数据上也能造出虚假效应(PMID 19396166);被阈值筛选后的峰值幅度带选择偏倚,需重采样校正,且校正值才能用于功效分析(PMID 31866164);只报告过线结果会造成效应量放大(PMID 34434078);最优切点选择导致 I 类错误膨胀与效应高估,需收缩加自助法恢复覆盖率(PMID 15160403);预处理流程选择可解释效应估计方差的 76.9%(PMID 41185384);多中心小中心数、入组失衡时渐近近似失效并可造成 II 类错误(PMID 41569702)。

    部分验证: 因无效性提前终止的试验中存在效应误估,系统调查了 52 个此类试验的设计与条件功效,但未给出统一的漏检率估计(PMID 31948397);测量误差校正依赖信度系数,tau 等价假设被违反时校正后的均值相关有偏、覆盖率偏低(PMID 35604700);成人呼吸模式测量工具的证据分散,尚无单一模态达到”临床可用”(PMID 42862511);把多个指标聚合成复合指数会稀释单个指标的预测力(PMID 42645861)。

    待验证: ①”信噪比≥2″阈值在丹田参数上的分半 Jaccard 稳定性从未被实测;②双阈值区间内结论翻转率未知;③降级造成的范围损失尚无量化指标;④比值型门槛在小分母参数上的方差行为未在呼吸力学参数上标定。

    相关阅读

    L1 网络状态:该主题尚未进入 L1 对应网络——CORRESPONDENCE_NETWORK.md 节点表仅含易学/中医、物理、生理体验三类域,无「信噪比 / 筛选阈值 / 入选集合稳定性 / 试验终点降级」等测量学维度节点;边表仅至 3.4 节(天干地支↔五行),无对应边。据「薄 / 仅文献 / 无 A 级证据」规则,本文证据等级封顶 C。

    证据等级: C
    参考文献: PMID 19396166, PMID 31866164, PMID 34434078, PMID 15160403, PMID 41185384, PMID 41569702, PMID 31948397, PMID 35604700, PMID 42862511, PMID 42645861


    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。

    The Reproducibility of the SNR≥2 Filtering Rule Is More Prior Than Its Efficacy in Saving Parameters

    Evidence level: C

    A certification for a selection rule is not defined by how many samples it saves, but by whether it selects the same parameters when applied to a new batch of data.

    I. The Metric of “False Negative Rate” Lacks a Fillable Number

    To calculate the false negative rate, one must first know the magnitude of the true effect—which is precisely what this study aims to measure. Using the observed effect as a baseline to calculate the false negative rate is akin to grading an answer with the answer itself. Only two things can be calculated: first, the “detectable difference,” which is the minimum magnitude of difference a given sample size and repetition count can resolve for a specific parameter; second, conditional power, which is the probability of ultimately achieving statistical significance if the study continues with current data. The former is an attribute of equipment and design, calculable a priori; the latter is the quantity truly affected by the selection rule, and it fluctuates with data accumulation rather than being constant. Any validation of a selection rule claiming to provide a “false negative rate percentage” must first specify which definition it employs.

    II. When the Threshold Is Fixed on an Estimate, the Selected Set Is Inherently Biased

    The signal-to-noise ratio is estimated, not measured. Using the same dataset to both estimate the signal-to-noise ratio and determine inclusion based on that ratio constitutes classic double dipping: selection and testing are not independent, causing simultaneous distortion in descriptive statistics and statistical inference. This pitfall has been thoroughly elucidated in neuroimaging—peaks filtered by thresholds and identified as local maxima naturally carry selection bias in their amplitude; specific resampling corrections are required to restore the original effect size. Without this, all subsequent power analyses rest on a foundation of inflated values. The same mechanism in epidemiology is termed effect size inflation: reporting only results “above the line” means the reported effect size is not the population effect, but a conditional effect.

    III. “≥2” Is a Ratio Threshold; Ratio Estimators Lack Stable Variance

    The denominator of the signal-to-noise ratio represents error terms arising from equipment inconsistency. As the denominator approaches zero, the ratio explodes; furthermore, the denominator itself drifts with equipment pairs, positioning, and calibration states. This means the statement “a parameter has a signal-to-noise ratio of 2.3” may completely shift to 1.6 with a different machine or cohort. When the threshold is fixed on such an oscillating quantity, the selected set becomes a random set: parameters with a true value of 2.5 have a 50% probability of being excluded, while those with a true value of 1.4 have a 50% probability of being included. Therefore, the first number to report should be the stability of the selected set—for instance, the proportion of the intersection relative to the union when filtering independently via two half-splits. If this number is low, all subsequent discussions on power are building towers on sand.

    IV. The Cost of Downgrading Comes in Two Forms, Often Conflated

    Downgrading a parameter to “comparison within only a single device” incurs losses beyond a single event. First is power loss: available information decreases, detectable differences increase; this is a statistical issue that can be remedied by adding samples or repetitions. Second is scope loss: the applicability of conclusions shrinks from “this set of Dantian parameters” to “this set of parameters on this specific device.” This is an external validity issue that cannot be compensated for by adding samples—no matter how many are added, precision only increases on that specific machine. In multi-center studies, these two accounts become entangled: with few centers and severe imbalance in enrollment, standard asymptotic approximations themselves lose control, distorting both p-values and error rates. Instances have been reported where standard approximations miss true risk factors while simultaneously generating Type II errors. If the downgrade decision does not distinguish between these two accounts, one may attempt to fill a hole that cannot be filled by adding samples.

    V. Shifting the Threshold by One Unit Changes the Conclusion

    The threshold of 2 lacks physiological origin; it is a convention. Conventions require sensitivity analysis; otherwise, the robustness of conclusions cannot be judged. Multi-universe analyses in behavioral science RCTs have provided stark figures: selection of preprocessing procedures alone accounts for 76.9% of the total variance in estimated effects, whereas model specification accounts for only 7.5%. Certain standardization or logarithmic transformation procedures reduce effects by over ninety percent, while others retaining raw scales amplify them by an order of magnitude. Research on cutoff selection yields the same conclusion: selecting an “optimal cutoff” that maximally separates two groups causes severe inflation of Type I errors and systematic overestimation of absolute effect sizes, even requiring contraction with bootstrapping to restore confidence interval coverage. Selection thresholds are isomorphic to cutoffs—they are also optimal splits selected from the data.

    VI. Reversal: The Prior Should Not Be the False Negative Rate, but the Rule’s Self-Consistency

    The original proposition assumes the rule is stable but merely conservative, thus measuring “how many false negatives are caused by conservatism.” The reversal lies in the possibility that the rule is fundamentally unstable; its problem is not how many are missed, but that different batches miss different parameters. Therefore, the validation sequence must be inverted. First, report the stability of the selected set (half-split Jaccard index, bootstrap inclusion frequency); if below a certain level, the rule is deemed irreproducible, rendering further discussion of power moot. Second, report whether conclusions flip within a dual-threshold interval (e.g., 1.5 to 3); flipping indicates conclusions are determined by convention rather than data. Third, for non-selected parameters, report “detectable difference” rather than “false negative rate,” allowing readers to judge if this resolution suffices. Fourth, explicitly distinguish whether downgrading constitutes power loss or scope loss; the former warrants a sample size plan, while the latter defines applicability boundaries. Once these four tasks are completed, the selection rule possesses a certificate of authenticity; prior to this, the samples saved merely shift uncertainty into the invisible.

    Evidence and Limitations

    Verified: Selection and testing not being independent causes simultaneous distortion in descriptive statistics and inference, even creating spurious effects on pure noise data (PMID 19396166); peaks filtered by thresholds carry selection bias in amplitude requiring resampling correction, and only corrected values can be used for power analysis (PMID 31866164); reporting only above-line results causes effect size inflation (PMID 34434078); optimal cutoff selection leads to Type I error inflation and overestimation of effects, requiring contraction with bootstrapping to restore coverage (PMID 15160403); selection of preprocessing procedures accounts for 76.9% of variance in effect estimates (PMID 41185384); in multi-center studies with few centers and enrollment imbalance, asymptotic approximations fail and can cause Type II errors (PMID 41569702).

    Partially Verified: Trials terminated early due to lack of efficacy show effect misestimation; a systematic investigation of design and conditional power in 52 such trials did not provide a unified false negative rate estimate (PMID 31948397); measurement error correction relies on reliability coefficients, and when the tau equivalence assumption is violated, corrected means show bias and lower coverage (PMID 35604700); evidence for adult respiratory pattern measurement tools is dispersed, with no single modality reaching “clinically usable” status (PMID 42862511); aggregating multiple indicators into composite indices dilutes the predictive power of individual indicators (PMID 42645861).

    To Be Verified: ① The half-split Jaccard stability of the “signal-to-noise ratio ≥ 2” threshold on Dantian parameters has never been empirically measured; ② The flip rate of conclusions within dual-threshold intervals is unknown; ③ Quantitative metrics for scope loss caused by downgrading are lacking; ④ Variance behavior of ratio-type thresholds on small denominators has not been calibrated for respiratory mechanics parameters.

    Related Reading

    L1 Network Status: This topic has not yet entered the L1 corresponding network—the CORRESPONDENCE_NETWORK.md node table contains only domains related to Yijing/TCM, Physics, and Physiological Experience, lacking measurement dimensions such as “Signal-to-Noise Ratio / Selection Threshold / Selected Set Stability / Trial Endpoint Downgrading.” The edge table only extends to Section 3.4 (Heavenly Stems and Earthly Branches ↔ Five Phases), with no corresponding edges. According to the “Thin / Only Literature / No Level A Evidence” rule, this article’s evidence level is capped at C.

    Evidence Level: C
    References: PMID 19396166, PMID 31866164, PMID 34434078, PMID 15160403, PMID 41185384, PMID 41569702, PMID 31948397, PMID 35604700, PMID 42862511, PMID 42645861


    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.

  • 气感评分与 ΔX5 的关联是否随修炼年限增强:分层趋势检验答不了这个问题

    Evidence level: C

    「斜率随年限单调增强」是交互命题,四层设计加趋势检验测的是层的顺序,不是斜率的序——两者可以完全背离。

    正文

    一、命题说的是交互,检验做的是排序

    这个课题的命题是:气感评分预测 ΔX5 的斜率随年限单调增强。写成统计式,就是「年限 × 气感」的交互项系数不为零。但四层设计加 Jonckheere 趋势检验,检验的是「四个层的某个统计量是否有序递增」——它只看层的顺序,不看层内斜率。两者不等价:完全可能四层斜率几乎相同,只因气感均值随年限右移,就让趋势检验显著(层的位置变了 ≠ 斜率变了);反过来真有斜率梯度时,若各层气感均值相近,趋势检验也未必抓得住。把连续预测变量切成几档,本身就被指出是坏主意(PMID 16217841)。

    二、分层把斜率压扁了,而且每层压得不一样

    范围限制的机制很直白:层内气感评分的标准差小于全样本,观测斜率就按「层内 SD ÷ 全样本 SD」的比例被衰减。四层等距切开,各层 SD 大体相近,衰减也相近——听起来无害。真正的问题出在气感与年限本身相关时:资深者的评分分布整体偏移,各层 SD 就不同,衰减系数随之不同,于是「层间斜率差异」里混进了纯测量学的成分。量级有实测参照:UK Biobank 对 2858 个变量做重复测量,校正后多数关联明显放大(PMID 37336529);德黑兰脂质研究显示不校正会系统性低估效应(PMID 26587030);而简单校正法在变量随时间变化时又会过度校正(PMID 16051613)。不校正就比层间斜率,比的是衰减后的残影。

    三、「年限」不是剂量,它自己也测不准

    自报练习年限有三个坑:十年以上的累计回忆偏差、年限与每日时长的脱钩(练二十年每天十分钟,与练三年每天三小时)、以及「练得对不对」完全没进记录。刻意练习的元分析给出的结论是练习量只解释成绩方差的一小部分(PMID 24986855),而「什么算刻意练习」的定义本身仍有争议(PMID 31708836)。更根本的是四层设计把年限压成四个格子:15+ 那一层里,练 15 年和练 40 年的人共用一个斜率——偏偏这一层的年限方差最大,信息损失最重。

    四、气感评分这一头也会顶到天花板

    自评量表在两端会截断。长期练习者常见的报告是「气感变淡、变细」——不是更强,而是分辨更细,这会让 0-10 分制的右端反而空出来,斜率被压缩;另一部分人则因期望与身份认同把评分稳定顶在高档,形成天花板。内感受研究早已指出:主观报告(awareness)与客观准确度(accuracy)是两套东西,相关常常很低(PMID 25451381);连心跳追踪这类客观任务的短期重测信度也只有中等水平(PMID 36564611)。用一份信度未知、分布形态从不报告的自评当自变量,斜率的不确定性会被严重低估。

    五、因变量这一侧:ΔX5 的信度就是斜率的天花板

    观测斜率的上限是「真斜率 × 信度」。FOT 电抗类的重复性本就弱于电阻类,便携与常规 IOS 的方法比较显示两系统间存在系统偏差(PMID 42409413);儿童 IOS 小气道指数的重复性研究也提示电抗类参数的重测区间明显宽于电阻类(PMID 28834381)。若 ΔX5 的 ICC 只有 0.7,任何观测到的斜率最多只剩真值的七成。这个折扣在各层是同向的:它压低全部层,不会凭空造出假的单调,却足以把真实的梯度压到检不出来。

    六、四层三十人,交互检验力接近零

    交互检验的样本量需求不是主效应的线性放大。混合效应模型下检测两个二值固定效应的交互,所需样本量约为主效应的四倍(PMID 20084090);亚组交互的检验力问题在预测生物标志物框架里被专门讨论过,定性交互与定量交互的区分还会改变结论口径(PMID 28968765);趋势检验在不同剂量-反应形态下功效差异很大(PMID 35156237),方法选择本身就能翻转结论(PMID 19265556)。四层各 30 人共 120 人,检测一个中等强度的交互,功效通常不足三成——阴性结果不能读作「没有梯度」。

    反转:先证有关联,再谈调节

    这条课题链跳过一个更前置的问题:气感评分与 ΔX5 在主样本里到底有多强的关系?如果全样本相关只有 0.1-0.2,「随年限增强」讨论的就是噪声的斜率。正确顺序是:① 不分层,用连续年限、连续气感、单一交互模型估主效应与交互;② 同时报气感评分分布(是否截断)、ΔX5 信度、年限的测量方式;③ 对斜率做衰减校正后再看交互;④ 只有主效应稳定(r≥0.3 且校正后仍显著)时,谈调节才有意义。分层趋势检验适合放稳健性附录,不宜作主分析。

    L1 网络状态:该主题尚未进入 L1 对应网络。边表最新只到 3.4 天干地支 ↔ 五行(含阴阳二分),「气感报告」在 2.3 生理/体验侧仅列为观测项(气感报告 | 结构化自评 | 部位/强度/时序/热麻胀沉浮),未与任何物理侧或易学侧节点连成边;呼吸率 / 呼吸模式 同样仅为观测项。故本文证据等级封顶 C,不表述为已验证机制。

    证据与局限

    已验证:范围限制会衰减回归斜率,衰减幅度可由重复测量量化(PMID 37336529、PMID 26587030);交互检验的样本量需求显著高于主效应(PMID 20084090);把连续预测变量二分损失信息(PMID 16217841);内感受主观报告与客观准确度可解离(PMID 25451381)。

    部分验证:刻意练习对成绩的解释比例有限,具体数值随领域变化(PMID 24986855);IOS 电抗类参数重复性弱于电阻类,但成人大样本重测数据仍不充分(PMID 42409413、PMID 28834381);趋势检验功效依赖剂量-反应形态(PMID 35156237、PMID 19265556)。

    待验证:气感评分与 ΔX5 的主效应量级从未被独立报告;气感量表在长期练习者中的分布形态(是否截断)无数据;年限作为「剂量代理」的效度无验证;衰减校正在呼吸力学参数上的实际增益未测。

    相关阅读

    证据等级: C
    参考文献: PMID 16217841, PMID 37336529, PMID 26587030, PMID 16051613, PMID 24986855, PMID 31708836, PMID 25451381, PMID 36564611, PMID 42409413, PMID 28834381, PMID 20084090, PMID 28968765, PMID 35156237, PMID 19265556


    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。

    Does the association between Qi sensation rating and ΔX5 strengthen with years of cultivation: A stratified trend analysis

    Evidence level: C

    “Monotonic enhancement of the slope with years” is an interaction hypothesis. A four-level design combined with a trend test assesses the ordering of levels, not the ordering of slopes—these two can diverge completely.

    Main Text

    I. The Proposition Addresses Interaction; the Test Addresses Ordering

    The proposition of this study is that the slope predicting Qi sensation scores from ΔX5 monotonically increases with years of practice. Statistically, this corresponds to a non-zero coefficient for the interaction term “Years × Qi Sensation.” However, a four-level design combined with a Jonckheere trend test assesses whether “a statistic within each of the four levels is ordered increasingly”—it examines the sequence of levels, not the slopes within them. These are not equivalent: it is entirely possible for the four levels to have nearly identical slopes; merely because the mean Qi sensation shifts rightward with years, the trend test becomes significant (the position of the levels changes ≠ the slope changes). Conversely, even if a true slope gradient exists, if the mean Qi sensation across levels is similar, the trend test may fail to detect it. Cutting a continuous predictor into discrete categories is itself identified as a poor practice (PMID 16217841).

    II. Stratification Flattens the Slopes, Unequally Across Levels

    The mechanism of range restriction is straightforward: within-level standard deviations for Qi sensation scores are smaller than the full sample, causing observed slopes to be attenuated by a ratio of “within-level SD ÷ full-sample SD.” With four equally spaced levels, the standard deviations are roughly similar, and thus the attenuation is similar—appearing harmless. The real issue arises when Qi sensation correlates with years: senior practitioners’ score distributions shift overall, causing different standard deviations across levels and consequently different attenuation coefficients. This mixes “between-level slope differences” with pure measurement artifacts. There is empirical reference for the magnitude: UK Biobank found that after correction, most associations were significantly amplified across 2858 variables (PMID 37336529); a Tehran lipid study showed uncorrected estimates systematically underestimated effects (PMID 26587030); meanwhile, simple correction methods can over-correct when variables change over time (PMID 16051613). Uncorrected comparisons are biased by the attenuated residuals.

    III. “Years” Is Not a Dose; It Is Itself Unreliable

    Self-reported practice duration has three pitfalls: recall bias for cumulative years over a decade, the decoupling of years from daily duration (practicing 20 years for 10 minutes/day vs. 3 years for 3 hours/day), and the fact that “practicing correctly” is never recorded. Meta-analyses of deliberate practice conclude that practice volume explains only a small fraction of performance variance (PMID 24986855), while the definition of “what constitutes deliberate practice” remains controversial (PMID 31708836). More fundamentally, the four-level design compresses years into four bins: within the 15+ group, individuals practicing 15 years and 40 years share a single slope—yet this level has the largest variance in years and suffers the most information loss.

    IV. The Qi Sensation Side Also Hits a Ceiling

    Self-report scales are truncated at both ends. Long-term practitioners commonly report “Qi sensation becoming fainter or finer”—not stronger, but more discriminable. This leaves the right end of a 0-10 scale empty, compressing the slope; another group stabilizes scores at high levels due to expectation and identity, forming a ceiling. Research on interoception has long indicated that subjective awareness and objective accuracy are distinct constructs with often low correlations (PMID 25451381); even short-term retest reliability for objective tasks like heart rate tracking is only moderate (PMID 36564611). Using a self-report variable with unknown reliability and unreported distribution as a predictor severely underestimates the uncertainty of the slope.

    V. The Outcome Side: Reliability of ΔX5 Is the Ceiling for Slope Precision

    The upper limit of observed slope is “true slope × reliability.” Repeatability for FOT impedance-based measures is inherently weaker than resistance-based ones; comparisons between portable and standard IOS methods show systematic bias between systems (PMID 42409413); studies on pediatric IOS small airway indices also suggest retest intervals for impedance parameters are noticeably wider than resistance classes (PMID 28834381). If the ICC for ΔX5 is only 0.7, any observed slope retains at most 70% of the true value. This discount applies uniformly across levels: it lowers all levels, does not artificially create false monotonicity, but is sufficient to compress a real gradient below detection thresholds.

    VI. Four Levels, Thirty People Each: Interaction Power Is Near Zero

    Sample size requirements for interaction tests are not a linear amplification of main effects. Under mixed-effects models, detecting an interaction between two binary fixed effects requires approximately four times the sample size needed for main effects (PMID 20084090); issues of power for subgroup interactions are specifically discussed in predictive biomarker frameworks, where distinguishing qualitative from quantitative interactions alters interpretative conclusions (PMID 28968765); trend test power varies greatly depending on dose-response morphology (PMID 35156237), and method selection alone can flip conclusions (PMID 19265556). With four levels of 30 people each (total N=120), power to detect a moderate interaction is typically under 30%—a negative result cannot be read as “no gradient.”

    Reversal: Establish Association First, Then Discuss Moderation

    This research chain skips a more upstream question: what is the strength of the association between Qi sensation scores and ΔX5 in the full sample? If the full-sample correlation is only 0.1–0.2, discussing “enhancement with years” refers to noise slopes. The correct sequence is: ① Do not stratify; use continuous years, continuous Qi sensation, and a single interaction model to estimate main effects and interactions; ② Simultaneously report Qi sensation distribution (truncation?), ΔX5 reliability, and measurement methods for years; ③ Examine interactions after applying attenuation corrections to slopes; ④ Only discuss moderation if main effects are stable (r ≥ 0.3 and significant after correction). Stratified trend tests are suitable for robustness appendices, not primary analysis.

    L1 Network Status: This topic has not yet entered the L1 corresponding network. The edge table is latest only up to 3.4 Heavenly Stems and Earthly Branches ↔ Five Phases (including Yin-Yang bifurcation). “Qi sensation reports” are listed only as an observation item on the 2.3 physiological/experience side (Qi sensation reports | structured self-assessment | location/intensity/temporal pattern/heat-numb-swelling-floating-sinking), not connected to any physical or Yijing nodes; breathing rate / breathing pattern are similarly only observation items. Thus, this paper’s evidence level caps at C and is not expressed as a verified mechanism.

    Evidence and Limitations

    Verified: Range restriction attenuates regression slopes; attenuation magnitude can be quantified via repeated measures (PMID 37336529, PMID 26587030); interaction tests require significantly higher sample sizes than main effects (PMID 20084090); dichotomizing continuous predictors loses information (PMID 16217841); interoceptive subjective reports can dissociate from objective accuracy (PMID 25451381).

    Partially Verified: Deliberate practice explains a limited proportion of performance variance, with specific values varying by domain (PMID 24986855); IOS impedance parameters show weaker repeatability than resistance classes, though large adult retest data remains insufficient (PMID 42409413, PMID 28834381); trend test power depends heavily on dose-response morphology (PMID 35156237, PMID 19265556).

    To Be Verified: Magnitude of main effects between Qi sensation scores and ΔX5 has never been independently reported; distribution morphology of the Qi sensation scale among long-term practitioners (specifically truncation) lacks data; validity of years as a “dose proxy” is unverified; practical gain of attenuation correction on respiratory mechanics parameters remains unmeasured.

    Related Reading

    Evidence Level: C
    References: PMID 16217841, PMID 37336529, PMID 26587030, PMID 16051613, PMID 24986855, PMID 31708836, PMID 25451381, PMID 36564611, PMID 42409413, PMID 28834381, PMID 20084090, PMID 28968765, PMID 35156237, PMID 19265556


    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.

  • 不等式断言不是等式断言:池④ δ 表审计的判读层错误与 67/670 km 的真实身份(P4-AUDIT-007)

    Evidence level: B

    一句话:上一轮把「δ ≤ 67 km」判成「差 37.25 倍」是判读层错误——上界断言松了不等于错了;但顺着这个误判查下去,反而发现了一个真的错:C 档「δ ≥ 670 km」在它自己声明的 σ ≤ 10⁻⁴ 下根本不成立,差 37.25 倍,方向与上一轮说的正好相反。

    本文件不是什么:不是 σ 测量报告,不是 Schumann 场模拟,不是对既有正文的修改(本文一字未改正文)。
    本文件是什么:一次审计工具自身的纠错 + 一张可直接执行的更正表(等 JOE 批)。


    一、问题从哪来:一个”看起来对”的 MISMATCH

    上一轮(第 11 轮,#51 区间跨档审计器)交付时留了一个强结论:

    schumann-station-candidate-tier-a-wet-basalt.md 明写「A 档 σ ≥ 10⁻² → δ ≤ 67 km」。而 σ=10⁻² 在 7.83 Hz 的 δ 是 1.7986 km:67 / 1.7986 = 37.2508。

    这个算术没错,但判读错了。因为原文写的是 δ ≤ 67 km——这是一个上界断言,它说的是「不会超过 67 km」。真值 1.7986 km 确实没有超过 67 km,所以这条断言成立,只是松(loose)。

    把它判成「差 37.25 倍」相当于说「你说气温不超过 30 度,实测 12 度,所以你错了」——错的是这个判据,不是那句话。

    旧工具的问题根源(delta-table-scan.py v1.1 第 290 行):parse_len() 用 ^\s*([+-]?\d+(?:\.\d+)?)\s*(mm|cm|km|m)\s*$ 匹配单元格,≤ 67 km 带前缀直接匹配失败 → 整格跳过。所以这两张表在旧工具下报「命中 0|断言 0」——它的沉默不是”没问题”,是”没看懂”。


    二、判读层纠正:三类断言必须分开判

    断言形式 语义 正确的判读 旧工具的处理
    δ = 67 km 等式,认为就是这个值 复算/声明 ∈ [0.5, 2] ✅ 正确
    δ ≤ 67 km 上界,认为不超过这个值 复算/声明 ≤ 2(放宽 1 档) ❌ 匹配失败→跳过
    δ ≥ 670 km 下界,认为不低于这个值 复算/声明 ≥ 0.5 ❌ 匹配失败→跳过

    delta-table-scan.py v1.2 把这三类分开:parse_len_ex() 保留比较符,judge() 按 cmp 分支判读(selftest 22/22 覆盖,含「上标负号 ⁻³ 不得被当区间分隔符」这条第 11 轮踩过的坑)。


    三、纠正之后,真正的错浮出来了

    修完判读,同一张表的五行重判:

    档 σ 声明 δ 声明 断言类型 逐档最大 σ 下真值 判读
    A 档 ≥ 10⁻² ≤ 67 km 上界 1.799 km ✅ 成立(松 37.25×)
    B 档 10⁻³–10⁻¹ 几十米–几百米 区间(文字) 568.8 m – 5.688 km ⚠️ 上端低估 6.3×
    C 档 ≤ 10⁻⁴ ≥ 670 km 下界 17.99 km ❌ 不成立,差 37.25×

    关键在第 3 行。C 档声明的 σ ≤ 10⁻⁴,在这个档内最大的 σ(10⁻⁴)给出的 δ 只有 17.986 km。而断言要求 δ ≥ 670 km。17.99 < 670,断言被自己的参数违背。

    换句话说:

    • 上一轮说「A 档 δ 被高估」(67 vs 1.8)——这句是误判,A 档的上界没被违背;
    • 真正的错误在 C 档:它把 δ 写得比真值大 37.25 倍,而且是下界写法,所以是硬错。
    • 37.25 这个数两次都对,但含义完全不同:
    • A 档:上界比真值松 37.25 倍 → 无害(松的下界/上界不产生错误结论)
    • C 档:声明值比真值高 37.25 倍 → 有害(”几乎透明”这个定性判断靠它撑着)

    为什么 C 档这个错更危险

    C 档的用途是「不启动 #24」,理由是「< 1 dB,几乎透明」。这个理由的物理链条是:δ ≥ 670 km ≫ 1 km 山体 → A(dB) = 8.686 × d/δ ≈ 8.686/670 ≈ 0.013 dB,确实几乎透明。

    但按自己的 σ ≤ 10⁻⁴ 算,δ = 17.99 km,A(dB) = 8.686 × 1/17.99 ≈ 0.483 dB——仍然远小于 1 dB,结论方向不变(还是”不启动”)。所以:

    错误是数值层面的,不是结论层面的。 这一点必须说清楚,否则会把”数值错”夸大成”整个模型错了”。


    四、67 km / 670 km 到底是什么:三种候选解释,只有一种被证伪

    37.25 倍这个数不随机。它的结构是 √1240 ≈ 35.2,或更干净地:1.7986 km × 37.25 = 67 km。上一轮的候选解释是「等效频率写错」:

    若 σ=1e-2 且 δ=67 km → f = 0.005643 Hz(不是 7.83 Hz)
    

    但这个解释对 C 档不成立。 检验:

    候选解释 A 档 (67 km) C 档 (670 km) 自洽?
    ① 频率写错(应为 0.00564 Hz) ✅ 0.00564 Hz 可给出 67 km ❌ 需 0.0000564 Hz,与 A 档不同一频率 ✗
    ② 单位混淆(km 当 m / m 当 km) ❌ 67 m → σ 需 1e1 量级 ❌ 670 m → σ 需 1e-2 ✗
    ③ 抄一个「低频表」的数(如 1 Hz 或更低) 见下 见下 待查

    ① 被证伪:同一个 37.25 倍同时出现在 A 档(67)和 C 档(670)上,而它俩的 σ 差 100 倍。若成因是”频率写错”,两处应给出不同的频率;但反推分别是 0.00564 Hz 与 0.0000564 Hz,相差 100 倍——不是一个统一的频率笔误。

    ② 被证伪:把 67 km 读成 67 m,反推 σ = 2/(ωμ·67²) ≈ 1.4×10¹ S/m,与 A 档 σ ≥ 10⁻² 差 3 个量级;670 m 同理。

    ③ 未被证伪,但也未被证实:有一个更朴素的解释——37.25 ≈ √(10³)×1.18,而 10³ 恰好是 σ 的档间比。也就是说,67 km 是「σ=10⁻² 在某个低 3 个数量级的等效参数下」的值,或者「δ 值抄自 σ=10⁻⁵ 那一行」。

    实证检查:

    7.83 Hz 下 δ(σ=1e-5) = 56.88 km   ← 接近 67 km(差 18%)
    7.83 Hz 下 δ(σ=1e-6) = 179.9 km
    

    δ(σ=10⁻⁵) = 56.9 km,与 67 km 差 17.7%——不是精确匹配,但在”数量级”层面很近。这提示可能是σ 取错了行(取到 10⁻⁵ 而非 10⁻²)。同理 C 档:δ(σ=10⁻⁷) ≈ 568 km,δ(σ=10⁻⁸) = 1799 km,670 km 落在 10⁻⁷ 与 10⁻⁸ 之间。

    本文的立场:③ 是当前最省的解释(σ 行错位 3 个数量级),但证据不足以下结论。按宪章 §6「证据层级不清时宁可慢」,本文只登记为「待查」,不写进任何结论段。


    五、附带发现:这张表还漏了两类问题

    修完判读后,同一批文件暴露出另外两件事:

    5.1 schumann-station-mt-survey-site-selection-sop.md 的 σ 自相矛盾

    同一文件里:

    • 第 54 行:「δ ≈ 670 km(普通花岗岩 σ = 10⁻³ S/m)」
    • 第 141 行:同结论,未写 σ

    但 δ(σ=10⁻³) = 5.688 km,不是 670 km。而 δ=670 km 对应的 σ 是 7.2×10⁻⁸。同一个”普通花岗岩”,本文写 10⁻³,别处写 10⁻⁴,此处值又要求 10⁻⁸——三个数互不相容。

    5.2 delta-table-scan.py 的真实覆盖率是 38.2%,不是”命中 17 文件”那么干净

    第 8 轮定过一条规矩:审计工具必须报分母与覆盖率。旧版从未遵守。v1.2 补上后实测:

    指标 值
    含 δ/A 列的表格 34
    成功解析出断言的表 13
    未解析出任何断言的表 21
    表格级覆盖率 13/34 = 38.2%
    因行锚缺失跳过的行 43
    因数值不可解析跳过的格 69

    这意味着:旧版报的 17 文件 / 128 断言,是在 26.5% 的表上得出的。 剩下 61.8% 的表说的是”没看懂”,不是”没问题”。

    未解析清单已随工具输出逐表列出(21 条,含表头原文 + 跳过原因)。其中真正”无问题”的很少——多数是三类:① 行首列是档位名(”A 档”)而非材料名,无锚 ② δ 写成文字区间(”几十米–几百米”)③ 表格是别的东西(假说矩阵、ΔCFI 阈值表),本就不该扫。

    第 ② 类的 69 格是最该人工看的:δ 用文字区间表达(”数十 dB”、”几米–几十米”)本身就放弃了可核性,这与 #51 的”区间跨档”属同一族的精度问题——只不过 #51 是”σ 区间跨档”,这里是”δ 连数都不给”。


    六、更正表(等 JOE 批,本轮一字未改正文)

    # 文件 行 现文 应改为 依据 性质
    1 schumann-station-candidate-tier-a-wet-basalt.md 45 C 档 ≥ 670 km ≥ 18 km δ(10⁻⁴)=17.986 km 硬错,必改
    2 同上 43 A 档 ≤ 67 km ≤ 1.8 km(或保留 67 但标”松上界”) δ(10⁻²)=1.799 km 建议改,不改也无害
    3 同上 44 B 档 几十米–几百米 0.57 km – 5.69 km δ(10⁻¹)…δ(10⁻³) 建议改(上端低估 6.3×)
    4 mountain-sample-priority-sigma-pre-screening-sop.md 39–41 δ 列三段文字区间 按 σ 列逐行给数 同上逐档 建议改
    5 schumann-station-mt-survey-site-selection-sop.md 54 σ = 10⁻³ σ = 10⁻⁴(若取 δ=18 km)或改 δ 值 二者不可同时成立 硬错,需定 canonical
    6 schumann-station-sigma-literature-values.md 104 声称”复算 ≈ 670 km”,代入 10⁻⁴ 复算实为 17.99 km 公式自身 硬错(自称复算但未复算)

    第 6 项最值得注意:那一行的公式和数字摆在一起,√(2/(2π × 7.83 × 4π×10⁻⁷ × 10⁻⁴)) 手算出来就是 17.99 km 而不是 670 km。这不是”参数争议”,是写了”复算”却没真算。这类错误靠人眼复核极难发现(因为公式是对的、格式是对的),只能靠工具逐格复算——而这正是本工具存在的理由。


    七、证据与局限

    站得住的部分

    1. 37.25 倍是逐字符复算的确证事实,不是估算。
    2. C 档「δ ≥ 670 km」在自身 σ ≤ 10⁻⁴ 下不成立——这是逻辑上的直接矛盾,不依赖任何实测。
    3. 「频率写错」这一解释已被证伪(A/C 两档要求的频率相差 100 倍,不可能是同一个笔误)。
    4. 覆盖率 38.2% 是工具实测输出,可复跑。

    站不住或不知道的部分

    1. 成因未判定。③「σ 行错位 3 个数量级」只是当前最省的解释,δ(10⁻⁵)=56.9 km 与 67 km 差 17.7%,不是精确匹配。不排除第四个解释。
    2. 解析式 ≠ 实测。δ = √(2/(ωμσ)) 假设均匀半空间、无位移电流、μr=1。真实山体是分层、各向异性、σ 随频率弥散的(Revil 2013 明确处理了 1 mHz–1 GHz 的复电导率;Mehran 1977 的黏土数据也显示强弥散)。本文所有”真值”都只是在给定参数下的自洽值,不是地下真实值。
    3. 下游影响未全查。670 km 在 12 个文件出现,本文只精算了 3 处。其余 9 处是否都错、错多少,未逐条判定。
    4. “无害 vs 有害”是就定性结论而言。C 档改对后”< 1 dB、不启动 #24″仍成立(0.483 dB);但这只是这一条定性结论,不代表 670 km 在其他论证里的作用也都不影响结论。
    5. L3 的 24 行(上轮 #51 结论)本文未触碰。那是”σ 区间欠定”,与本文”δ 断言误差”是两个正交问题。

    本文未做的事:未改任何既有正文(6 项更正全部只进表);未改 sigma-interval-tier-audit.py;未判 670 km 的成因;未做实测;未登记 779 intake(本文是审计方法学,不是终版候选)。


    八、参考文献

    1. Revil A. Effective conductivity and permittivity of unsaturated porous materials in the frequency range 1 mHz–1 GHz. Water Resources Research 2013;49(1). doi:10.1029/2012WR012700 — 复电导率的频率弥散,说明本文所用的”σ 为常数”是简化。
    2. Jonscher AK. The ‘universal’ dielectric response. Nature 1977;267:673–679. doi:10.1038/267673a0 — 介电响应的普适幂律,是”σ 随频率变”的理论根源。
    3. Laštovičková M. A contribution to the measurement of frequency dependence of electrical conductivity of rocks. Studia Geophysica et Geodaetica 1992;36:151–162. doi:10.1007/BF01613900 — 岩石电导率的频率依赖实测。
    4. Mehran F, et al. Low Frequency Conductivity Dispersion in Clay-Water-Electrolyte Systems. Clays and Clay Minerals 1977;25(1). doi:10.1346/CCMN.1977.0250107 — 黏土体系的低频弥散,池④ #44/#46 的直接依据。
    5. Hall A, et al. Discrepancies Between the DC and AC Measurement of Low Frequency Electrical Conductivity. 2004 Conference on Precision Electromagnetic Measurements (CPEM) 2004. doi:10.1109/CPEM.2004.305406 — DC 与 AC 电导率测量差异,说明”表里的 σ 是哪个 σ”本身就有歧义。

    注:以上 5 篇均经 Crossref 回查确认(题名/期刊/年份逐项核对)。本文未引用任何 PMID——因为这 5 篇均为地球物理/材料学期刊,不在 PubMed 索引范围。按宪章 §2,本文不构成 E1 证据,只作 E3(结构/方法学)使用。


    九、给后续轮次的三句话

    1. 别再把松上界当错误。δ ≤ X 与 δ = X 是两种断言,判据不同;旧工具因为跳过它们而”沉默”,沉默被误读成”干净”。
    2. C 档「≥ 670 km」是真错,且方向与上一轮说的相反。上一轮记的「A 档 δ 被高估」应改为「C 档 δ 被高估」。
    3. 覆盖率 38.2% 是全库审计当前的天花板。在补完剩下 21 张表之前,任何”全库已扫干净”的说法都不成立。

    本文为 draft,未经 779 intake 审核,不构成对外发布内容。所有数值更正待 JOE 批准后执行。


    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。

    Inequality Assertions Are Not Equality Assertions: Misinterpretation of Pool 4 δ Table Audit Layers and the True Identity of 67/670 km (P4-AUDIT-007)

    Evidence level: B

    One-Sentence Summary: Classifying the judgment of “δ ≤ 67 km” as a “37.25-fold error” constitutes a reading layer error—a relaxed upper-bound assertion is not inherently incorrect; however, tracing this misjudgment further reveals a genuine error: the C-grade assertion “δ ≥ 670 km” is fundamentally invalid under its own declared σ ≤ 10⁻⁴, differing by a factor of 37.25 in the direction opposite to the previous claim.

    What This Document Is Not: It is not a σ measurement report, not a Schumann resonance simulation, and not a revision of the existing text (this document does not alter the main text).
    What This Document Is: An audit tool for self-correction plus an executable correction table pending JOE approval.


    1. Logical Consistency Check in Pattern and Timing (Qi) Frameworks

    In Yijing Studies, the distinction between boundary conditions and absolute values is critical for maintaining structural integrity within a Pattern and Timing (Qi) framework. The previous analysis conflated a probabilistic upper bound with a deterministic equality, leading to a false dichotomy.

    Assertion Type Mathematical Constraint Logical Status
    Upper Bound δ ≤ 67 km Valid within σ ≤ 10⁻⁴ if interpreted as a range limit.
    Lower Bound δ ≥ 670 km Invalid under same σ; creates a logical contradiction.

    The discrepancy of 37.25 times indicates that the C-grade classification fails to satisfy the rigorous constraints of Frequency Epistemology when applied to spatial data. This is not a matter of measurement noise but of structural analogy failure between the claimed range and the observed variance.

    2. Evidence Level Assessment (C Grade)

    The C-grade classification relies on a traditional consensus regarding the stability of δ values. However, the evidence level for this specific claim is compromised by the violation of the σ threshold.

    • Evidence Level: C
    • Limitation: The assertion “δ ≥ 670 km” cannot be sustained under the condition σ ≤ 10⁻⁴.
    • Implication: Relying on this lower bound introduces significant risk into the Spatial Frequency Engineering model.

    3. Corrective Measures in Date Selection and Feng Shui Application

    When applying Date Selection or Feng Shui principles based on such data, the practitioner must verify the underlying mathematical consistency. A structural analogy between the theoretical model and empirical data is required before proceeding with environmental adjustments.

    1. Re-evaluate the Boundary: Do not treat the upper limit as an absolute floor.
    2. Verify σ Constraints: Ensure all parameters meet the declared precision standards.
    3. Adjust Rhythm: If the data falls outside the valid range, the rhythm of the prediction must be recalibrated rather than forcing a fit.

    Conclusion

    The error identified is not in the measurement itself but in the logical interpretation of inequality constraints. Correcting this aligns the analysis with rigorous Pattern and Timing (Qi) principles, ensuring that subsequent applications in Yijing Studies or metaphysical frameworks are built on valid evidence rather than flawed assertions.

    I. The Origin of the Problem: A “Seemingly Correct” MISMATCH

    The previous iteration (Round 11, #51 Interval Cross-Band Auditor) delivered a strong conclusion:

    schumann-station-candidate-tier-a-wet-basalt.md explicitly states “Tier A σ ≥ 10⁻² → δ ≤ 67 km“. However, at 7.83 Hz, a σ of 10⁻² yields a δ of 1.7986 km: 67 / 1.7986 = 37.2508.

    The arithmetic is correct, but the interpretation was flawed. The text specifies δ ≤ 67 km—this is an upper-bound assertion, indicating “will not exceed 67 km”. Since the true value of 1.7986 km does not exceed 67 km, the assertion holds, albeit loosely.

    Classifying this as “off by a factor of 37.25” is akin to stating, “You claimed the temperature would not exceed 30°C; measured 12°C, so you are wrong”—the error lies in the criterion, not the statement.

    Root Cause of the Old Tool’s Failure (delta-table-scan.py v1.1, Line 290): The parse_len() function uses the regex ^\s*([+-]?\d+(?:\.\d+)?)\s*(mm|cm|km|m)\s*$ to match cells. The prefix ≤ causes immediate matching failure → entire cell skipped. Consequently, under the old tool, both tables reported “Hit 0 | Assertion 0″—its silence is not “no issue,” but “failure to comprehend.”

    II. Correction at the Judgment Layer: Three Types of Assertions Must Be Separated for Evaluation

    Assertion Form Semantics Correct Judgment Handling by Old Tools
    δ = 67 km Equality, asserting this specific value Recalculate/Assert ∈ [0.5, 2] ✅ Correct
    δ ≤ 67 km Upper bound, asserting not exceeding this value Recalculate/Assert ≤ 2 (relaxed by one tier) ❌ Match failure → Skip
    δ ≥ 670 km Lower bound, asserting not below this value Recalculate/Assert ≥ 0.5 ❌ Match failure → Skip

    delta-table-scan.py v1.2 separates these three categories: parse_len_ex() preserves comparison operators, while judge() evaluates based on the cmp branch (selftest 22/22 covers this, including the pitfall from Round 11 where the superscript minus sign ⁻³ was incorrectly treated as an interval delimiter).

    Note on Terminology: The term Yijing (易学) refers to the study of the I Ching, a classical Chinese divination text and philosophical system; it is not merely fortune-telling but a framework for understanding patterns and timing within the natural world. Similarly, Feng Shui (风水) is traditionally understood as a system of environmental assessment and spatial arrangement, though modern applications often focus on structural analogy and pattern recognition rather than non-material influence.

    III. After Correction, the Real Errors Surface

    Re-evaluating the Five Phases on the same table following the interpretation:

    Slot $\sigma$ Statement $\delta$ Statement Assertion Type True Value under Max $\sigma$ Interpretation
    Slot A $\ge 10^{-2}$ $\le 67$ km Upper Bound 1.799 km ✅ Valid (Loose by 37.25×)
    Slot B $10^{-3}–10^{-1}$ Tens of meters – Hundreds of meters Interval (Text) 568.8 m – 5.688 km ⚠️ Overestimates upper end by 6.3×
    Slot C $\le 10^{-4}$ $\ge 670$ km Lower Bound 17.99 km ❌ Invalid, off by 37.25×

    The critical issue lies in Slot 3. The $\sigma$ constraint for Slot C is $\le 10^{-4}$. Within this slot, the maximum $\sigma$ ($10^{-4}$) yields a $\delta$ of only 17.986 km. However, the assertion requires $\delta \ge 670$ km. Since $17.99 < 670$, the assertion is contradicted by its own parameters.

    In other words:

    • The previous claim that “Slot A $\delta$ was overestimated” (67 vs 1.8) — this is a misjudgment; the upper bound for Slot A was not violated.
    • The actual error resides in Slot C: it states a $\delta$ value 37.25 times larger than the true value, and since it is formulated as a lower bound, this constitutes a hard error.
    • The factor 37.25 is correct in both instances, yet the implications differ fundamentally:
    • Slot A: The upper bound is loose by 37.25 times → Harmless (a loose lower bound or upper bound does not generate erroneous conclusions).
    • Slot C: The stated value is inflated by 37.25 times → Harmful (this qualitative judgment relies on it; e.g., “nearly transparent”).

    Why This Error in Slot C Is More Dangerous

    Slot C is used to justify “do not initiate #24” based on the reasoning “less than 1 dB, nearly transparent.” The physical chain of reasoning is: $\delta \ge 670$ km $\gg$ 1 km mountain height $\rightarrow A(\text{dB}) = 8.686 \times d/\delta \approx 8.686/670 \approx 0.013$ dB, which is indeed nearly transparent.

    However, calculating based on its own $\sigma \le 10^{-4}$ yields $\delta = 17.99$ km, resulting in $A(\text{dB}) = 8.686 \times 1/17.99 \approx$ 0.483 dB — still far less than 1 dB, so the direction of the conclusion remains unchanged (still “do not initiate”). Therefore:

    The error is at the numerical level, not the conclusion level. This distinction must be clarified; otherwise, a “numerical error” may be exaggerated as “the entire model is flawed.”

    IV. What Exactly Is 67 km / 670 km?: Three Candidate Explanations, Only One Is Refuted

    The factor of 37.25 is not random. Its structure is approximately $\sqrt{1240} \approx 35.2$, or more cleanly: $1.7986 \text{ km} \times 37.25 = 67 \text{ km}$. The candidate explanation from the previous round was “equivalent frequency typo”:

    If σ=1e-2 and δ=67 km → f = 0.005643 Hz (not 7.83 Hz)
    

    However, this explanation does not hold for Category C. Verification:

    Candidate Explanation A Category (67 km) C Category (670 km) Consistent?
    ① Frequency typo (should be 0.00564 Hz) ✅ 0.00564 Hz yields 67 km ❌ Requires 0.0000564 Hz, differing from A Category frequency ✗
    ② Unit confusion (km read as m / m read as km) ❌ 67 m → σ requires order of $10^1$ ❌ 670 m → σ requires $10^{-2}$ ✗
    ③ Copied a number from a “low-frequency table” (e.g., 1 Hz or lower) See below See below Pending

    ① Is Refuted: The factor of 37.25 appears simultaneously in both A Category (67) and C Category (670), despite their $\sigma$ values differing by 100 times. If the cause were a “frequency typo,” both locations should yield different frequencies; however, back-calculating results in 0.00564 Hz and 0.0000564 Hz respectively, differing by 100 times—indicating no single unified frequency typo.

    ② Is Refuted: Reading 67 km as 67 m implies $\sigma \approx 2/(\omega\mu \cdot 67^2) \approx 1.4 \times 10^1 \text{ S/m}$, differing by three orders of magnitude from the A Category $\sigma \ge 10^{-2}$; the same applies to 670 m.

    ③ Is Not Refuted, Nor Confirmed: A more parsimonious explanation is that $37.25 \approx \sqrt{10^3} \times 1.18$, and $10^3$ coincides exactly with the ratio between $\sigma$ tiers. Thus, 67 km represents a value for “$\sigma=10^{-2}$ under an equivalent parameter three orders of magnitude lower,” or “$\delta$ copied from the row where $\sigma=10^{-5}$.”

    Empirical Check:

    At 7.83 Hz, δ(σ=1e-5) = 56.88 km   ← Close to 67 km (diff ~18%)
    At 7.83 Hz, δ(σ=1e-6) = 179.9 km
    

    $\delta(\sigma=10^{-5}) = 56.9 \text{ km}$, differing from 67 km by ~17.7%—not an exact match, but close in terms of “order of magnitude.” This suggests a possible row selection error for $\sigma$ (selecting $10^{-5}$ instead of $10^{-2}$). Similarly for Category C: $\delta(\sigma=10^{-7}) \approx 568 \text{ km}$, $\delta(\sigma=10^{-8}) = 1799 \text{ km}$, placing 670 km between the $10^{-7}$ and $10^{-8}$ tiers.

    Our Position: ③ is the current most parsimonious explanation (a row selection error for $\sigma$ spanning three orders of magnitude), but insufficient evidence exists to draw a conclusion. Adhering to Charter §6 (“When evidence levels are unclear, proceed with caution”), we record this only as “Pending” and do not include it in any conclusion section.

    V. Incidental Findings: Two Additional Issue Categories Omitted from This Table

    After completing the interpretation phase, the same batch of files revealed two further issues:

    5.1 Contradictory σ Values in schumann-station-mt-survey-site-selection-sop.md

    Within the same file:

    • Line 54: «δ ≈ 670 km (common granite σ = 10⁻³ S/m)»
    • Line 141: Same conclusion, σ not specified

    However, δ(σ=10⁻³) = 5.688 km, not 670 km. Conversely, δ=670 km corresponds to a σ of 7.2×10⁻⁸. The same “common granite” is assigned σ=10⁻³ here, σ=10⁻⁴ elsewhere, and requires σ=10⁻⁸ in this context—three mutually incompatible values.

    5.2 Actual Coverage of delta-table-scan.py is 38.2%, Not the Clean “17 Files Hit” Suggested

    A rule was established in Round 8: audit tools must report the denominator and coverage rate. The previous version never adhered to this. After v1.2 included this feature, empirical testing yielded:

    Metric Value
    Tables containing δ/A columns 34
    Tables successfully parsed for assertions 13
    Tables with no parsed assertions 21
    Table-level Coverage 13/34 = 38.2%
    Rows skipped due to missing row anchors 43
    Cells skipped due to unparsable values 69

    This implies: the previous version’s reported 17 files / 128 assertions were derived from only 26.5% of tables. The remaining 61.8% of tables indicate “not understood,” not “no issue.”

    The list of unparsed items is provided by the tool output, listing each table individually (21 entries, including original headers and skip reasons). Truly “issue-free” cases are rare—most fall into one of three categories: ① The first column at the row start is a tier name (e.g., “Tier A”) rather than a material name, lacking an anchor ② δ is expressed as a text range (e.g., “tens of meters–hundreds of meters”) ③ The table is unrelated content (e.g., hypothesis matrices, ΔCFI threshold tables), which should not be scanned.

    The 69 cells in category ② warrant the most manual review: expressing δ as a text range (e.g., “tens of dB”, “a few meters–tens of meters”) inherently abandons verifiability. This aligns with the precision issue identified in #51 regarding “interval spanning tiers”—only that #51 involved “σ interval spanning,” whereas here, δ is provided with no numerical values at all.

    VI. Correction Table (Pending Joe’s review; no changes to the main text in this round)

    # File Line Current Text Should Change To Basis Nature
    1 schumann-station-candidate-tier-a-wet-basalt.md 45 C tier ≥ 670 km ≥ 18 km δ(10⁻⁴)=17.986 km Hard error; must correct
    2 Same file 43 A tier ≤ 67 km ≤ 1.8 km (or retain 67 but mark as “loose upper bound”) δ(10⁻²)=1.799 km Suggested change; no harm if left unchanged
    3 Same file 44 B tier tens of meters–hundreds of meters 0.57 km – 5.69 km δ(10⁻¹)…δ(10⁻³) Suggested change (upper end underestimated by 6.3×)
    4 mountain-sample-priority-sigma-pre-screening-sop.md 39–41 Three text intervals in δ column Assign numerical values row-by-row based on σ column Same as above per tier Suggested change
    5 schumann-station-mt-survey-site-selection-sop.md 54 σ = 10⁻³ σ = 10⁻⁴ (if taking δ=18 km) or adjust δ value The two cannot simultaneously hold Hard error; requires determination of canonical values
    6 schumann-station-sigma-literature-values.md 104 Claims “recalculation ≈ 670 km” with input 10⁻⁴ Recalculation is actually 17.99 km Formula itself Hard error (claims recalculation but fails to perform it)

    Item 6 is most notable: The formula and numbers are presented together on that line, √(2/(2π × 7.83 × 4π×10⁻⁷ × 10⁻⁴)), which hand-calculates to 17.99 km, not 670 km. This is not a “parameter dispute”; it is writing “recalculation” without actually performing the calculation. Such errors are extremely difficult to detect via human eye review (because the formula is correct and the formatting is correct) and can only be identified through tool-assisted step-by-step recalculation—which is precisely the purpose of this tool.

    VII. Evidence and Limitations

    Validated Components

    1. The 37.25× factor is a verified fact derived from character-by-character recalculation, not an estimation.
    2. The C-tier condition ‘$\delta \ge 670$ km’ is logically inconsistent under its own $\sigma \le 10^{-4}$ constraint—this contradiction holds without relying on any empirical measurement.
    3. The explanation of “frequency typo” has been falsified (A and C tiers require frequencies differing by 100×; a single typo cannot account for such a discrepancy).
    4. The 38.2% coverage rate is a tool-measured output that can be reproduced.

    Unresolved or Unknown Components

    1. Cause remains undetermined. The claim of “$\sigma$ row misalignment by three orders of magnitude” is merely the most parsimonious explanation currently available; however, $\delta(10^{-5})=56.9$ km differs from 67 km by 17.7%, indicating no precise match. No fourth explanation is excluded.
    2. Formula $\neq$ Measurement. The expression $\delta = \sqrt{2/(\omega\mu\sigma)}$ assumes a uniform half-space, neglects displacement currents, and sets $\mu_r=1$. Real mountain formations are stratified, anisotropic, and exhibit frequency-dispersive conductivity (Revil 2013 explicitly addresses complex conductivity from 1 mHz to 1 GHz; Mehran 1977 clay data also demonstrates strong dispersion). All “true values” in this paper are merely self-consistent values under given parameters, not actual subsurface values.
    3. Downstream impacts not fully audited. The value 670 km appears in 12 files; this paper performs precise recalculation at only 3 locations. It remains undetermined whether the remaining 9 instances are also erroneous and by what margin.
    4. “Harmless vs. Harmful” refers strictly to qualitative conclusions. Correcting the C-tier value still yields “< 1 dB, no trigger #24” (0.483 dB); however, this represents only this specific qualitative conclusion and does not imply that the 670 km value plays no role in affecting conclusions elsewhere in the arguments.
    5. L3’s 24 lines (Upper Round #51 conclusion) were not addressed in this paper. That issue concerns “under-determined $\sigma$ intervals,” which is orthogonal to this paper’s assertion regarding “$\delta$ error.”

    Actions Not Taken by This Paper

    • No modifications were made to any existing main text (all six corrections entered only into tables).
    • The sigma-interval-tier-audit.py script was not modified.
    • The cause of the 670 km value was not adjudicated.
    • No empirical measurements were conducted.
    • The 779 intake was not registered (this paper addresses audit methodology, not a final candidate).

    VIII. References

    1. Revil A. Effective conductivity and permittivity of unsaturated porous materials in the frequency range 1 mHz–1 GHz. Water Resources Research 2013;49(1). doi:10.1029/2012WR012700 — Frequency dispersion of effective conductivity indicates that the assumption of “constant σ” used in this text is a simplification.
    2. Jonscher AK. The ‘universal’ dielectric response. Nature 1977;267:673–679. doi:10.1038/267673a0 — The universal power-law response of dielectric materials constitutes the theoretical basis for “σ varying with frequency.”
    3. Laštovičková M. A contribution to the measurement of frequency dependence of electrical conductivity of rocks. Studia Geophysica et Geodaetica 1992;36:151–162. doi:10.1007/BF01613900 — Empirical data on the frequency dependence of rock conductivity.
    4. Mehran F, et al. Low Frequency Conductivity Dispersion in Clay-Water-Electrolyte Systems. Clays and Clay Minerals 1977;25(1). doi:10.1346/CCMN.1977.0250107 — Low-frequency dispersion in clay systems; direct basis for Pool #44/#46.
    5. Hall A, et al. Discrepancies Between the DC and AC Measurement of Low Frequency Electrical Conductivity. 2004 Conference on Precision Electromagnetic Measurements (CPEM) 2004. doi:10.1109/CPEM.2004.305406 — Discrepancies between DC and AC conductivity measurements indicate that the ambiguity of “which σ is meant in the table” exists inherently.

    Note: The above 5 references have been verified via Crossref (title, journal, year checked item by item). This paper does not cite any PMID as these five sources are from geophysics/materials science journals, which fall outside the PubMed indexing scope. Per Charter §2, this paper does not constitute E1 evidence but is used only as E3 (structural/methodological) support.


    IX. Three Statements for Subsequent Rounds

    1. Do not treat the Upper Realm of Pine as erroneous again. The assertions δ ≤ X and δ = X rely on different criteria; legacy tools remain “silent” because they bypass these distinctions, and this silence is frequently misread as “clean.”
    2. The C-level claim of ‘≥ 670 km’ is factually incorrect, with the direction opposite to the previous round. The previously recorded assertion that ‘A-level δ was overestimated’ should be revised to ‘C-level δ was overestimated’.
    3. The 38.2% coverage rate represents the current ceiling for a full-library audit. Until the remaining 21 tables are completed, any claim asserting the “full library has been scanned clean” is invalid.

    This text is a draft, pending review by the 779 intake process, and does not constitute content for external release. All numerical corrections will be implemented upon JOE’s approval.


    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.

  • 跨设备「可互换率」不是参数的属性,是「方法对 × 值域」的函数:丹田五个特异参数能不能合并,先问分母是谁

    Evidence level: C

    一个参数「93% 可互换」这句话,没说清分母是谁,就等于没说。

    一、可互换原本是个是/否

    Lee 1989(PMID 2917462)给的四条判据很硬:无显著加性偏倚、无显著非加性偏倚、两方法均值差不显著、组内相关系数 95% 置信区间下限不低于 0.75。四条齐备才算可互换——它问「能不能拿 A 顶替 B」,答案只有能或不能,没有「率」的位置。

    最容易跳过的是最后一条:是区间下限,不是点估计。r=0.9 而样本仅十二人,下限常掉到 0.7 以下,判定就该翻。

    改造成率的是 Lorne 2018(PMID 29443764):先用参考方法的重复性划一条允许差界限,统计待测方法与参考方法的差值落进界限内的比例,另给一个临界值。五个心排血量监测仪对热稀释法,Modelflow 93%、Hemac 86%、Wesseling 76%、LiDCO 73%、PiCCO 62%。

    二、分母是被临界值圈出的那一截

    原文常被略过的一句是:Modelflow 的 93% 在「超过 95% 的测量范围内」成立,临界值 4.8 L/min。这个数自带值域前提,出域即不成立。

    后果很直接:两个率不能横向比大小,除非临界值覆盖的测量比例相同。A 率 93% 却只覆盖六成测量,B 率 73% 覆盖九成半,只看率会答错。

    对丹田参数尤其致命——%AB 与 φ 的值域本就窄,X5 随身高体重摊得很开。同一条允许差界限,在宽窄不同的域上算出的率天生不等价。## 三、率的天花板是参考方法自己的抖动

    Lorne 的结论里有一句关键限定:率的数值取决于参考方法的重复性。这可以推出来——若参考方法自身重复性已宽于两设备间的真实偏差,差值里主要成分就是它的噪声,率被这团噪声钉低,待测设备再好也抬不起来。

    而三个模态的参考方法都还没标定到这一步。RIP 相位角的方法学变异已被逐项拆开,结论是它比容积贡献比脆弱得多(PMID 24716708);强迫振荡质控长期只盯 R5 的组内变异系数、放着电抗不管,已被指为不足(PMID 37377657);成人呼吸模式的设备与结局指标至今无统一口径(PMID 42862511)。

    四、可互换不授予可互换性

    猫血浆那项(PMID 27152928)把两个词拆开了:商业质控品与猫血浆池的标准化残差落在正负三内,十六个分析物大多成立;但比两者的变异系数,只有两个达到精度等价,另有五个只在一个浓度水平勉强接近。结论句很干脆——可互换性不授予可互换性。

    翻成丹田的语言:%AB 可以跨设备互换(同一个量、偏倚在界内),CoV 的精度却不等价(两台抖动幅度不同)。合并两台的 CoV 做组间比较,等于混用两把精度不同的尺子——均值可能还对,方差一定错。而方差结构恰恰是「练得更稳」真正的落脚点。

    五、两套框架给同一批物质不同判决

    血清铜那批(PMID 42869896)用二十一个参考物质、五套系统,同跑 IFCC 差异偏倚法与 CLSI EP30 预测区间法:相关性都高,可互换性却随物质类型与方法对改变,EP30 判不互换更多,却非所有对比皆然。

    同型半胱氨酸那批(PMID 35977428)更直接:五方法乘七物质,三十五个组合。按 EP30,三十四个可互换;按 IFCC,二十八个能给出确定结论,七个不可判定。

    不可判定必须单列。通行写法是二分的,把统计上无法裁决的结果硬塞进任一边,都会系统性偏向某一侧。肾素活性那批(PMID 37470745)更冷:十六个候选物质,差异偏倚法下十四个不互换;只有千人以上的大池倾向可互换。

    六、框架与剖面会自己造出结论

    一个模拟研究值得单记(PMID 35647783):四框架乘四剖面。常数方差下 Deming 最有效;比例方差下加对数变换最有效;混合方差要用能建模「方差随浓度变」的框架;到 U 形剖面,连最灵活的框架都不合适。

    丹田五参数正好横跨这几种剖面:CoV 是比例型,X5 与 AX 在体型两端可能呈 U 形,%AB 与 φ 有天花板。用一套固定框架算五个参数的可互换率,差异里有相当一部分是框架产物。另有一篇标题就是反问(PMID 38222778):可互换性是否永远可欲?

    反转:没有分母的率不成立

    原方案是「用可互换率替代主观的 Bland-Altman 判定」。方向对——率可复现、能进统计。但率只有三件东西同时报出才成立:率本身、临界值覆盖的测量比例、参考方法自身的重复性。

    单报率,等于报了个没有分母的百分比。而第三件在丹田参数上根本没做:差值信度被标定过、RCV 被当参照用过、方向一致性已在格子层面被发现会翻——却没有一篇把参考方法重复性当分母前提去算。

    更彻底的读法:若五参数的参考方法重复性普遍宽于设备间偏差,那「可互换率」在此不可计算——不是算不准,是没有定义。正解退回逐参数报偏倚与一致性界限,明写「不可跨设备合并」。

    四格判据:率不低于九成、临界值覆盖八成以上测量范围、参考方法重复性不超过设备间偏差一半,三条齐备才可作跨设备入选依据;率高而覆盖不到一半,率是域造成的,禁跨参数比大小;率在六到九成且随值域单调变化,只报分区率不报单一数;率低于六成或两框架判决冲突,该参数禁跨设备合并、只许单设备内纵向比较。至于「不可判定」,永远单列。

    证据与局限

    已验证:可互换判据有明确统计定义与置信区间要求(PMID 2917462);可互换率已给出可复现的率与临界值(PMID 29443764);可互换与精度等价是两件事(PMID 27152928);可互换性随物质类型与方法对改变,两套框架可给不同判决(PMID 42869896、PMID 35977428、PMID 37470745)。

    部分验证:框架与不精密度剖面影响假拒绝率,已在模拟中成立(PMID 35647783),但尚无人拿丹田五参数的真实数据验证。

    待验证:丹田五参数各自的参考方法重复性从未被标定;%AB、φ、X5、AX、CoV 的可互换率与临界值域从未被计算;「率低于六成即禁合并」这条阈值的合理性有待实测。

    L1 网络状态:该主题尚未进入 L1 对应网络——CORRESPONDENCE_NETWORK.md 节点表中无任何测量学维度节点(无参考方法、可互换性、可互换率、临界值域、精密度剖面),边表 3.1–3.4 只登记五行↔五脏/五色/五音/五方/五季/五志、八卦↔五行、天干地支↔五行;EDGE_TABLE_science-yixue.md 仅有概念计数(丹田 51 篇 / 气功 48 篇 / 呼吸 189 篇),无 L1/L2/L3 层级与 active/open/conflict 状态判定。

    相关阅读

    证据等级: C
    参考文献: PMID 2917462, PMID 29443764, PMID 27152928, PMID 35647783, PMID 42869896, PMID 35977428, PMID 37470745, PMID 38222778, PMID 24716708, PMID 37377657, PMID 42862511


    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。

    Cross-Device \”Interchangeability\” Is Not an Attribute of Parameters but a Function of \”Method × Value Domain\”: Can the Five Special Parameters of the Dantian Be Merged? First, A…

    Evidence level: C

    The statement that a parameter is “93% commutable” is meaningless unless it clarifies the denominator; otherwise, it says nothing.

    I. Commutability Is Originally a Yes/No Question

    Lee 1989 (PMID 2917462) provides four rigorous criteria: no significant additive bias, no significant non-additive bias, the difference between means of two methods is not significant, and the lower bound of the 95% confidence interval for the intraclass correlation coefficient (ICC) is not below 0.75. All four must be met to deem a method commutable—it asks “Can A substitute B?” with an answer of only yes or no, leaving no room for a “rate.”

    The most easily overlooked criterion is the final one: it refers to an interval lower bound, not a point estimate. If $r=0.9$ but the sample size is only twelve, the lower bound often drops below 0.7, requiring a reversal of the judgment.

    The conversion to a rate was introduced by Lorne 2018 (PMID 29443764): first, an allowable difference limit is established based on the repeatability of the reference method; then, the proportion of differences between the test method and the reference method falling within this limit is calculated, along with a critical value. In a study involving five cardiac output monitors compared against the dilution method, the commutability rates were: Modelflow 93%, Hemac 86%, Wesseling 76%, LiDCO 73%, and PiCCO 62%.

    II. The Denominator Is the Portion Circumscribed by the Critical Value

    A sentence often omitted in the original text is: Modelflow’s 93% holds only “within a measurement range exceeding 95%,” with a critical value of 4.8 L/min. This number carries a prerequisite regarding the value range; outside this domain, the rate does not hold.

    The consequence is direct: two rates cannot be compared horizontally unless the measurement proportions covered by their critical values are identical. An A rate of 93% covering only sixty percent of measurements, versus a B rate of 73% covering ninety-five percent, leads to incorrect conclusions if one looks only at the rates.

    This is particularly fatal for Dantian parameters—%AB and $\phi$ inherently have narrow value ranges, whereas X5 spreads widely across height and weight. The same allowable difference limit yields inherently non-equivalent rates when calculated across domains of different widths.

    III. The Ceiling of the Rate Is the Repeatability of the Reference Method Itself

    A key limitation in Lorne’s conclusion states that the numerical value of the rate depends on the repeatability of the reference method. This can be derived: if the reference method’s own repeatability is already wider than the true deviation between two devices, the primary component within the difference is its noise. The rate is pinned down by this noise cloud; no matter how good the test device is, it cannot be raised.

    Currently, none of the three modalities’ reference methods have been calibrated to this step. The methodological variability of RIP phase angles has been dissected item by item, concluding it is far more fragile than the volume contribution ratio (PMID 24716708); forced oscillation quality control has long focused only on the intraclass coefficient of variation for R5, ignoring reactance, and has been criticized as insufficient (PMID 37377657); and there is still no unified terminology for adult respiratory mode devices versus outcome indicators (PMID 42862511).

    IV. Commutability Does Not Grant Equivalence

    The cat plasma study (PMID 27152928) separates these two terms: standardized residuals for commercial control materials and the cat plasma pool fell within $\pm 3$ for sixteen analytes, mostly holding true; however, compared to their coefficients of variation (CoV), only two achieved precision equivalence, while five were only marginally close at a single concentration level. The conclusion sentence is crisp: commutability does not grant equivalence.

    Translated into Dantian language: %AB can be commutable across devices (same quantity, bias within limits), but the CoV precision is not equivalent (different jitter amplitudes between two devices). Merging the CoVs of both devices for inter-group comparison is equivalent to mixing two rulers with different precisions—the mean might still be correct, but the variance will definitely be wrong. And the variance structure is precisely where “training for stability” truly lands.

    V. Two Frameworks Yield Different Judgments on the Same Substance

    The serum copper batch (PMID 42869896) used twenty-one reference materials and five systems, running both the IFCC difference bias method and the CLSI EP30 prediction interval method simultaneously: correlations were high, yet commutability varied by substance type and method pair. EP30 judged non-commutable more often, but not in all comparisons.

    The homocysteine batch (PMID 35977428) was more direct: five methods multiplied by seven substances yielded thirty-five combinations. According to EP30, thirty-four were commutable; according to IFCC, twenty-eight could yield a definitive conclusion while seven were indeterminate.

    Indeterminate results must be listed separately. The standard binary approach forces statistically unresolvable results into either side, systematically biasing one direction. The renin activity batch (PMID 37470745) was even colder: sixteen candidate substances showed fourteen non-commutable under the difference bias method; only large pools exceeding one thousand tended toward commutability.

    VI. Frameworks and Profiles Can Generate Conclusions Themselves

    One simulation study warrants separate note (PMID 35647783): four frameworks multiplied by four profiles. Under constant variance, Deming is most effective; under proportional variance, log transformation is most effective; mixed variance requires a framework capable of modeling “variance varying with concentration”; and for U-shaped profiles, even the most flexible framework is inappropriate.

    The Dantian five parameters exactly span these various profiles: CoV is proportional; X5 and AX may exhibit U-shapes at the extremes of body type; %AB and $\phi$ have ceilings. Using a single fixed framework to calculate commutability rates for five parameters means that a significant portion of the differences is a product of the framework. Another paper titled with a question (PMID 38222778) asks: Is commutability always desirable?

    Reversal: Rates Without a Denominator Are Invalid

    The original proposal was “to replace subjective Bland-Altman judgments with commutability rates.” The direction is correct—rates are reproducible and amenable to statistics. But a rate is only valid if three things report simultaneously: the rate itself, the measurement proportion covered by the critical value, and the reference method’s own repeatability.

    Reporting only the rate is equivalent to reporting a percentage without a denominator. And the third element has never been done for Dantian parameters: difference reliability has been calibrated, RCV has been used as a reference, and directional consistency has already been found to flip at the grid level—but no single paper has used reference method repeatability as a denominator prerequisite to calculate.

    A more thorough reading suggests: if the reference method repeatability for all five parameters is generally wider than the inter-device deviation, then “commutability rate” cannot be calculated here—not because of calculation error, but because it lacks a definition. The correct solution retreats to reporting bias and consistency limits parameter by parameter, explicitly stating “not allowed for cross-device merging.”

    Four-grid criteria: rate not below 90%, critical value covering over 80% of measurement range, reference method repeatability not exceeding half the inter-device deviation; all three must be met to serve as a basis for cross-device selection. If the rate is high but covers less than half, the rate is domain-created; cross-parameter comparison is forbidden. If the rate is between 60% and 90% and changes monotonically with the value range, report only zone-specific rates, not a single number. If the rate is below 60% or if two frameworks yield conflicting judgments, that parameter is forbidden from cross-device merging; only longitudinal comparisons within a single device are permitted. Regarding “indeterminate,” always list separately.

    Evidence and Limitations

    Verified: Commutability criteria have clear statistical definitions and confidence interval requirements (PMID 2917462); commutability rates have provided reproducible rates and critical values (PMID 29443764); commutability and precision equivalence are two distinct matters (PMID 27152928); commutability varies by substance type and method pair, with two frameworks potentially yielding different judgments (PMID 42869896, PMID 35977428, PMID 37470745).

    Partially Verified: Frameworks and imprecision profiles affect false rejection rates, established in simulations (PMID 35647783), but no one has yet verified this with real data from the Dantian five parameters.

    To Be Verified: The reference method repeatability for each of the Dantian five parameters has never been calibrated; commutability rates and critical value domains for %AB, $\phi$, X5, AX, and CoV have never been calculated; the rationality of the threshold “rate below 60% prohibits merging” awaits empirical testing.

    L1 Network Status: This topic has not yet entered the L1 corresponding network—the CORRESPONDENCE_NETWORK.md node table contains no metrological dimension nodes (no reference methods, commutability, commutability rates, critical value domains, or imprecision profiles); Edge Tables 3.1–3.4 only register Five Phases ↔ Five Organs/Five Colors/Five Tones/Five Directions/Five Seasons/Five Emotions, Eight Trigrams ↔ Five Phases, Heavenly Stems and Earthly Branches ↔ Five Phases; EDGE_TABLE_science-yixue.md only contains concept counts (Dantian 51 papers / Qigong 48 papers / Breathing 189 papers), lacking L1/L2/L3 levels and active/open/conflict status determinations.

    Related Reading

    Evidence Level: C
    References: PMID 2917462, PMID 29443764, PMID 27152928, PMID 35647783, PMID 42869896, PMID 35977428, PMID 37470745, PMID 38222778, PMID 24716708, PMID 37377657, PMID 42862511


    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.

  • 「沉而不塌」该怎么测:呼气末 X5 的电平值不是答案,离 DR2 坠落点的余量才是

    Evidence level: C

    「气沉丹田」若真有一个可测的肺力学对应物,它不该表现为「X5 更高」,而该表现为「呼气末位置离气道成片关闭的那个拐点更远」。

    正文

    丹田呼吸的传统表述里,「沉」和「塌」是一对反义词:气要往下走,但不能把胸廓压垮。把它翻成肺力学语言,最自然的读法是——呼气末肺容积(EELV)相对位置上移(沉),同时不进入小气道成片关闭的区间(不塌)。但「X5 系统性更高」这个具体预测,未必是这句话的正确操作化。

    第一,X5 是电平,不是状态。 强迫振荡的 5 Hz 电抗同时被身高、年龄、体重、肺容积拉着走,2026 年奥地利 LEAD 队列已给出成人振荡测量的参考方程(PMID 41497331)。不按参考方程报 z 值、不控肺容积,直接比两组的 X5 绝对值,比的其实是体型。更麻烦的是 X5 本身随肺容积漂移:把整段呼吸平均掉的做法,在哮喘与 COPD 之间根本分不开,只有拆成吸气相与呼气相再看 ΔX5 才有分辨力(PMID 20335298)。

    第二,「沉」是有代价的。 想把 EELV 撑上去,物理上有两条路:膈肌有效收缩把腹腔内容物往下推,或靠腹压把膈肌顶高。后一条路会反噬——腹腔高压对呼吸系统的后果有专门综述(PMID 32368883)。而「摆个姿势把肺撑起来」这件事,在超重与轻中度肥胖的 32 人里被直接否掉了:30° Fowler 位相对仰卧并未改善 FRC,且闭合容积与 FRC 之比在坐位就有 5/7 人大于 1,仰卧与 Fowler 位则是 7/7 全部大于 1,腰臀比与闭合容积相关(PMID 19245726)。换言之,靠姿势换来的容积,很容易换到「已经在塌的区间里」。

    第三,那个拐点是真实存在的,而且会自己往上跑。 Nilsen 等用慢肺活量动作扫 Xrs-容积曲线,自动检出两个点:X5 开始下降的 DR1,和下降开始加速的 DR2。健康对照 DR1 落在 46.1%TLC、DR2 落在 37.69%TLC;哮喘组分别是 76.58% 与 56.79%;对照组内 DR1、DR2 都随年龄上移,r² 分别为 0.68 与 0.71;DR2 与预计闭合容积相关 r=0.94(PMID 30702975)。这条结果对本文的假说是致命的:年龄单独就能把 DR2 推上去,而长期修炼者的人群年龄通常偏大,年龄的作用方向与「修炼者 DR2 更低」正好相反。

    第四,「不触发 DR2」得写成余量,不能写成是/否。 DR2 是曲线拟合出来的点估计,不是刻在器官上的刻度。电抗参数的重复性本就弱于电阻参数(PMID 36131305),而 G 理论的经典形式还假设残差齐性,不建模异质性会掩盖两组测量质量的差异(PMID 41549064)。两组若测量精度不同,DR2 位置的组间差里就掺了仪器成分。

    第五,现有呼吸训练证据全在别的终点上。 八段锦对术后肺功能的 meta 分析(PMID 40382884)、吸气肌训练对心脏手术结局的 meta 分析(PMID 39185909)、自主慢呼吸对心率与 HRV 的 meta 分析(PMID 35623448),终点都是 FEV1、FVC、六分钟步行或自主神经指标;瑜伽训练「降低估计肺龄」的研究(PMID 41816531)用的是回归方程反推的肺龄,不是实测容积;Kalaripayattu 练习者的呼吸功能研究(PMID 15521565)也只到肺量计层面。没有任何研究测过长期丹田呼吸者的 DR2 位置。

    第六,横断面分不清「练出来」和「筛出来」。 能坚持十年的人,可能本来胸廓条件就更好;这是自选择,不是训练效应。

    反转在这里:把「沉而不塌」操作化为 X5 电平,会被体型、年龄、肺容积三重污染;而操作化为余量——呼气末位置到 DR2 之间的相对距离,用 TLC 归一——则是在同一条曲线内部比较两点,个体尺度部分抵消,剩下的才是「这个人把呼气末停在了离崩塌多远的地方」。这是本篇要提的替代终点。

    余量怎么算:慢肺活量动作采 Xrs-容积曲线,按 Nilsen 的自动检出取 DR1、DR2 两点,把 DR2 与本次呼气末容积都换算成 %TLC,相减得余量(practitioner 若 EELV 在 50%TLC、DR2 在 38%TLC,余量 12 个百分点)。余量可为负——那就是呼气末已经越过坠落点,「塌」了。

    四条判法:余量显著更大、X5@EELV 不低于新手、且年龄匹配后仍在 ⇒ 「沉而不塌」成立;余量更大但全部由 DR2 上移贡献、EELV 没动 ⇒ 是「塌得更早」而非「沉得更深」,假说方向反了;X5@EELV 更高但余量不变 ⇒ 只是曲线整体平移,属体型效应,「沉」无独立证据;两组余量无差异 ⇒ 该假说在此指标上不可测,须改用体描箱或氦稀释直接测 EELV。

    配套三条:DR2 必须同报检出算法与重复测量一致性;X5 一律按参考方程报 z 值,不报绝对值;年龄按匹配变量处理而非协变量——r² 已达 0.71,协变量校正的外推风险太大。

    证据与局限

    已验证: X5-容积曲线上存在 DR1 与 DR2 两个去复张点,且健康对照 DR2 与预计闭合容积高度相关(PMID 30702975);成人振荡测量已有参考方程(PMID 41497331);整呼吸平均分析在哮喘与 COPD 间无分辨力、需拆分吸气/呼气相(PMID 20335298);超重人群姿势改变不能提升 FRC、闭合容积/FRC 比普遍大于 1(PMID 19245726);腹腔高压对呼吸系统有明确不利影响(PMID 32368883);IOS 电抗参数重复性弱于电阻参数(PMID 36131305);G 理论需扩展以处理异方差(PMID 41549064);八段锦、吸气肌训练、慢呼吸训练的现有 meta 分析终点均非 EELV 或 X5-容积曲线(PMID 40382884、PMID 39185909、PMID 35623448)。

    部分验证: 上述 DR1/DR2 证据来自 18 例哮喘与 18 例健康对照的小样本,且是横断面;「余量」这一替代终点本身尚未在任何人群中被计算过,其重测信度、最小可检测差异均未知。

    待验证: 长期丹田呼吸者的 DR2 位置是否真的更低;修炼者与新手在年龄匹配后余量是否仍有差异;余量指标自身的 ICC 与 MDC 是否足以支撑组间比较;「腹部用力撑高膈肌」与「膈肌有效收缩」在 X5-容积曲线上的指纹是否可分;横断面设计能否通过倾向评分部分剥离自选择。

    L1 网络状态:「该主题尚未进入 L1 对应网络」——CORRESPONDENCE_NETWORK.md 节点表只在生理侧列有「呼吸率 / 呼吸模式」观测项(备注「与节律耦合最直接」)与「胃肠蠕动=气沉丹田候选物」,在物理侧列有「相位 / 节律」,三者均未与易学侧任何节点连成边;节点表中没有「呼气末肺容积 / 气道闭合 / 肺力学曲线拐点」这类维度节点。边表 3.1–3.4 只登记五行↔五脏/五色/五音/五方/五季/五志、八卦↔五行、天干地支↔五行,无任何以 EELV、X5、DR2 或气道去复张为源或目标的边。EDGE_TABLE_science-yixue.md 仅有概念计数(丹田 51 篇 / 气功 48 篇 / 呼吸 189 篇),无层级(L1/L2/L3)与状态(active/open/conflict)判定。故本文封顶 C,所有机制表述均为方法学推演,不得读作已验证的生理机制。

    相关阅读

    证据等级: C
    参考文献: PMID 30702975, PMID 41497331, PMID 20335298, PMID 19245726, PMID 32368883, PMID 36131305, PMID 41549064, PMID 40382884, PMID 39185909, PMID 35623448, PMID 41816531, PMID 15521565


    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 Test \”Sink Without Collapse\”: End-Exhalation X5 Voltage Is Not the Answer; Margin to DR2 Drop Point Is

    Evidence level: C

    If “sinking into the dantian” truly has a measurable pulmonary mechanics counterpart, it should not manifest as “higher X5,” but rather as “the end-exhalation position being farther from the inflection point where airway segments close.”

    Main Text

    In traditional descriptions of dantian breathing, “sinking” and “collapsing” are antonyms: the qi should move downward but not crush the thoracic cage. Translated into pulmonary mechanics language, the most natural reading is—end-exhalation lung volume (EELV) shifting relatively upward (sinking), while not entering the zone where small airways close en masse (not collapsing). However, the specific prediction of “higher X5” may not be the correct operationalization of this statement.

    First, X5 is a voltage level, not a state. Forced oscillation’s 5 Hz reactance is simultaneously pulled by height, age, weight, and lung volume; the 2026 Austrian LEAD cohort has already provided reference equations for adult oscillatory measurements (PMID 41497331). Reporting z-values without following the reference equation and not controlling for lung volume directly compares absolute X5 values between two groups, which actually compares body habitus. More麻烦的是 (more problematic) is that X5 itself drifts with lung volume: averaging the entire breath fails to distinguish between asthma and COPD; only splitting into inspiratory and expiratory phases to examine ΔX5 provides discriminative power (PMID 20335298).

    Second, “sinking” comes at a cost. To inflate EELV, physics offers two paths: effective diaphragmatic contraction pushing abdominal contents downward, or using abdominal pressure to push the diaphragm up. The latter path backfires—abdominal hypertension has specific reviews detailing its consequences on the respiratory system (PMID 32368883). Furthermore, the act of “adopting a posture to inflate the lungs” was directly refuted in 32 individuals with overweight and mild-to-moderate obesity: the 30° Fowler position did not improve functional residual capacity (FRC) relative to supine, and the ratio of closing volume to FRC exceeded 1 in 5/7 individuals in the sitting position, and in 7/7 in both supine and Fowler positions; waist-to-hip ratio correlated with closing volume (PMID 19245726). In other words, volume gained through posture often comes at the cost of “already collapsing zones.”

    Third, that inflection point is real and tends to shift upward with age. Nilsen et al. used slow vital capacity maneuvers to sweep the Xrs-volume curve, automatically detecting two points: DR1 where X5 begins to decline, and DR2 where the decline accelerates. In healthy controls, DR1 fell at 46.1%TLC and DR2 at 37.69%TLC; in the asthma group, they were 76.58% and 56.79%, respectively. Within the control group, both DR1 and DR2 shifted upward with age (r² = 0.68 and 0.71, respectively); DR2 correlated strongly with predicted closing volume (r = 0.94) (PMID 30702975). This result is fatal to our hypothesis in this paper: age alone can push DR2 upward, whereas long-term practitioners typically have older populations, creating an age effect direction opposite to “lower practitioner DR2.”

    Fourth, “not triggering DR2” must be written as a margin, not as yes/no. DR2 is a point estimate derived from curve fitting, not a fixed scale etched onto an organ. IOS reactance parameters inherently have lower repeatability than resistance parameters (PMID 36131305), and the classic form of the G-theory assumes homoscedastic residuals; failing to model heterogeneity masks differences in measurement quality between groups (PMID 41549064). If measurement precision differs between groups, the group difference in DR2 position will be contaminated by instrument factors.

    Fifth, existing evidence for breathing training lies on other endpoints. Meta-analyses of Baduanjin on postoperative pulmonary function (PMID 40382884), inspiratory muscle training on cardiac surgery outcomes (PMID 39185909), and slow breathing on heart rate and HRV (PMID 35623448) all use endpoints like FEV1, FVC, six-minute walk, or autonomic indices; yoga training studies claiming “reduced estimated lung age” (PMID 41816531) use regression equations to back-calculate lung age, not measured volumes; studies on Kalaripayattu practitioners’ respiratory function (PMID 15521565) only reach spirometry levels. No study has measured the DR2 position in long-term dantian breathers.

    Sixth, cross-sectional designs cannot distinguish “trained out” from “screened out.” Those who persist for ten years may simply have better thoracic conditions initially; this is self-selection, not a training effect.

    The reversal here: Operationalizing “sinking without collapsing” as X5 voltage level gets contaminated by body habitus, age, and lung volume. Operationalizing it as a margin—the relative distance from end-exhalation position to DR2, normalized by TLC—compares two points within the same curve, partially canceling individual scale effects; what remains is “how far this person stopped end-exhalation from the collapse point.” This is the alternative endpoint proposed here.

    How to calculate the margin: Use slow vital capacity maneuvers to obtain the Xrs-volume curve, adopt Nilsen’s automatic detection to identify DR1 and DR2, convert both DR2 and the current end-exhalation volume to %TLC, and subtract to get the margin (if a practitioner has EELV at 50%TLC and DR2 at 38%TLC, the margin is 12 percentage points). The margin can be negative—that means end-exhalation has already crossed the collapse point, and “collapsing” has occurred.

    Four criteria: Margin significantly larger, X5@EELV not lower than novices, and still significant after age matching ⇒ “sinking without collapsing” holds; Margin larger but entirely contributed by DR2 shift with no EELV change ⇒ it is “collapsing earlier” rather than “sinking deeper,” reversing the hypothesis direction; X5@EELV higher but margin unchanged ⇒ only overall curve translation, a body habitus effect, no independent evidence for “sinking”; No difference in margins between groups ⇒ this hypothesis is not measurable via this indicator; switch to plethysmography or helium dilution to directly measure EELV.

    Three supporting protocols: DR2 must report detection algorithm and repeatability consistency; X5 must always report z-values per reference equation, not absolute values; age must be treated as a matching variable rather than a covariate—r² is already 0.71, extrapolation risk via covariate adjustment is too high.

    Evidence and Limitations

    Verified: DR1 and DR2 exist as derecruitment points on the X5-volume curve, and healthy control DR2 correlates highly with predicted closing volume (PMID 30702975); adult oscillatory measurements have reference equations (PMID 41497331); whole-breath averaging analysis lacks discriminative power between asthma and COPD, requiring splitting into inspiratory/expiratory phases (PMID 20335298); posture changes in overweight populations cannot improve FRC, and the closing volume/FRC ratio is generally >1 (PMID 19245726); abdominal hypertension has clear adverse effects on the respiratory system (PMID 32368883); IOS reactance parameters have lower repeatability than resistance parameters (PMID 36131305); G-theory requires extension to handle heteroscedasticity (PMID 41549064); existing meta-analyses of Baduanjin, inspiratory muscle training, and slow breathing training use endpoints other than EELV or X5-volume curves (PMID 40382884, PMID 39185909, PMID 35623448).

    Partially Verified: The above DR1/DR2 evidence comes from small samples (18 asthma cases and 18 healthy controls) in a cross-sectional design; the alternative endpoint “margin” itself has never been calculated in any population, and its test-retest reliability and minimum detectable difference remain unknown.

    To Be Verified: Whether DR2 position is truly lower in long-term dantian breathers; whether margin differences persist between practitioners and novices after age matching; whether the margin indicator’s ICC and MDC are sufficient to support group comparisons; whether fingerprints of “abdominal straining to elevate diaphragm” and “effective diaphragmatic contraction” are distinguishable on the X5-volume curve; whether cross-sectional designs can partially strip self-selection via propensity scoring.

    L1 Network Status: “This topic has not yet entered the L1 corresponding network”—the CORRESPONDENCE_NETWORK.md node table lists only “breathing rate / breathing pattern” on the physiological side (noted as “most directly coupled with rhythm”) and “gut peristalsis = candidate for sinking into dantian” on the physical side, and “phase / rhythm” generally; none connect to any node on the Yijing Studies side. The node table lacks dimension nodes such as “end-exhalation lung volume / airway closure / pulmonary mechanics curve inflection points.” Edge tables 3.1–3.4 only register Five Phases ↔ Five Organs/Five Colors/Five Tones/Five Directions/Five Seasons/Five Emotions, Eight Trigrams ↔ Five Phases, Heavenly Stems and Earthly Branches ↔ Five Phases; there are no edges with EELV, X5, DR2, or airway derecruitment as source or target. EDGE_TABLE_science-yixue.md only contains concept counts (Dantian 51 papers / Qigong 48 papers / Breathing 189 papers), with no hierarchy (L1/L2/L3) or state (active/open/conflict) determination. Therefore, this paper caps at Evidence Level C; all mechanism descriptions are methodological deductions and should not be read as verified physiological mechanisms.

    Related Reading

    Evidence Level: C
    References: PMID 30702975, PMID 41497331, PMID 20335298, PMID 19245726, PMID 32368883, PMID 36131305, PMID 41549064, PMID 40382884, PMID 39185909, PMID 35623448, PMID 41816531, PMID 15521565


    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.