Category: 科学易学

  • 遍数缺失机制分层的覆盖率裁决:均匀浮动、与被试相关、与结果相关三档缺失下,例数口径与混合模型口径何时失守

    Evidence level: C

    缺了百分之几不重要,为什么缺才重要:同一缺失率换一个缺失机制,例数口径与混合模型口径的覆盖率可以一个稳如泰山、一个直接塌方。

    正文

    前一篇已经论证过:气功外气判读里每例的「遍数」天然会不齐,样本量申报必须带上「遍数完整性」字段。但那条论证的第一轮仿真有一个隐藏假设——缺失是均匀浮动的:缺哪一遍与被试是谁、测出来是什么都无关。统计学上这叫 MCAR(完全随机缺失)。真实判读现场很少这么干净:补读往往集中在状态波动或结果可疑的例上,也就是缺失概率取决于那个没测到的值本身——这叫 MNAR(非随机缺失)。中间还有一档 MAR(随机缺失):缺失概率只取决于已经观测到的信息,比如被试的基础状态。三档机制的区分是重复测量模型无偏性讨论的标准前提(PMID 32033617)。

    关键在于:缺失机制的档位换一档,两种口径的命运就完全不同。 2025 年一项五法比较仿真给出了清晰的分界线(PMID 40247172):在 4-10% 的 MCAR 缺失下,所有方法都无偏、置信区间覆盖都在 95% 附近;而在高达 40% 的 MAR 脱落下,线性混合模型与协方差模式模型仍然无偏、覆盖仍接近 95%,而按例数分析的朴素口径(完全例分析、配对 t 检验)偏倚随脱落率上升、覆盖与把握度同步塌方。也就是说,在 MCAR 和 MAR 档,混合模型口径守得住,例数口径先失守。

    但 MNAR 档是另一回事。Yang 等 2014 年的 Monte Carlo 仿真(PMID 24079928)表明:传统混合模型在 MAR 下确实表现良好,可一旦缺失与结果本身相关(MNAR),混合模型同样失守——而且没有一种方法能在所有缺失机制下都无偏,作者的实际建议是先问「数据为什么缺」,再做跨机制假设的敏感性分析。更硬的证据来自解析推导:Yang、Wang、Maxwell 2015 年直接从数学上证明(PMID 25098455),在「随机斜率依赖」的 MNAR 缺失下,典型线性混合模型的均值截距与斜率估计都有偏,且斜率估计会被系统性地拉向完成者那一侧——偏倚不是抽样的运气问题,是权重结构被 MNAR 破坏的必然结果。

    对共振意识场这条研究线的直接含义是:判读现场「补读集中在可疑例」不是理论假设,而是 MNAR 的真实现场形态。如果遍数不齐数据的裁决只建立在均匀浮动假设的第一轮仿真上,那结论只能覆盖 MCAR 一档。所以这一题的裁决设计是:缺失机制三档 × 会话数两档 × 会话内相关两档的网格仿真,同批数据分别用例数口径与混合模型口径估计,记录各自置信区间覆盖率与偏差。裁决判据:若 MNAR 档混合模型覆盖率跌破 0.90 或绝对偏差超过 0.02,则遍数完整性申报必须加「缺失机制」字段,MNAR 场景一律改用敏感性分析框架申报;若 MNAR 档混合模型仍守住,则说明该场景的 MNAR 强度不足以击穿似然法的稳健性,单字段申报即可放行。

    证据与局限

    已验证:缺失机制三档的界定与重复测量模型无偏性条件(PMID 32033617);MCAR/MAR 档混合模型无偏且覆盖正常、例数口径在 MAR 档失守(PMID 40247172)。
    部分验证:MNAR 档传统混合模型会失守、无万全方法、须敏感性分析(PMID 24079928,Monte Carlo 仿真层面)。
    待验证:随机斜率依赖 MNAR 下混合模型的解析偏倚方向(PMID 25098455)能否推广到「遍数不齐 + 补读集中」这一具体场景;判读现场补读的真实 MNAR 强度从未被实测——这正是本篇裁决仿真要补的缺口。

    相关阅读

    L1 网络状态:本主题(缺失机制分层 × 口径覆盖率裁决)尚未进入 L1 对应网络;最近邻边 E-061「体验通道」当前状态为「L1→L2 依据具(工具就绪 + 跨情境失效风险已标注)|open」。本文为纯测量学方法学课题,证据等级限 C。

    证据等级: C
    参考文献: PMID 32033617, PMID 40247172, PMID 24079928, PMID 25098455

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索


    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。

    Coverage Decision of Missingness Mechanism Stratification: When Do Case Count and Mixed Model Approaches Fail Under Uniform, Subject-Related, and Outcome-Related Missingness?

    Evidence level: C

    The proportion of missing data is secondary; why the data is missing is critical. Under identical missingness rates, different missing mechanisms can cause the coverage of sample-based and mixed-model approaches to remain stable or collapse entirely.

    1. Introduction

    In Yijing Studies (易学), the application of pattern and timing (理气) often relies on specific structural analogies (结构类比). However, when integrating these traditional frameworks with modern statistical models—such as those used in Feng Shui (风水) or Bazi (八字)—researchers frequently encounter missing data. This section examines how the mechanism of missingness, rather than just its quantity, determines whether the sample-based approach and the mixed-model approach maintain their intended coverage.

    2. Stratified Coverage Decisions

    The decision on which model to apply depends on the nature of the missingness:

    Missingness Type Description Impact on Coverage
    Uniform Missingness Data is missing randomly across all subjects. Both sample-based and mixed-model approaches maintain stable coverage.
    Subject-Related Missingness Data is missing based on specific subject characteristics (e.g., birth year in Bazi). Coverage remains robust for both approaches.
    Outcome-Related Missingness Data is missing based on the result itself (e.g., whether a Feng Shui prediction was accurate). Critical failure: Sample-based coverage collapses; mixed-model coverage remains stable.

    3. Analysis of Mechanisms

    3.1 Uniform Floating Gaps

    When missingness is uniform, the distribution of the remaining data remains representative of the original population. Consequently, the sample-based approach (relying on observed cases) and the mixed-model approach (incorporating imputed or structural values) both retain their theoretical coverage.

    3.2 Subject-Related Gaps

    In contexts like Bazi, where certain pillars may be incomplete due to birth date constraints, the missingness is tied to the subject’s identity. As long as the missingness mechanism is independent of the outcome variable, both the sample-based and mixed-model frameworks continue to provide reliable coverage.

    3.3 Outcome-Related Gaps

    This scenario represents the most significant challenge in evidence-level (证据等级) analysis. If missingness is related to the outcome—such as when a Feng Shui form sha (形煞) is only recorded if it causes a negative event—the sample-based approach suffers from selection bias. In this case, the coverage of the sample-based method can collapse entirely, whereas the mixed-model approach, by accounting for the structural analogy of the missing data, maintains its coverage.

    4. Conclusion

    The stability of statistical inference in Chinese metaphysics research hinges on understanding the missingness mechanism. Uniform and subject-related gaps allow for flexible modeling, but outcome-related gaps necessitate the use of mixed models to preserve evidence integrity.

    Body

    The preceding article established that in the interpretation of external Qi Gong readings, the number of repetitions per case (“遍数”) is inherently uneven; therefore, sample size declarations must include a “repetition completeness” field. However, the first simulation round of that argument relied on a hidden assumption: missingness is uniformly floating—meaning the absence of a repetition is unrelated to which subject was tested or what value was measured. Statistically, this is known as MCAR (Missing Completely at Random). In real-world interpretation settings, such cleanliness is rare: supplementary readings often concentrate on cases exhibiting state fluctuations or suspicious results. In these instances, the probability of missingness depends on the unmeasured value itself—this is termed MNAR (Missing Not at Random). A middle category, MAR (Missing at Random), applies when missingness depends only on observed information, such as the subject’s baseline state. Distinguishing these three mechanisms is the standard prerequisite for discussing unbiasedness in repeated-measurement models (PMID 32033617).

    The critical point is: shifting the missingness mechanism by one level fundamentally alters the fate of the two analytical approaches. A 2025 comparative simulation involving five methods established a clear boundary (PMID 40247172): under MCAR with 4–10% missingness, all methods remain unbiased and confidence interval coverage stays near 95%; however, under MAR with up to 40% dropout, linear mixed models and covariance pattern models remain unbiased with coverage close to 95%, whereas the naive count-based approach (complete-case analysis, paired t-tests) suffers increasing bias as dropout rises, with coverage and power collapsing simultaneously. Thus, in MCAR and MAR regimes, the mixed-model approach holds steady while the count-based approach fails first.

    The MNAR regime is different. Yang et al.’s 2014 Monte Carlo simulation (PMID 24079928) showed that traditional mixed models perform well under MAR but also fail once missingness correlates with the outcome itself (MNAR)—and no single method remains unbiased across all missingness mechanisms. The authors’ practical recommendation is to first ask “why the data is missing,” followed by sensitivity analyses across different mechanisms. Stronger evidence comes from analytical derivation: Yang, Wang, and Maxwell (2015) proved mathematically (PMID 25098455) that under “random-slope-dependent” MNAR missingness, the mean intercept and slope estimates in typical linear mixed models are both biased, with slope estimates systematically pulled toward the side of completers. This bias is not due to sampling luck but is an inevitable consequence of MNAR destroying the weighting structure.

    For the research line on resonance consciousness fields, this has direct implications: the observation that “supplementary readings concentrate on suspicious cases” is not a theoretical assumption but the true MNAR form of the interpretation setting. If decisions based on uneven repetition counts rely solely on the first simulation round assuming uniform missingness, conclusions can only cover the MCAR case. Therefore, this decision framework is designed as follows: a grid simulation varying three missingness mechanisms (MCAR, MAR, MNAR) × two session counts × two intra-session correlation levels, applying both count-based and mixed-model estimation to the same dataset, recording respective confidence interval coverage and bias. The decision criterion is: if under MNAR the mixed-model coverage drops below 0.90 or absolute bias exceeds 0.02, then the repetition completeness declaration must add a “missingness mechanism” field, and MNAR scenarios shall always be reported using the sensitivity analysis framework; if the mixed model remains robust under MNAR, it indicates that the MNAR intensity in this scenario is insufficient to break the likelihood-based robustness, allowing single-field declaration.

    Glossary Note (non-promissory): Mixed-model approach refers to statistical methods that account for repeated measures and correlations within subjects, rather than treating each measurement as independent.

    Evidence and Limitations

    Verified: Definition of the three-tier mechanism gap and unbiasedness conditions for repeated-measurement models (PMID 32033617); MCAR/MAR tier mixed models remain unbiased with adequate coverage of normal cases, though the MAR tier fails under sample-size constraints (PMID 40247172).
    Partially Verified: Traditional mixed models for the MNAR tier exhibit bias, lack a universal solution, and require sensitivity analysis (PMID 24079928, Monte Carlo simulation level).
    Pending Verification: Whether the analytical bias direction of mixed models under MNAR with random slope dependence (PMID 25098455) generalizes to the specific scenario of “incomplete readings plus concentrated make-up readings”; the true MNAR intensity of on-site make-up readings has never been empirically measured—this represents the gap this article’s simulation aims to fill.

    Related Reading

    L1 Network Status: This topic (missingness mechanism layering × criterion coverage adjudication) has not yet entered the corresponding L1 network. The nearest-neighbor edge E-061 “Experience Channel” is currently in the state “L1→L2 basis established (tools ready + cross-context failure risk flagged) | open”. This article is a pure measurement methodology topic; evidence level is limited to C.

    Evidence Level: C
    References: PMID 32033617, PMID 40247172, PMID 24079928, PMID 25098455

    Evidence level: A = direct PubMed support; B = scientific analogy / indirect evidence; C = traditional consensus / preliminary exploration.


    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.

  • 数功龄比高低之前,先过「熟练测试者偏差」这一关:修炼者与新手的测量学习速率差如何污染气功生理横向比较

    Evidence level: C

    把功龄当分层变量之前,得先证明两组人对测量任务本身的学习速率一样快,否则横向比较里悄悄混进了「谁更会应付测试」这一项。

    正文

    气功生理研究里最常见的设计是按功龄分层:把修炼者和新手拉到一起,比心率变异性、皮温、脑电,然后宣布「功龄越长,指标越不一样」。这个设计有一个默认假设——测的是功,不是「会做实验」的能力。但测量本身有学习曲线,这个假设站不站得住,得先看数据。

    练习效应有多普遍?健康成年人重复做同一套认知测试,早期高频测试阶段的学习效应 Cohen’s d 达 0.36–1.19,而且前两三次最猛,之后进入平台(PMID 20846444)。不止认知测试:隔一周重测,连日常生活功能量表和操作类测量都显著提升(PMID 40844017)。也就是说,「第一次测」和「第二次测」根本不是同一个量——任何人第一次进实验室,都带着一条还没走完的学习曲线。

    麻烦在于,专家和新手的学习速率不一样。视运动适应任务里的直接证据:新手靠更快的隐式学习过程,专家的显式策略成分更强、隐式学习更慢但保持更好(PMID 25713526)。fMRI 元分析显示,长期运动训练者与新手的脑激活差异集中在动作计划与动作理解区域(PMID 25450866);运动员在多目标追踪上的优势幅度还会被任务参数(目标数、追踪时长)系统调节(PMID 39209919)。把这些迁移到气功情境:修炼者长年练习呼吸控制和注意调节,面对「按提示呼吸同时测 HRV」这类任务,可能天然带着「预习过」的状态;新手却要现场学怎么配合仪器。这就是熟练测试者偏差——两组人面对同一个任务,处于学习曲线的不同位置。

    污染方向还不止一个,这正是它阴险的地方。方向一:新手首测紧张、不会配合,基线被学习增益压低,组间差异被夸大——看起来像「功龄拉开差距」,其实是「新手还没学会被测」。方向二:修炼者对测量情境早已熟透、首测即平台,而新手复测时迅速追平一部分差距,此时只看单次测量反而低估了真实差异。哪个方向占上风,取决于「任务熟悉度」和「功龄」哪个是主导变量——不专门测出来,就永远不知道该往哪边修。

    气功测量还有两个放大器,让这个问题比一般运动科学更尖锐。第一,气功的核心操作(调息、调心)和测量任务的要求高度重叠:让被试「缓慢深呼吸」对新手是一需要现场学的技能,对修炼者是练了十年的日常——同一句指导语,两组人接收到的任务难度完全不同。第二,得气强度这类主观指标靠自评量表,而量表本身也有使用学习:第一次填的人往往不知道「麻」「胀」「沉」该对应几分,第二次就知道往哪个锚点靠。这两条叠加意味着,气功横向比较里的学习增益污染通道比普通认知测试更多、更隐蔽。

    所以可测的方案是现成的:首测–复测差值按功龄分层。同一批被试隔至少一周测两次,把首复测差当作各自的学习增益估计,再拟合「修炼年限 × 学习曲线斜率」的混合模型。判据可以这样定:若学习增益与功龄无关(交互项斜率置信区间跨零且效应量小于 0.1),功龄分层的横向比较可以不校正;若修炼者的学习增益系统性低于新手(差距达到新手增益的一半以上),那么任何不带回测的功龄分层比较都必须标注「污染未校正」,而且首测数据不能单独当基线用。

    一句话:比功龄之前,先让测量本身对两组人一样公平。

    证据与局限

    已验证:练习效应在认知测试与功能测量中普遍存在且量级可观(PMID 20846444、40844017);专家与新手的隐式/显式学习过程结构确实不同(PMID 25713526)。

    部分验证:学习速率差异在运动技能领域可量化,且受任务参数调节(PMID 25450866、39209919)——但「气功修炼者」作为一类被试的学习速率特征尚无直接测量。

    待验证:功龄与测量学习增益之间的剂量关系;污染的方向(夸大还是缩小组间差异);首复测间隔的最短定标(多久之后学习曲线才进入平台)。

    相关阅读

    证据等级: C
    参考文献: PMID 20846444, PMID 40844017, PMID 25713526, PMID 25450866, PMID 39209919

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索

    L1 网络状态:该主题尚未进入 L1 对应网络(CORRESPONDENCE_NETWORK / EDGE_TABLE_science-yixue 中无「功龄 × 测量学习速率污染」对应边;最近邻 E-061 呼吸节律↔自主神经谱,L2/open)


    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。

    Before comparing proficiency ratios, overcome the \”proficiency tester bias\”: how the measurement learning rate gap between practitioners and novices contaminates cross-sectional…

    Evidence level: C

    Before comparing proficiency-to-age ratios, the “proficiency-tester bias” must be addressed

    Before treating years of practice as a stratification variable, it is necessary to demonstrate that both groups exhibit identical learning rates on the measurement task itself; otherwise, lateral comparisons inadvertently conflate performance with “who is better at taking the test.”

    Body

    The most common design in qigong physiological research stratifies by practice duration: bringing practitioners and novices together to compare heart rate variability, skin temperature, and EEG, then declaring that “longer practice yields different metrics.” This design rests on a default assumption—that the measurement captures qigong, not the ability to perform experiments. Yet measurement itself has a learning curve; whether this assumption holds must be judged by the data.

    How pervasive is the practice effect? In healthy adults repeatedly taking the same cognitive test, the early high-frequency testing phase shows a learning effect with Cohen’s d ranging from 0.36 to 1.19, peaking in the first two or three trials before plateauing (PMID 20846444). Beyond cognitive tests: retesting after one week also significantly elevates daily living function scales and performance-based measures (PMID 40844017). In other words, the “first test” and the “second test” are not measuring the same quantity—anyone entering the lab for the first time brings an unfinished learning curve.

    The complication is that experts and novices learn at different rates. Direct evidence comes from visuo-motor adaptation tasks: novices rely on faster implicit learning processes, whereas experts exhibit stronger explicit strategy components and slower but better-maintained implicit learning (PMID 25713526). fMRI meta-analysis shows that long-term exercise trainees differ from novices primarily in brain activation within action planning and action understanding regions (PMID 25450866); athletes’ advantage in multiple-object tracking is systematically modulated by task parameters (target number, tracking duration) (PMID 39209919). Translating this to qigong contexts: practitioners, having long trained breath control and attention regulation, may approach tasks like “breathe according to prompts while measuring HRV” with a pre-learned state; novices must learn on the spot how to coordinate with instruments. This is the proficient tester bias—two groups facing the same task occupy different positions on the learning curve.

    The direction of contamination is not merely one-sided; this is where it becomes insidious. Direction one: novices are tense during the first test, unable to cooperate, so their baseline is depressed by learning gains, exaggerating between-group differences—making it appear as though “practice duration creates gaps,” when in fact “novices haven’t learned how to be tested.” Direction two: practitioners are thoroughly familiar with the measurement context, reaching plateau on the first test, while novices rapidly close part of the gap during retesting; in this case, relying on single measurements underestimates the true difference. Which direction dominates depends on whether “task familiarity” or “practice duration” is the dominant variable—without specifically measuring this, one cannot know which direction to correct.

    Qigong measurement also has two amplifiers that make this issue sharper than in general exercise science. First, the core operations of qigong (breath regulation, mind regulation) highly overlap with the requirements of measurement tasks: instructing subjects to “breathe slowly” is a skill novices must learn on the spot, whereas for practitioners it is a daily routine honed over ten years—the same instruction yields entirely different task difficulty for the two groups. Second, subjective indicators like qi intensity rely on self-rating scales, which themselves have usage learning: those filling the scale for the first time often do not know how many points correspond to “numbness,” “distension,” or “heaviness,” whereas those taking it the second time know which anchor point to aim for. Together, these mean that learning-gain contamination channels in qigong cross-group comparisons are more numerous and more concealed than in general cognitive tests.

    Therefore, the measurable solution is straightforward: compute the difference between first and second tests stratified by practice duration. The same batch of subjects is tested twice, separated by at least one week; the first-second difference serves as an estimate of individual learning gain, and a mixed model fitting “practice duration × learning curve slope” is constructed. The criterion can be set as follows: if learning gain is independent of practice duration (the interaction term slope confidence interval crosses zero and the effect size is less than 0.1), then cross-group comparisons stratified by practice duration need no correction; if practitioners’ learning gain is systematically lower than that of novices (the gap exceeds half the novice gain), then any practice-duration-stratified comparison without retest must be labeled “uncorrected for contamination,” and first-test data cannot be used alone as baseline.

    In one sentence: before comparing by practice duration, first make the measurement itself equally fair for both groups.

    Evidence and Limitations

    Verified: Practice effects are ubiquitous and of substantial magnitude in cognitive testing and functional measurement (PMID 20846444, 40844017); the implicit/explicit learning processes of experts and novices indeed differ in structure (PMID 25713526).

    Partially Verified: Differences in learning rates are quantifiable in the domain of motor skills and modulated by task parameters (PMID 25450866, 39209919)—however, the learning rate characteristics of “Qigong practitioners” as a subject group have not yet been directly measured.

    Pending Verification: The dose-response relationship between years of practice and measured learning gains; the direction of contamination (whether it exaggerates or attenuates between-group differences); and the minimum calibration for the first retest interval (after how long the learning curve enters a plateau).

    Related Reading

    Evidence level: C
    References: PMID 20846444, PMID 40844017, PMID 25713526, PMID 25450866, PMID 39209919

    Evidence level: A = directly supported by PubMed, B = scientific analogy or indirect evidence, C = traditional consensus / preliminary exploration

    L1 network status: This topic has not yet entered the L1 correspondence network (no edge corresponding to “practice duration × measurement learning-rate contamination” exists in CORRESPONDENCE_NETWORK / EDGE_TABLE_science-yixue; the nearest neighbor E-061 links respiratory rhythm ↔ autonomic spectrum, L2/open).


    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.

  • 幂律、有理式还是样条:s*(m) 曲线参数化形式的模型选择裁决

    Evidence level: C

    一条拟合曲线的形状不是画出来的,是选出来的——选错了形状,外推到高 m 档就是系统性错路,而且错得很自信。

    正文(1000-1500字)

    同一批八档 m 仿真网格数据,三条候选曲线都”能拟合”:幂律(两头平滑、可外推)、有理式(能出渐近线、能出拐点)、样条(贴数据最紧、但外推区行为完全由端点惯性决定)。拟合优度上样条几乎永远赢——参数多、自由度大,把每个网格点的残差都吃掉了。但本系列前一篇已经说过:我们关心的不是网格内的贴合度,是外推到 m=24、m=32 之外的 α 预告值。样条在训练区赢来的每一分贴合度,都可能在外推区变成一份没有约束的任性。

    所以裁决标准必须换成:留一外推误差。做法是同一套规则跑三遍:每次从八档里留出一档(优先留高端档,因为外推误差在高端最要命),用剩下七档拟合,再去预告留出档,记录误差。三个候选各跑完,比谁的外推误差小。这不是我们发明的招——生态学里 548 个物种的大规模模型选择就是这么做的,逐观测留一交叉验证(LOOCV)被证明比 WAIC 这类廉价替代品更可靠(PMID 32324930);交叉验证驱动的模型选择算法本身就是为了防”过拟合数据”这一件事而设计的(PMID 25505354)。

    但有三个坑必须提前写进规则里。

    坑一:留一验证在小样本上自己也会乐观。 方法学研究用多个临床数据集实打实比过:样本小时,leave-1-out 交叉验证给出的校正估计有偏,带重复的交叉验证和 bootstrap 表现更稳(PMID 24966219)。八档 m 是很小的数据,所以裁决不能只跑一轮 LOOCV 就拍板——须在仿真母体上做重复生成(不同随机种子重抽网格),报外推误差的分布而不是单点值。哪条曲线的误差分布右尾更短,谁才配当默认形式。

    坑二:交叉验证的切法必须贴着外推任务来。 空气质量预测领域有现成教训:随机切分的十折交叉验证把 RMSE 报成 2.10,换成按空间距离留出远处站点的切法,RMSE 立刻涨到 3.11——随机切分高估了外推能力约 48%(PMID 33122961)。对应到我们的网格:如果随机留档,等于测”插值”;要想测”外推”,必须只留高端档(m=24、32),让模型用低中档去够高端。这是本裁决与一般模型选择最大的不同点。

    坑三:拐点检验先于形式选择。 幂律和有理式的关键分歧在 m=12/24 附近有没有拐点。判法用残差符号游程检验:三候选各自拟合后,把残差按 m 排序看符号序列——若幂律拟合的残差符号出现”先正后负再正”的系统性游程,就说明数据里有幂律装不下的弯曲,拐点存在,裁决应倒向有理式或样条。游程检验是每个统计软件都有的老工具,成本为零。

    三条规则合起来,裁决流程就定型了:同一网格数据、三形式、高端留档的重复 LOOCV、残差游程检验拐点、误差分布右尾定胜负。胜者成为 s*(m) 的默认参数化形式,写进后续所有外推预算(包括前一篇的几何因子传播带宽表)的底层假设;败者不删,留作敏感性分析的对照组。

    对整条”气功外气”测量线来说,这一步看着枯燥,其实是地基:曲线形式选错,后面所有 α 预告、所有带宽、所有”m=32 档 0.02 传到 0.3″的算术都建立在一个错误形状上。测量学文献反复验证的那句话在这里同样成立——模型选择的正确做法不是看谁拟合得漂亮,是看谁在没见过的数据上错得少(PMID 32324930、25505354)。外推世界里,谦虚的曲线比自信的曲线活得久。

    证据与局限

    已验证: 留一交叉验证作为模型选择裁判在大规模真实数据上可行且优于 WAIC 等替代品(PMID 32324930);小样本下 LOOCV 存在乐观偏差、重复 CV 与 bootstrap 更稳(PMID 24966219);随机切分交叉验证会系统性高估外推能力,须用与外推任务同构的切法(PMID 33122961);CV 驱动的模型选择可有效抑制过拟合(PMID 25505354)。

    部分验证: 高端留档切法在空间外推领域有直接先例(空间 CV),移植到”高档 m 留出”属同类结构,但未在本系列网格上实测。

    待验证: 三候选(幂律/有理式/样条)在同一八档网格上的外推误差头对头;m=12/24 拐点的残差符号游程检验;重复生成下误差分布右尾比较。全部为本系列提出的仿真裁决方案,结果未出。

    相关阅读

    证据等级: C
    参考文献: PMID 32324930, PMID 24966219, PMID 25505354, PMID 33122961

    L1 网络状态:该主题尚未进入 L1 对应网络。最近邻边 E-061(呼吸节律 0.1 Hz 谱峰 ↔ Mayer 波/自主神经谱,层级 L2+,状态 open)——本文为纯测量学方法学课题,不依赖任何传统对应边,故按规则取 C 级。


    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。

    Power law, rational function or spline: Model selection for the parameterization of the s*(m) curve

    Evidence level: C

    The shape of a fitted curve is not drawn; it is selected. Selecting the wrong shape leads to systematic error when extrapolating to high m values, and this error is often made with confidence.

    1. Introduction: The s*(m) Framework and Model Selection

    The study of the relationship between spatial frequency $s$ and magnitude $m$—denoted as $s^*(m)$—is central to understanding the underlying architecture of complex systems, from neural networks to urban planning. This relationship is often modeled using parametric curves, but the choice of model significantly impacts the interpretation of results.

    Three primary models are frequently considered:
    1. Power Law: $s^(m) \propto m^\alpha$
    2.
    Rational Function: $s^(m) = \frac{P(m)}{Q(m)}$
    3. Spline: A piecewise polynomial interpolation.

    Each model offers a different structural analogy for how the system organizes itself. The power law implies scale invariance; the rational function suggests a saturation or asymptotic behavior; and the spline allows for arbitrary local variations. However, these are not merely mathematical conveniences; they represent competing hypotheses about the frequency epistemology of the system.

    2. The Power Law Model: Scale Invariance and Its Limits

    The power law model posits that $s^(m)$ follows a simple scaling relationship:
    $$ s^
    (m) = A \cdot m^\alpha $$
    where $A$ is a scaling constant and $\alpha$ is the exponent.

    Traditional Consensus: In many natural and social systems, power laws are observed as a signature of self-organized criticality or fractal geometry.

    Evidence Level: A (Strong empirical support in physics and biology; mixed in social sciences).

    Limitations:
    While elegant, the power law model often fails when extrapolating beyond the observed range of $m$. If the underlying process is not truly scale-invariant, the power law fit may appear good over a limited range but diverge significantly at higher $m$ values. This divergence is not a minor deviation; it represents a systematic error in the model’s assumptions.

    3. The Rational Function Model: Saturation and Asymptotes

    The rational function model proposes that $s^(m)$ is the ratio of two polynomials:
    $$ s^
    (m) = \frac{P(m)}{Q(m)} $$
    where $P(m)$ and $Q(m)$ are polynomials of degrees $n$ and $d$, respectively.

    Traditional Consensus: Rational functions are useful for modeling systems with saturation effects, such as enzyme kinetics or traffic flow.

    Evidence Level: B (Moderate support; often fits data better than power laws in constrained environments).

    Limitations:
    The rational function model can capture asymptotic behavior, which the power law cannot. However, it requires more parameters and may overfit if the degrees of $P$ and $Q$ are not justified by the underlying pattern and timing (Qi) of the system.

    4. The Spline Model: Flexibility and Overfitting

    The spline model uses piecewise polynomials to interpolate the data points, allowing for arbitrary curvature.

    Traditional Consensus: Splines are flexible and can fit any set of points exactly.

    Evidence Level: C (Weak; high risk of overfitting).

    Limitations:
    While splines can fit the observed data perfectly, they lack predictive power for unobserved $m$ values. Extrapolating a spline often leads to unrealistic behavior, such as oscillations or unbounded growth, which contradicts the physical or logical constraints of the system.

    5. Model Selection Criteria: Beyond Goodness-of-Fit

    Choosing the correct model for $s^*(m)$ requires more than just minimizing the sum of squared errors. We must consider:

    • Extrapolation Behavior: How does the model behave at high $m$?
    • Parameter Parsimony: Is the added complexity justified by the evidence?
    • Structural Analogy: Does the model reflect the underlying frequency epistemology of the system?

    Table 1: Comparison of Models for $s^*(m)$

    Model Equation Extrapolation Behavior Evidence Level Structural Analogy
    Power Law $s^*(m) = A \cdot m^\alpha$ Unbounded growth or decay; no saturation A Scale invariance, fractal geometry
    Rational Function $s^*(m) = \frac{P(m)}{Q(m)}$ Asymptotic convergence; saturation possible B Saturation, finite capacity
    Spline Piecewise polynomial Arbitrary; often oscillatory or unbounded C Local variation, no global structure

    6. Conclusion: The Importance of Model Selection

    The choice of model for $s^*(m)$ is not trivial. A power law fit may be statistically adequate over a limited range but fail catastrophically when extrapolated. A rational function may capture saturation but require unjustified complexity. A spline may fit perfectly but predict nonsense.

    Therefore, model selection must be guided by both empirical evidence and theoretical plausibility. The frequency epistemology of the system—its underlying pattern and timing—must inform the choice of model, not just the fit to observed data.

    In summary, the shape of $s^*(m)$ is not drawn; it is selected. And selecting the wrong shape leads to systematic error, often with great confidence.

    Body (1000–1500 words)

    Using the same set of eight-grid m simulation mesh data, all three candidate curves—power law, rational function, and spline—can “fit”: the power law is smooth at both ends and allows extrapolation; the rational function can produce asymptotes and inflection points; the spline adheres most tightly to the data but its extrapolation behavior is entirely determined by endpoint inertia. In terms of goodness-of-fit, the spline almost always wins—its many parameters and high degrees of freedom absorb residuals at every mesh point. Yet, as the previous article in this series noted, what matters is not fit within the mesh but extrapolation beyond m=24 and m=32 to predict α values. Every point of fit the spline gains in the training region may become unconstrained arbitrariness in the extrapolation region.

    Therefore, the decision criterion must shift to leave-one-out extrapolation error. The procedure applies the same rules three times: each run leaves out one of the eight grids (prioritizing high-end grids, because extrapolation error is most severe there), fits the remaining seven, predicts the left-out grid, and records the error. After running all three candidates, we compare whose extrapolation error is smallest. This is not an invention of ours: ecology has used leave-one-out cross-validation (LOOCV) for massive model selection across 548 species, and LOOCV has been shown to be more reliable than cheaper alternatives like WAIC (PMID 32324930). Model selection algorithms driven by cross-validation are themselves designed to prevent “overfitting data” (PMID 25505354).

    However, three pitfalls must be built into the rules in advance.

    Pitfall 1: LOOCV is inherently optimistic on small samples. Methodological studies using multiple clinical datasets have shown that with few samples, leave-one-out cross-validation yields biased corrected estimates; repeated cross-validation and bootstrap methods perform more stably (PMID 24966219). Eight grids constitute a very small dataset, so the decision cannot rest on a single LOOCV run. Instead, we must generate repeats from the simulation population (re-drawing the mesh with different random seeds) and report the distribution of extrapolation errors rather than a single point estimate. The curve whose error distribution has the shorter right tail earns default status.

    Pitfall 2: The cross-validation split must align with the extrapolation task. Air quality prediction offers a clear lesson: random ten-fold cross-validation reported RMSE = 2.10, but when the split respected spatial distance—leaving out distant stations—the RMSE jumped to 3.11; random splitting overestimated extrapolation ability by about 48% (PMID 33122961). Translating to our mesh: random leave-one-out measures “interpolation”; to measure “extrapolation,” we must only leave out high-end grids (m=24, 32), forcing the model to reach for higher values using low and mid grids. This is the key distinction between our decision and general model selection.

    Pitfall 3: Inflection point testing precedes form selection. The critical divergence between power law and rational function lies in whether an inflection point exists near m=12/24. The test uses residual sign run tests: after each candidate fits the data, we sort residuals by m and examine the sequence of signs. If the power-law fit shows a systematic run such as “positive, then negative, then positive,” it indicates curvature the power law cannot capture; an inflection point exists, and the decision should favor the rational function or spline. Run tests are a standard statistical tool available in every software package, costing nothing.

    Combining these three rules yields the finalized decision workflow: same mesh data, three forms, repeated LOOCV with high-end leave-outs, residual run test for inflection points, and right-tail comparison of error distributions to determine the winner. The winner becomes the default parameterization s*(m) and underpins all subsequent extrapolation budgets—including the geometric factor bandwidth table from the previous article. Losers are not discarded; they serve as controls for sensitivity analysis.

    For the entire “qigong external qi” measurement line, this step may appear dry, yet it is foundational: choosing the wrong curve form invalidates all α predictions, all bandwidths, and all arithmetic such as “m=32 grid 0.02 propagates to 0.3.” The measurement literature repeatedly confirms that correct model selection is not about who fits beautifully but about who errs least on unseen data (PMID 32324930, 25505354). In the extrapolation world, humble curves outlive confident ones.

    Evidence and Limitations

    Verified: Leave-one-out cross-validation serves as a robust model selection criterion on large-scale real-world data, outperforming alternatives such as WAIC (PMID 32324930); under small-sample conditions, leave-one-out cross-validation exhibits optimistic bias, making repeated cross-validation and bootstrap methods more stable (PMID 24966219); random splitting cross-validation systematically overestimates extrapolation capacity and requires splits isomorphic to the extrapolation task (PMID 33122961); cross-validation-driven model selection effectively suppresses overfitting (PMID 25505354).

    Partially Verified: High-end leave-one-out splitting has direct precedent in spatial extrapolation (spatial cross-validation); transposing this to “high-end m left-out” constitutes a structurally analogous approach, though it has not been empirically tested on the current series of grids.

    Pending Verification: Head-to-head extrapolation error comparison among three candidates (power law, rational function, spline) on the same eight-tier grid; residual sign run tests at m=12/24 inflection points; right-tail distribution comparison under repeated generation. These are all simulation-based adjudication schemes proposed in this series; results are not yet available.

    Related Reading

    Evidence level: C
    References: PMID 32324930, PMID 24966219, PMID 25505354, PMID 33122961

    L1 network status: This topic has not yet entered the corresponding L1 network. The nearest neighbor edge E-061 (respiratory rhythm 0.1 Hz spectral peak ↔ Mayer wave / autonomic spectrum, level L2+, status open) — this article is a pure measurement methodology topic and does not rely on any traditional corresponding edge, hence it is assigned level C according to the rules.


    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.

  • 遍数完整性元数据在存量判读数据里的完备度审计与补录成本估计

    Evidence level: C

    申报口径升级之前,先回答一个更基本的问题:历史判读记录里到底还找不找得到每一遍的原始读数——如果逐遍数据没存下来,任何针对”不齐场景”的加权或混合模型口径都只是纸面设计。

    正文

    在一个以重复判读为基础的测量体系里,”每例测了几遍”不是一个可有可无的备注,而是决定统计口径的地基。前文已经论证过:当各例遍数不齐时,把例数当成样本量的朴素口径会悄悄改变有效信息量,需要改用加权或混合模型口径来修正。但这条修正路线有一个很少被正视的前置条件——逐遍数据必须可回溯。如果历史记录在入库时只存了每例的均值或汇总分,逐遍的原始读数已经丢失,那么不齐口径再正确也无从实施,因为它需要的输入根本不存在。

    这正是为什么口径升级的第一步不是写统计公式,而是做一次存量数据的完备度审计:抽取足够多的历史判读会话记录,把每一例的遍数元数据分成三档清点——”逐遍数据可回溯”、”仅有均值汇总”、”遍数记录本身缺失”。三档的占比直接决定下一步走哪条路:若逐遍可回溯的比例足够高(例如不低于八成),不齐口径可以直接升级,剩下两成按缺失处理;若比例不够,就必须先定补录规则,并给补录标出真实成本。

    这个”先审计完备度、再谈方法升级”的次序,在医学登记库的数据质量研究里是被反复验证的标准动作。挪威急性心梗国家登记库做过一次系统的变量级审计:审计护士对 280 例患者的原始病历重新摘录,与登记库已录数据逐变量比对,结果发现多数变量一致性优良,但症状起病时间、家族史、体重指数等关键字段的缺失率高得超出预期——也就是说,不看原始记录,登记库自己不会告诉你哪些字段其实没存好(PMID 27574467)。急性卒中登记库的原型评估走得更细:把数据元素分成十二类逐段统计完备率,最低的”溶栓治疗”段只有约 53.5%,最高段接近 99.6%,同一套字段在不同采集方式下完备度差异巨大,而且回溯式采集与前瞻式采集在完备度上并无稳定优势(PMID 17178303)。赫尔辛基创伤登记库的内部审计则把审计本身拆成了四个可报告的维度——病例完整性、数据完备性、准确性、正确性,并建议登记库定期系统性地报告这四项,而不是只笼统说”数据质量还行”(PMID 31694457)。

    这三个先例对判读遍数元数据的审计有两条直接启示。第一,完备度在字段之间高度不均:核心字段往往接近全备,而逐遍原始读数这类过程性字段,恰恰是最容易在汇总入库时被丢弃的一层——卒中登记里缺失率最高的正是过程性最强的字段,这个模式在判读台账里同样会重演。所以审计不能只抽均值层面的记录,必须专门检查逐遍层是否留存。第二,审计工具自身也要校准:一项对电子病历审计工具的信度研究显示,即使经过培训,审计员之间的一致性也只是中等水平(PMID 34320964)。如果几个审计员对”这一例逐遍数据算不算可回溯”给出不同答案,三档占比就不可信;审计开始前须先在小样本上做审计员间比对,把判定规则写成操作手册。

    补录成本的估计同样有先例可循。心梗登记库的审计结论明确指出,精确的字段定义与摘录员培训是保住数据质量的前提——这意味着补录的大头成本不在重新判读本身,而在把判定规则写清楚、把人员训练到位。据此可把补录成本拆成三块:逐遍可回溯的例(零成本)、需回看原始录像或波形重新计遍的例(按工时计价)、原始载体已丢失只能降级为近似口径的例(成本转嫁为统计口径的损失)。三块占比从审计里直接读出,补录总成本就是一份可量化的预算,而不是一句”回头补一下”。

    对这套判读体系而言,这次审计的意义超出单一课题:它是把口径升级从纸面推向可执行的第一道闸门,也是元数据治理规范(窗长、缺失机制、遍数完整性三个字段)里最先要回填真实数据的一个格子。

    证据与局限

    已验证: 医学登记库领域”变量级审计揭示字段级缺失不均”是成熟范式(PMID 27574467、17178303、31694457);审计工具自身的评分者间信度须先行校准有直接文献支撑(PMID 34320964)。

    部分验证: 完备度三档划分与”逐遍层最易丢失”的模式推断,系从登记库字段完备度模式类比而来,判读台账的实际分布未测。

    待验证: 判读存量记录的三档占比本身(本课题核心判据);补录三档的成本单价;审计判定规则的手册化能否把审计员间一致性抬到优良档。

    相关阅读

    L1 网络状态:该主题尚未进入 L1 对应网络(CORRESPONDENCE_NETWORK 中无「遍数完整性元数据 × 不齐口径可实施性」对应边;最近邻为体验通道 L2/open 层级的边),按规则本文证据等级封顶为 C。

    证据等级: C
    参考文献: PMID 27574467, PMID 17178303, PMID 31694457, PMID 34320964

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索


    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。

    Completeness Audit and Imputation Cost Estimation of Inventory Assessment Data Using Fullness Metadata

    Evidence level: C

    Completeness Audit of Iteration Counts in Stock Interpretation Data and Estimation of Supplemental Recording Costs

    Before upgrading the reporting scope, address a more fundamental question: Can original readings for every iteration still be located within historical interpretation records? If sequential iteration data is not preserved, any weighted or hybrid model scope targeting “incomplete scenarios” remains merely theoretical design.

    Metric Definition / Evidence Level
    Iteration count Number of complete cycles recorded (Evidence Level A)
    Data gap Missing readings between iterations (Evidence Level B)
    Supplemental cost Estimated effort to reconstruct missing data (Evidence Level C)
    • Iteration count: Number of complete cycles recorded.
    • Data gap: Missing readings between iterations.
    • Supplemental cost: Estimated effort to reconstruct missing data.

    Methodology for Completeness Audit

    1. Review existing records: Examine all available historical data to identify gaps in iteration counts.
    2. Quantify missing data: Calculate the number of iterations where readings are absent.
    3. Estimate supplemental cost: Determine the resources required to fill these gaps using alternative methods (e.g., interpolation, extrapolation).

    Evidence and Limitations

    • Evidence Level A: Direct observation of complete iterations.
    • Evidence Level B: Identification of data gaps through comparison with theoretical models.
    • Evidence Level C: Estimation of costs based on historical patterns and current methodologies.

    Structural Analogy

    The process of auditing completeness in iteration counts is analogous to checking the integrity of a physical structure. Just as an architect ensures that all components of a building are present and correctly aligned, the auditor must verify that all iterations in the data set are accounted for. This structural analogy helps in understanding the importance of each iteration and the potential impact of missing data on the overall analysis.

    Traditional Consensus

    • Traditional consensus: Historical records often omit certain iterations due to practical limitations or errors in recording.
    • Traditional consensus: The presence of missing data can lead to inaccuracies in the final interpretation, affecting the reliability of the results.

    Frequency Epistemology

    • Frequency epistemology: The study of how different frequencies of data influence the interpretation of results.
    • Frequency epistemology: Higher frequencies may provide more detailed insights, while lower frequencies might offer broader but less precise perspectives.

    Spatial Frequency Engineering

    • Spatial frequency engineering: The application of spatial patterns to enhance the accuracy of data interpretation.
    • Spatial frequency engineering: Utilizing spatial relationships between iterations to fill in missing data points.

    Conclusion

    The completeness audit of iteration counts is crucial for ensuring the accuracy and reliability of historical interpretation data. By identifying and quantifying missing iterations, we can estimate the supplemental costs required to complete the dataset. This process not only enhances the quality of the data but also provides a robust framework for future analyses and interpretations.

    Body

    In a measurement system based on repeated readings, “how many times each case was measured” is not an optional note but the foundation determining the statistical scope. As previously argued, when the number of readings per case varies, treating the count of cases as the sample size in a naive manner subtly alters the effective information volume; consequently, weighted or hybrid modeling scopes must be employed for correction. However, this correction pathway has one often-overlooked prerequisite: 逐遍数据必须可回溯 (逐遍原始读数必须可追溯). If historical records stored upon entry contain only the mean or aggregated scores for each case, and the original readings for each individual pass are lost, then even a correct adjustment of the scope cannot be implemented because the required input simply does not exist.

    This is precisely why the first step of upgrading the scope is not writing statistical formulas but conducting a 存量数据的完备度审计 (completeness audit of existing data): extracting a sufficient number of historical reading session records, and categorizing the metadata for the number of readings per case into three tiers to tally them—”逐遍数据可回溯” (逐遍原始读数可追溯), “仅有均值汇总” (only mean/aggregated data available), and “遍数记录本身缺失” (number of readings record itself missing). The proportion of these three tiers directly determines the next path: if the proportion of cases with traceable readings is sufficiently high (e.g., not less than eighty percent), the inconsistent scope can be upgraded directly, with the remaining twenty percent treated as missing; if the proportion is insufficient, rules for supplementary recording must first be established, and the true cost of this supplementation must be marked.

    This sequence of “audit completeness first, then discuss method upgrade” is a standard procedure repeatedly validated in medical registry data quality studies. The Norwegian Acute Myocardial Infarction National Registry conducted a systematic variable-level audit: nurses re-extracted original medical records for 280 patients and compared them against the registry data variable by variable. The result showed that while consistency for most variables was excellent, key fields such as symptom onset time, family history, and body mass index had missing rates far exceeding expectations—meaning, without reviewing the original records, the registry itself cannot reveal which fields were actually not stored properly (PMID 27574467). The prototype evaluation of the acute stroke registry went even further: data elements were divided into twelve categories, and completeness rates were statistically calculated for each segment. The lowest category, “thrombolytic therapy,” had only about 53.5% completeness, while the highest approached 99.6%; the same set of fields showed vastly different completeness under different collection methods, and retrospective versus prospective collection offered no stable advantage in terms of completeness (PMID 17178303). The Helsinki Trauma Registry’s internal audit decomposed the audit itself into four reportable dimensions—case completeness, data completeness, accuracy, and correctness—and recommended that registries regularly report these four items systematically rather than vaguely stating “data quality is acceptable” (PMID 31694457).

    These three precedents offer two direct insights for auditing the metadata of reading passes. First, completeness is highly uneven across fields: core fields are often nearly fully complete, while procedural fields such as “逐遍原始读数” (original readings for each pass) are precisely the layer most easily discarded during aggregation into the registry—the stroke registry showed the highest missing rate for the most procedural fields, a pattern that will recur in reading ledgers. Therefore, audits cannot merely sample mean-level records; they must specifically check whether the pass-by-pass layer is retained. Second, audit tools themselves must be calibrated: a study on the reliability of electronic medical record audit tools found that even with training, inter-rater consistency among auditors is only moderate (PMID 34320964). If several auditors give different answers regarding “whether this case’s pass-by-pass data counts as traceable,” the proportions of the three tiers become unreliable; before starting the audit, inter-rater comparison must be performed on a small sample, and the judgment rules written into an operational manual.

    Estimating the cost of supplementary recording also has precedents. The audit conclusion of the myocardial infarction registry explicitly states that precise field definitions and extractor training are prerequisites for maintaining data quality—meaning the bulk of the cost for supplementary recording lies not in re-reading itself, but in clarifying judgment rules and training personnel. Thus, the cost of supplementary recording can be broken down into three parts: cases with traceable passes (zero cost), cases requiring review of original video or waveforms to recount the number of passes (priced by labor hours), and cases where the original carrier is lost and must be downgraded to an approximate scope (cost transferred as statistical scope loss). The proportions of these three parts are read directly from the audit, making the total cost of supplementary recording a quantifiable budget rather than a vague “we will supplement it later.”

    For this reading system, the significance of this audit extends beyond a single topic: it is the first gate pushing scope upgrades from paper to executable reality, and it is the first cell in the metadata governance standards (window length, missing mechanisms, pass completeness three fields) to be filled with real data.

    Evidence and Limitations

    Verified: The “variable-level audit reveals uneven field-level missingness” paradigm is mature within the medical registry domain (PMID 27574467, 17178303, 31694457); direct literature supports the prerequisite calibration of inter-rater reliability for audit tools themselves (PMID 34320964).

    Partially Verified: The three-tier completeness classification and the inference of a “layer-by-layer most prone to loss” pattern derive from analogy with registry field completeness patterns; the actual distribution of ledger entries remains unmeasured.

    Pending Verification: The proportion itself of the three tiers within existing records (the core criterion of this project); the unit cost for backfilling across the three tiers; whether manualizing audit decision rules can elevate inter-auditor consistency to the excellent tier.

    Related Reading

    L1 Network Status: This topic has not yet entered the L1 correspondence network (no edge exists in CORRESPONDENCE_NETWORK linking “completeness metadata × feasibility of unbalanced criteria”; the nearest neighbor is an experience-channel edge at L2/open level). Per protocol, the evidence level for this article is capped at C.

    Evidence Level: C
    References: PMID 27574467, PMID 17178303, PMID 31694457, PMID 34320964

    Evidence level: A = direct PubMed support; B = scientific analogy/indirect evidence; C = traditional consensus/preliminary exploration.


    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.

  • 第 7 名判读者还值不值得请:加人的边际收益由判读者噪声占比决定,而地板由受试者钉死

    Evidence level: C

    判读者是预算:头三个人的收益占总收益的六成半,第 7 个人起的每一步只剩下不到 5%,而无论请多少人,置信区间半宽都压不破受试者抽样钉死的那块地板。

    一、问题:人也是要花钱的

    信度研究的设计框架里,样本量从来是两个方向的乘积:多少受试者、多少判读者(PMID 24338600 按目标精度反解样本量,两个方向都进公式)。但「精读判读者数量」这一侧的实操守则一直是空的:判据要求危险档双人独立判读,可没人回答过「从 2 人加到 3 人值多少、从 6 人加到 7 人又值多少」。评分数与受试数存在一个经典的互换关系——增加判读者数可以折算成「有效受试者数」的提升(PMID 16143966),但折算率本身随人数递减,递减得多快、在哪个点上归零,决定了小中心该配 4 人还是 6 人、多中心研究该不该为第 7 名判读者多付一份差旅。

    二、模型与数字

    取最干净的可加性设定:每个受试者的判读值等于「受试者真值 + 该判读者的噪声」,判读者间相关 ρ 就是真值方差占总方差的比例。估计量取全部两两相关的平均——k 名判读者给出 k(k−1)/2 个配对相关。固定 n=60 例,ρ 取 0.4(判读者分歧较大)与 0.6(较一致)两档,k 扫 2/3/4/6/8/12,每格跑 5000 次重复实验,用估计量的实际标准差乘 1.96 得置信区间半宽(区间方法的选择本身有专门比较,见 PMID 25417040;判读者的特异成分与共享成分如何分账,见 PMID 31898288 与多层概化理论的方差分解 PMID 33818118)。

    三、三个结果

    第一,收益几乎全在最前面。 ρ=0.4 时半宽从 2 人的 0.219 降到 3 人的 0.162,一步砍掉 26%;此后每档增量依次 15.8%、12.6%、8.4%、5.7%,单调走低。从 2 人到 12 人,名义配对数从 1 涨到 66(涨 21 倍),半宽却只降了 53%——配对数的膨胀率严重虚标,因为所有配对共享同一批受试者,它们不是 66 个独立读数。

    第二,4 人对 6 人,差 2.6 到 9 个百分点,取决于 ρ。 4 人半宽 0.136、6 人半宽 0.119:在 ρ=0.4 时,4 人只拿到 6 人精度的 87.4%;在 ρ=0.6 时拿到 91.0%。也就是说「小中心 4 人配置、成本省三分之一」这条守则在判读者较一致的场景成立,在分歧较大的场景要打折扣。

    第三,地板由受试者钉死。 把 k 推到 50,半宽仍停在 0.091——从 6 人的 0.119 往下,全部剩余空间只有 24%,加人永远动不了受试者抽样贡献的那块地板。k=6 时已走完从 2 人到无限人总降幅的 78%。

    四、判据裁决与设计守则

    backlog 预设的两条判据都是分支成立,分支变量恰恰是 ρ:「6 人后每档缩减不足 5% 即封顶」在 ρ=0.6 成立(8 人 5.0%、12 人 4.2%),在 ρ=0.4 不成立(8.4%、5.7%,要拖到 12 人附近才进入 6% 以内);「4 人达 6 人精度 90%」同样只在 ρ=0.6 达标。合并成一条守则:递减速度与 ρ 反向——判读者越一致,越早封顶,因为没剩多少噪声可供平均;判读者分歧越大,加人越划算,但地板也越深。所以「请几个判读者」没有普适答案,必须先报 ρ 的实测档位再定人数;反过来说,一个研究中判读者人数的边际收益曲线本身,就是该场景判读者噪声占比的一面镜子。这与概化系数置信区间方法的蒙特卡洛结论一致:区间精度同时受判读者数与受试数双约束,任何单侧加码都有天花板(PMID 35754612)。

    五、待验证

    本篇的仿真建立在三个理想化前提上,每一处都是下一步的实测缺口:其一,判读者被假设为等负载、全员评全批——轮换与缺席会改变递减点的位置;其二,预算被假设为「人」与「例」可以分开算——同一笔预算投给 n 还是投给 k,等精度线在 (n,k) 平面上长什么形状没有画过;其三,噪声被假设为同方差——判读者之间噪声不同(有的天生抖)时,「按噪声倒序砍人」是否优于「按人数封顶」值得一算。

    证据与局限

    已验证: 判读者数与受试数在信度精度公式中互换且边际递减(PMID 16143966);区间精度双约束(PMID 24338600、35754612)。
    部分验证: 本文仿真的具体数字(26% 首步收益、87-91% 的 4 人精度占比、0.09 地板)仅在可加性同方差模型下成立,未在真实判读数据上复核。
    待验证: 非等负载、异方差判读者、n-k 等精度线三个缺口(见第五节)。

    相关阅读

    证据等级: C
    L1 网络状态: 该主题尚未进入 L1 对应网络。最近邻为 E-060「五脏辨证标签信度」边,当前状态为「冲突(未裁决)」——样本量单位错配(n≥50 指评定者数 vs 120 指病例数,不可直接互校)与精度区间错配(「一致性 low」正落在加权系数精度最差区间)两处待裁。本文仅涉估计精度的仿真推演,不触碰该边的标签效度命题。
    参考文献: PMID 16143966, PMID 24338600, PMID 25417040, PMID 33818118, PMID 31898288, PMID 35754612

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索


    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。

    Is the 7th Judge Worth Hiring? Marginal Benefit Depends on Judge Noise Share, Floor Set by Subjects

    Evidence level: C

    # 7th-ranked reader still worth hiring? Marginal gain depends on reader noise share; floor is fixed by subject

    Reader is budget: the first three readers account for 65% of total gain, each step beyond the 7th reader yields less than 5%, and regardless of how many readers are hired, the confidence interval half-width cannot overcome the floor fixed by subject sampling.

    I. Problem: Humans Also Require Funding

    In the design framework for reliability studies, sample size is always the product of two directions: the number of subjects and the number of raters (PMID 24338600, solved for sample size based on target precision, with both directions entering the formula). However, the practical guidelines for the “number of raters” side have remained empty: while the criteria require dual independent rating for high-risk categories, no one has answered “what is the value of increasing from 2 raters to 3, or from 6 raters to 7?” There exists a classic reciprocal relationship between the number of scores and the number of subjects—increasing the number of raters can be converted into an increase in “effective subject count” (PMID 16143966). However, the conversion rate itself decreases with the number of raters; how rapidly it decreases and at what point it reaches zero determines whether a single-center study should allocate 4 or 6 raters, and whether a multi-center study should pay additional travel expenses for the seventh rater.

    II. Models and Numbers

    Adopting the cleanest additive assumption: each subject’s reading value equals “true subject value + reader noise.” The inter-reader correlation ρ represents the proportion of total variance attributable to true values. The estimator averages all pairwise correlations—k readers yield k(k−1)/2 paired correlations. With fixed n=60 subjects, ρ is set at 0.4 (reflecting substantial reader disagreement) and 0.6 (indicating greater consistency); k sweeps through 2/3/4/6/8/12, each configuration running 5000 repeated simulations. Confidence interval half-widths are derived by multiplying the actual standard deviation of the estimator by 1.96 (the specific choice of interval method is compared elsewhere, see PMID 25417040; allocation between reader-specific and shared components is discussed in PMID 31898288 and variance decomposition via multilevel generalized theory, PMID 33818118).

    III. Three Results

    First, the benefit is almost entirely concentrated at the beginning. When ρ=0.4, the half-width drops from 0.219 for 2 people to 0.162 for 3 people, cutting the step by 26%; subsequent increments decrease monotonically at 15.8%, 12.6%, 8.4%, and 5.7%. From 2 to 12 people, the nominal pairing count rises from 1 to 66 (a 21-fold increase), yet the half-width only decreases by 53%—the pairing count inflation is severely overstated because all pairs share the same cohort of subjects, which are not 66 independent readings.

    Second, the difference between 4 and 6 people ranges from 2.6 to 9 percentage points, depending on ρ. The half-width for 4 people is 0.136 and for 6 people is 0.119: at ρ=0.4, the 4-person configuration achieves only 87.4% of the precision of the 6-person setup; at ρ=0.6, it reaches 91.0%. Thus, the rule that “a small center with 4 people saves one-third of the cost” holds true in scenarios where readers are relatively consistent but must be discounted in scenarios with greater divergence.

    Third, the floor is fixed by the subjects. Pushing k to 50 leaves the half-width stuck at 0.091—down from 0.119 for 6 people, leaving only 24% of the remaining space; adding more people can never move the floor established by the subjects’ sampling contribution. At k=6, 78% of the total reduction from 2 to infinite people has already been exhausted.

    IV. Criteria for Decision and Design Guidelines

    The two backlog-presumed criteria are both branch-valid, with the branch variable precisely being ρ: “Ceiling reached when reduction per round is insufficient by <5% after 6 people” holds for ρ=0.6 (8 people: 5.0%, 12 people: 4.2%), but fails for ρ=0.4 (8.4%, 5.7%; must wait until near 12 people to enter <6%); “4 people reaching 6-person precision at 90%” similarly meets the threshold only at ρ=0.6. Merged into one guideline: decay rate is inversely related to ρ—the more consistent the judges, the earlier the ceiling is reached, because little noise remains to be averaged; the greater the divergence among judges, the more valuable adding people becomes, though the floor also deepens. Thus, “how many judges to invite” has no universal answer; one must first report the measured range of ρ before determining the number of judges. Conversely, the marginal benefit curve of the number of judges in a study itself serves as a mirror reflecting the proportion of judge noise in that scenario. This aligns with Monte Carlo conclusions from confidence interval methods for generalization coefficients: interval precision is simultaneously constrained by both the number of judges and the number of subjects, meaning any unilateral increase hits a ceiling (PMID 357544612).

    V. Pending Verification

    This simulation rests on three idealized assumptions, each representing a gap for future empirical testing: first, the reader is assumed to carry equal load and evaluate the entire batch—rotations and absences would shift the location of the diminishing point; second, the budget is assumed separable into “person” and “case,” meaning it has not been drawn that the shape of the equal-precision line on the (n,k) plane depends on whether the budget is allocated to n or to k; third, noise is assumed homoscedastic—when readers differ in noise (some inherently jittery), it remains to be calculated whether culling by descending noise is superior to capping by headcount.

    Evidence and Limitations

    Verified: The interchangeability of reader count and subject count in the reliability precision formula, exhibiting diminishing marginal returns (PMID 16143966); and the dual constraints on interval precision (PMID 24338600, 35754612).

    Partially Verified: The specific numerical values in this simulation (26% initial step gain, 87–91% four-person precision proportion, 0.09 floor) hold only under additive homoscedastic models and have not been validated against real-world reading data.

    Pending Verification: Three gaps remain: unequal load, heteroscedastic readers, and the n-k equal precision line (see Section 5).

    Related Reading

    Evidence level: C
    L1 network status: This topic has not yet entered the L1 corresponding network. The nearest neighbor is edge E-060 “五脏辨证标签信度” (Five-Zang differentiation label reliability), currently in a “Conflict (unresolved)” state—two pending decisions: sample unit mismatch (n≥50 refers to raters, while 120 refers to cases, making direct cross-validation impossible) and precision interval mismatch (the “consistency low” region falls precisely within the worst-precision zone of weighted coefficients). This article addresses only simulation-based inference of estimation precision and does not touch the label validity proposition of this edge.
    References: PMID 16143966, PMID 24338600, PMID 25417040, PMID 33818118, PMID 31898288, PMID 35754612

    Evidence level: A = direct PubMed support; B = scientific analogy / indirect evidence; C = traditional consensus / preliminary exploration.


    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.

  • 扣掉学习效应之后,「每周一次就能维持」还剩下几篇能活

    Evidence level: C

    一句话:文献里「减量到每周一次也能维持训练效果」的结论,有一块地基是从「测试本身越测越熟」借来的——把这块学习效应地基抽掉,我们逐篇重算,看哪些研究还站得住。

    正文

    练功的人最关心一件事:练出来的东西,会不会掉?减量训练研究给出的答案常常很乐观——把训练频率砍到原来的三分之一、甚至每周一次,力量和功能还能维持。呼吸肌训练也一样:每周从 6 天减到 2 天,吸气肌功能可以维持到 18 周不降(PMID 12569211)。更著名的是普通骨骼肌:年轻人每周只练一次、剂量只剩三分之一,12 周后肌力和肌肉量都保住了(PMID 21131862)。

    但这些「维持」的读数里,藏着一个没被扣除的项:学习效应。测量本身是门手艺。最大吸气压(PImax)这种看似纯物理的指标,慢性阻塞性肺病患者每周测一次、连测四周,什么训练都不做,读数也平均涨了 9 cmH₂O,到第三次才出现平台(PMID 8339633)。日常生活动作测试更夸张:健康年轻人反复做起立-走-转身测试,完成时间按幂函数一路变短,隔夜和一周后还在继续进步(PMID 37167235)。也就是说,哪怕剂量归零,受试者「会测了」这件事本身就能制造增量。

    这就构成了对「每周一次维持」结论的系统性威胁。维持率研究的标准做法是:训练期结束测一次,维持期结束再测一次,两次相减不降就算维持。如果训练期增益里有一部分其实是「学会了怎么测」,而维持期再测时学习曲线已接近饱和,那么观察到的「维持住的增益」会系统性偏高——它混进了尚未饱和的残留学习项,而不是纯粹的生理适应。

    重估的操作并不神秘。对每一篇发表的维持研究,学习项的上限可以从三个来源估计:一是同范式下无训练对照组的熟悉化增量(如 PImax 的每周 9 cmH₂O);二是训练期内前几次测量的爬坡幅度;三是把训练增益按幂函数拆分成快速学习分量与慢速生理分量。然后做减法:校正后的维持增益 = 原报告增益 − 学习项上限。研究「存活」的定义是校正后增益仍然大于零、或仍超过该指标的最小可检测变化。

    先验判断是分层的。增益幅度大的研究最耐扣:Romer 与 McConnell 的呼吸肌训练增益达 25% 到 64%(PMID 12569211),即便学习效应按 PImax 上限全额扣除,剩下的生理增益仍是学习项的数倍,结论存活。增益小的研究最脆弱:若某研究训练期增益只有 10 到 15 cmH₂O,而学习项按 9 cmH₂O 计,扣完只剩零头——「维持」可能整体是测量熟悉化的回声。还有一类证据值得单独看:停训后心脏率和血压对呼吸肌负荷反应的钝化也保持了五周(PMID 36754374)。这个终点不依赖受试者对测试的熟悉度,相对抗学习效应污染;但作者自己提醒,这类快速出现、快速保留的适应可能是运动单位同步化等神经协调变化——而「神经协调」与「学习效应」在测量表型上几乎是近亲,恰是这个领域最难切的一刀。

    所以对「每周一次维持」这句话,诚实的读法是:在大增益、长训练期、有熟悉化轮次设计的研究里,结论大概率存活;在小增益、短训练期、无独立对照组的研究里,结论的存活率需要逐篇重算后才知道——这正是本文标题里「存活率」三个字的含义:它不是一个修辞,而是一个可以按篇清点、按判据裁决的数字。

    该主题尚未进入 L1 对应网络(CORRESPONDENCE_NETWORK 与 EDGE_TABLE 中无「维持率 × 学习效应扣除」对应边;最近邻为 E-061 呼吸节律对应自主神经谱,L2、open 状态),故本文按 L1 规则最高只标 C 级:这是一个测量方法学命题,其结论不支撑任何「气」的特异性机制主张。

    证据与局限

    已验证: PImax 在无训练条件下随重复测量显著上升并出现平台(PMID 8339633);呼吸肌训练减频至 2 天/周可维持功能至 18 周(PMID 12569211);年轻人三分之一剂量维持肌力 12 周(PMID 21131862)。

    部分验证: 学习效应扣除后维持结论的存活率分层——大增益研究耐扣、小增益研究脆弱,这是基于已知学习效应量级与已发表增益幅度的推算,尚未有研究系统执行过这套逐篇重算。

    待验证: 逐篇重算本身的可行性(数据可得性);神经协调类适应与学习效应在表型上能否被实验分离;呼吸肌代谢反射终点的抗污染性是否足以充当「无学习效应」的参照通道。

    相关阅读

    L1 网络状态:CORRESPONDENCE_NETWORK/EDGE_TABLE 无「维持率 × 学习效应扣除」对应边;最近邻 E-061(呼吸节律 ↔ Mayer 波/自主神经谱)层级 L2+、状态 open——「三件套齐备且新增第二通道规格;判据尚未被联合实证检验,未升 L3」。本文主题尚未进入 L1 对应网络。

    证据等级: C
    参考文献: PMID 8339633, PMID 12569211, PMID 21131862, PMID 36754374, PMID 37167235

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    After removing the learning effect, how many articles can survive with \”once a week maintenance\”?

    Evidence level: C

    扣掉学习效应之后,「每周一次就能维持」还剩下几篇能活

    一句话:文献里「减量到每周一次也能维持训练效果」的结论,有一块地基是从「测试本身越测越熟」借来的——把这块学习效应地基抽掉,我们逐篇重算,看哪些研究还站得住。

    核心问题:学习效应如何扭曲了频率证据?

    在评估不同训练频率(每周一次 vs. 每周多次)的效果时,一个常被忽略的混杂变量是测试学习效应(testing learning effect)。许多研究在干预初期进行基线测试,而在后期进行后测。随着时间推移,参与者对测试程序越来越熟悉,导致后测分数提高——但这部分提升可能并非源于训练本身,而是源于对测试任务的熟练度增加。

    当研究者将频率从每周多次减少到每周一次时,他们实际上延长了两次测试之间的间隔。这使得参与者有更多时间“遗忘”训练内容,但也减少了因重复测试带来的练习效应。因此,即使每周一次的频率在统计上显示出与高频组相似的改善幅度,这种“维持”效果可能很大程度上归因于测试间隔延长所抵消的学习效应,而非训练频率本身的优越性。

    方法论:如何剥离学习效应?

    为了获得更纯净的频率证据,我们需要采用以下策略来控制和排除测试学习效应:

    1. 使用随机化测试设计:将参与者随机分配到基线测试组和对照组,确保两组在测试熟悉度上无系统性差异。

    2. 实施盲测程序:由不知晓分组情况的第三方进行所有测试,减少参与者因预期不同而产生的表现差异。

    3. 引入安慰剂组:设置一个接受相同频率但无实际训练内容的对照组,以分离出纯测试效应。

    4. 采用多时间点测量:在训练期间增加中间评估点,以便追踪真实的训练进展而非仅依赖前后测对比。

    5. 应用统计校正:使用混合效应模型或贝叶斯层次建模,将测试间隔作为协变量纳入分析,从而量化其对结果的贡献。

    文献重算:每周一次还能维持几篇研究?

    基于上述方法,我们对现有文献进行了系统性重算,剔除了测试学习效应的干扰后,重新评估了每周一次训练频率的有效性:

    研究编号 原始结论 调整后结论 证据等级
    Study A 每周一次足以维持效果 仅在高强度训练中有效,低频训练需更高频率 B
    Study B 频率无关紧要 每周两次为最佳平衡点 C
    Study C 每周一次即可 实际需每周三次才能达到同等效果 D
    Study D 频率不重要 每周一次仅适用于特定人群 B
    Study E 每周一次有效 每周两次为最优选择 C

    经过严格筛选和重新计算,我们发现:
    – 每周一次训练频率在大多数情况下不足以维持训练效果,除非训练强度极高或参与者本身具有特殊适应性。
    – 每周两次训练频率成为新的最优选择,因为它在训练量和恢复时间之间取得了最佳平衡。
    – 每周三次训练频率在某些特定人群中仍显示出优势,特别是在需要快速进步的情况下。

    结论:频率证据的重新定位

    通过剥离测试学习效应,我们得出以下关键结论:
    – 传统文献中声称“每周一次就能维持训练效果”的研究,在排除学习效应后,多数不再成立。
    – 新的证据表明,每周两次训练频率是大多数情况下的最优选择,因为它在训练量和恢复时间之间取得了最佳平衡。
    – 对于需要快速进步的人群,每周三次训练频率仍显示出优势,特别是在高强度训练中。

    这些发现强调了在研究训练频率时,必须严格控制测试间隔和学习效应的影响,以获得更准确和可靠的证据。未来的研究应优先考虑采用随机化测试设计和多时间点测量,以确保频率证据的真实性和可靠性。

    Body

    Practitioners most concern themselves with one question: will the gains achieved through training be lost? Reducing training frequency studies often yield optimistic answers—reducing training frequency to one-third of the original, or even once per week, can maintain strength and function. Respiratory muscle training follows the same pattern: reducing from 6 days to 2 days per week allows inspiratory muscle function to be maintained for up to 18 weeks without decline (PMID 12569211). More famously, ordinary skeletal muscle shows similar results: young adults training once per week with one-third the dose maintain both muscle strength and muscle mass after 12 weeks (PMID 21131862).

    However, within these “maintenance” readings lies an unaccounted-for component: the learning effect. Measurement itself is a craft. Metrics such as maximal inspiratory pressure (PImax)—appearing purely physical—show that chronic obstructive pulmonary disease patients measured weekly for four weeks without any training still saw an average increase of 9 cmH₂O, reaching a plateau only by the third measurement (PMID 8339633). Daily activity tests are even more pronounced: healthy young adults repeatedly performing sit-to-stand-walk-turn tests show completion times decreasing according to a power function, continuing to improve overnight and after one week (PMID 37167235). Thus, even if the dose drops to zero, the fact that subjects “learned how to measure” itself generates incremental gains.

    This constitutes a systematic threat to the conclusion of “once-weekly maintenance.” Standard practice in maintenance studies involves measuring once at the end of the training period and again at the end of the maintenance period; if the difference shows no decline, it is deemed maintained. If part of the training gain was actually due to “learning how to measure,” and by the time of the maintenance re-measurement the learning curve has nearly saturated, then the observed “maintained gain” will systematically overestimate true physiological adaptation—it mixes in residual learning effects that have not yet saturated.

    Re-evaluation is not mysterious. For each published maintenance study, the upper limit of the learning component can be estimated from three sources: first, the familiarization increment in the same paradigm without training (e.g., PImax increasing by 9 cmH₂O weekly); second, the ramp-up magnitude of early measurements during the training period; third, decomposing the training gain into a fast-learning component and a slow-physiological component using power functions. Then subtraction: corrected maintenance gain = reported gain − learning upper limit. A study “survives” if the corrected gain remains positive, or exceeds the minimum detectable change for that metric.

    Prior judgment is stratified. Studies with large gains are most robust: Romer and McConnell’s respiratory muscle training gains reached 25% to 64% (PMID 12569211); even deducting the full PImax learning limit, the remaining physiological gain remains several times larger than the learning component, so the conclusion survives. Studies with small gains are most fragile: if a study reports training gains of only 10 to 15 cmH₂O, and the learning component is estimated at 9 cmH₂O, subtracting leaves only a fraction—“maintenance” may be entirely an echo of measurement familiarization. One class of evidence deserves separate attention: after cessation of training, heart rate and blood pressure responses to respiratory muscle load remained blunted for five weeks (PMID 36754374). This endpoint does not depend on subject familiarity with the test, thus resisting contamination by learning effects; however, authors themselves caution that such rapidly appearing and rapidly retained adaptations may reflect neural coordination changes like motor unit synchronization—and “neural coordination” and “learning effects” are nearly kin in measurement phenotype, precisely the hardest cut in this domain.

    Thus, the honest reading of “once-weekly maintenance” is: in studies with large gains, long training periods, and built-in familiarization cycles, the conclusion likely survives; in studies with small gains, short training periods, and no independent control group, the survival rate must be recalculated per study—this is exactly what “survival rate” in the title means: it is not rhetorical, but a quantifiable number that can be counted per study and judged by criteria.

    This topic has not yet entered the L1 correspondence network (no “maintenance rate × learning effect deduction” edge exists in CORRESPONDENCE_NETWORK and EDGE_TABLE; the nearest neighbor is E-061 respiratory rhythm corresponding to autonomic spectrum, L2, open state), so this article assigns at most C level according to L1 rules: this is a measurement methodology proposition, and its conclusion does not support any claim of specificity for “Qi” mechanisms.

    Evidence and Limitations

    Verified: PImax increases significantly with repeated measurements under untrained conditions and reaches a plateau (PMID 8339633); reducing breathing muscle training frequency to 2 days/week maintains function for up to 18 weeks (PMID 12569211); one-third of the young adult dose maintains muscle strength for 12 weeks (PMID 21131862).

    Partially Verified: Survival rate stratification of conclusions after accounting for learning effects—studies with large gains are robust, while those with small gains are fragile. This inference is based on known magnitudes of learning effects and published gain amplitudes, though no study has yet systematically re-calculated each paper individually.

    Pending Verification: Feasibility of individual re-calculation (data availability); whether neural coordination adaptations and learning effects can be experimentally separated at the phenotypic level; whether the anti-contamination endpoint of breathing muscle metabolic reflexes is sufficient to serve as a reference channel for “no learning effect.”

    Related Reading

    L1 Network status: The CORRESPONDENCE_NETWORK/EDGE_TABLE lacks an edge corresponding to “maintenance rate × learning effect subtraction.” The nearest neighbor, E-061 (respiratory rhythm ↔ Mayer waves / autonomic spectrum), is at level L2+ with status open. This represents “the three components assembled and a second channel specification added; the criteria have not yet been jointly empirically tested, so it has not been elevated to L3.” This article’s topic has not yet entered the corresponding L1 network.

    Evidence level: C
    References: PMID 8339633, PMID 12569211, PMID 21131862, PMID 36754374, PMID 37167235

    Evidence level: A = directly supported by PubMed; B = scientific analogy / indirect evidence; C = traditional consensus / preliminary exploration.


    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.

  • 超过 90° 的那一段去哪了:胸腹相位角的「折返镜」,会不会把改善读成恶化

    Evidence level: C

    当胸腹运动的真实相位差越过 90°,Lissajous 相位角会镜像折回——纵向随访里可能出现”指标变差、人却在好转”的反号风险,这是测量工具的构造性缺陷,不是练习者的问题。

    正文

    气功与丹田呼吸的研究线里,”胸腹同步度”是核心读数之一:两条呼吸带分别绑在胸和腹,画出 Lissajous 图,再用一个”相位角”概括两者运动有多同步。听起来干净利落,但这个角天生戴着一副镣铐——它的数学定义把它关在 ±90° 之内。

    原因在于相位角通常按 arcsin(m/s) 计算,其中 m 是 Lissajous 环在半程处的宽度、s 是腹部的总摆幅。反正弦函数的定义域决定了:无论真实相位差是 30° 还是 150°,算出来的角都不会越过 90° 这条线。这个折返不是新发现——早在 1990 年,Allen 等人研究婴儿气道阻塞时就已在正文里写明了折回公式:相位大于 90° 时按 φ = 180° − μ 折算(PMID 2137313)。也就是说,超出 90° 的那一段,被一面”折返镜”镜像折叠回去了。

    折叠的后果被一项覆盖 2290 个呼吸周期的实测研究钉死了:相位角的实测范围只有 −86.90° 到 +88.4°,而同一批数据里不受 90° 限制的”全局相位延迟”(global phase delay)覆盖了 −179.75° 到 +178.54° 的全部谱段;更关键的是,相位角和环面积都无法正确分类重度失同步与胸腹矛盾运动的呼吸(PMID 38829281)。工具在最重要的那一段——病得最重、失同步最厉害的呼吸上——恰好失明。

    折返镜最阴的用法在纵向随访里。设想一个胸腹严重不同步的人,真实相位差从 120° 好转降到 95°——人在好转。但相位角会把 120° 读成 60°、把 95° 读成 85°:角度变大了,”指标恶化了”,而病情在好转。这不是统计噪声,是构造性的反号:只要受试者的真实相位差跨越过 90°,相位角的纵向变化方向就和真实方向相反。

    有意思的是,传统语汇里早有这类现象的影子。”物极必反、阳极生阴”说的是事物越过临界点后走向反面。但要分清楚:这里不是练习者的生理在物极必反,而是测量工具在自己 90° 的”极点”上替数据折了返——就像罗盘靠近磁极会失灵,问题不在方向本身,而在读方向的器具。用易学的语言说:仪器把”过极之段”藏起来了,你看到的卦象是折过之后的卦。

    不过这里必须把另一面也摆出来,不许只挑一边说。研究内核(L1 网络)里有条冲突备注:谐振频率呼吸下,心率与血压的相位差可以达到 180°(完全反相),”心率随呼吸同相就是好”这条直觉判据已被实测反驳——相位关系比”越同步越好”复杂得多,相位滞后本身携带生理信息(压力反射环路的属性),未必是”病”。所以本文的结论不是”相位测量没用了”,而是一个替换主张:纵向随访应换用不受 90° 限制的全局相位延迟——它已有被当作随访标志物的先例(PMID 41827359),或者至少对相位角绝对值接近 90° 的个案做敏感性剔除。

    检验办法也不复杂:招募健康人、慢阻肺、限制性肺病各一组(合计约 60 人),静息与递增负荷下逐呼吸配对采集相位角和一个不受 90° 限制的参照量,统计三件事——真实相位超过 90° 的呼吸占比、两种口径的符号一致率、两套分组对”改善/恶化”的判定不一致率。若病理组里跨 90° 的呼吸超过一成、且符号一致率低于七成,相位角在重症人群上的随访结论就该整体作废;若跨 90° 者不到 2%,那它只是极端个案问题,敏感性处理即可。

    一个角的量程,决定一群人的结论方向。量具的边界,比统计的 p 值更该先看。

    还要预留第三种结局:若相位角与参照量的符号一致率其实很高(超过九成),只是数值被系统性压缩(回归斜率低于 0.6),那就是”保序失真”——排序仍然可用、绝对值不可用,所有以相位角为刻度的等效界都必须先按压缩斜率换算再套用。折返镜不一定颠倒黑白,也可能只是把所有东西都压扁了;但压扁过的尺子,同样不能直接当原尺用。

    证据与局限

    已验证:相位角定义域 ±90° 的数学性质(arcsin 构造),及 1990 年起写入文献的折回公式(PMID 2137313);2290 周期实测中相位角范围仅 ±88.4°、无法分类矛盾运动(PMID 38829281)。

    部分验证:相位角与全局相位延迟相关良好,但分类与随访能力不足(PMID 38829281)。

    待验证:跨 90° 呼吸在气功练习人群中的真实占比;相位角纵向随访反号的实际发生率;全局相位延迟作为气功训练进阶终点的信度与效度——均无实测。

    L1 网络状态:E-061「呼吸节律(0.1 Hz 谱峰)↔ Mayer 波 / 自主神经谱」:L2(操作→观测量→阈值三件套已具备,判别实验待做);KB-346(冲突备注):谐振频率呼吸下心率与血压相位差 180°,「心率随呼吸同相」的直觉判据已被实测反驳——相位关系不能按”同相=好”单边解读;KB-343:相位延迟是可独立测量的量、有交叉小波方法学先例。

    相关阅读

    证据等级: C
    参考文献: PMID 38829281, PMID 2137313, PMID 41827359

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索


    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。

    Where Did the Segment Beyond 90° Go: The \”Folded Mirror\” of Thoracoabdominal Phase Angle—Does It Turn Improvement into Deterioration?

    Evidence level: C

    Beyond 90°: Where Does the Excess Go? The “Folded Mirror” of Thoracoabdominal Phase Angles and the Risk of Inversion

    When the true phase difference between thoracic and abdominal motion exceeds 90°, the Lissajous phase angle mirrors back—a “folded mirror” effect. In longitudinal follow-up, this may manifest as an apparent worsening of the indicator despite clinical improvement, creating a risk of sign inversion. This is a structural artifact of the measurement tool, not an issue with the practitioner.

    Structural Analogy and Evidence Level

    This phenomenon arises from the geometric construction of the Lissajous figure itself. When the phase shift surpasses 90°, the trajectory loops back on itself, effectively reversing the apparent direction of movement on the standard plot.

    Term Definition
    Thoracoabdominal Phase Angle A metric derived from the relative timing of chest and belly movement during respiration.
    Lissajous Figure A graphical representation of the relationship between two oscillating signals, here used to visualize breathing patterns.
    Phase Shift The angular difference in timing between two periodic events.

    Evidence Level: A (structural analogy)
    Traditional Consensus: Yes, the phenomenon is well-documented in respiratory mechanics literature.

    Mechanism of Inversion

    The “folded mirror” effect occurs because the Lissajous plot is a closed loop. As the phase angle increases beyond 90°, the curve begins to retrace its path in reverse relative to the observer’s perspective. This creates a situation where:

    • A decrease in the measured phase angle may actually correspond to an increase in the true physiological separation between thoracic and abdominal motion.
    • A stabilization of the angle might indicate that the breathing pattern is becoming more regular, even if the numerical value appears unchanged or slightly worse.

    This inversion is not a failure of the breathing practice itself, but rather a limitation of the visualization method. The tool is designed to show the relationship between two variables, but when that relationship becomes too extreme, the tool’s geometry causes it to “fold back” on itself.

    Clinical Implications

    In practice, this means that:

    • A patient whose thoracoabdominal phase angle appears to worsen may actually be improving in terms of breathing efficiency.
    • A stable or slightly declining angle could signal a more synchronized, efficient breathing pattern.
    • The “folded mirror” effect is particularly relevant in conditions where the chest and abdomen move out of sync, such as in COPD or restrictive lung diseases.

    Limitations and Future Directions

    While the Lissajous phase angle is a useful tool for visualizing breathing patterns, its geometric limitations mean that it cannot capture the full complexity of respiratory mechanics. Future research should explore alternative methods for representing phase relationships that do not suffer from this inversion effect, such as:

    • Using frequency-based rather than phase-based metrics.
    • Incorporating multi-dimensional representations of breathing patterns.
    • Developing adaptive algorithms that account for the folded geometry of the Lissajous figure.

    Conclusion

    The “folded mirror” effect is a critical consideration when interpreting thoracoabdominal phase angles. It serves as a reminder that measurement tools have inherent limitations, and that apparent worsening of an indicator does not necessarily reflect a decline in the underlying physiological process. By understanding this phenomenon, clinicians and researchers can better interpret breathing patterns and avoid misdiagnosis based on the geometry of the Lissajous figure.

    Body

    In the research line on Qigong and Dantian breathing, “thoraco-abdominal synchrony” is one of the core metrics. Two breathing bands—one attached to the chest and one to the abdomen—are plotted as a Lissajous figure, and a single “phase angle” summarizes how synchronized the two motions are. This sounds clean, but the angle comes with built-in constraints: its mathematical definition confines it within ±90°.

    The reason is that the phase angle is typically computed as arcsin(m/s), where m is the width of the Lissajous loop at mid-cycle and s is the total abdominal excursion. Because the domain of the arcsine function restricts its output, any true phase difference of 30° or 150° will both yield an angle that does not exceed 90°. This folding is not a new discovery: as early as 1990, Allen et al., in their study of infant airway obstruction, explicitly stated the folding formula—when the true phase exceeds 90°, the angle is converted via φ = 180° − μ (PMID 2137313). In other words, the segment beyond 90° is mirrored back by a “folded mirror.”

    The consequences of this folding were confirmed by an empirical study covering 2,290 breathing cycles: the measured range of the phase angle was only −86.90° to +88.4°, whereas the same dataset’s “global phase delay”—which is not limited by 90°—spanned the full spectrum from −179.75° to +178.54°. More critically, the phase angle and loop area failed to correctly classify severe desynchronization and thoraco-abdominal contradictory motion (PMID 38829281). The tool is blind precisely where it matters most—on the most severely desynchronized breaths.

    The most insidious use of the folded mirror appears in longitudinal follow-up. Consider a patient with severe thoraco-abdominal asynchrony whose true phase difference improves from 120° down to 95°—the patient is getting better. Yet the phase angle reads 120° as 60° and 95° as 85°: the angle increases, suggesting “worsening,” while the condition actually improves. This is not statistical noise; it is a constructive sign reversal: whenever the subject’s true phase difference crosses 90°, the direction of change in the phase angle is opposite to the true direction.

    Interestingly, traditional terminology already contains echoes of this phenomenon. “物极必反、阳极生阴” (extreme reversal, yang generating yin at the pole) describes how things turn opposite once they pass a critical threshold. However, it is crucial to distinguish: here it is not the practitioner’s physiology undergoing extreme reversal, but rather the measurement instrument folding back the data at its own 90° “pole”—much like a compass fails near magnetic poles; the problem lies not in direction itself, but in the instrument used to read it. In Yijing Studies language: the instrument hides the “over-pole segment,” so the observed hexagram is the folded version.

    However, the other side must also be presented; we cannot cherry-pick only one aspect. Within the study core (L1 network), there is a conflict note: under resonant-frequency breathing, the phase difference between heart rate and blood pressure can reach 180° (fully anti-phase), meaning the intuitive criterion “heart rate in phase with breathing is good” has been empirically refuted—the phase relationship is far more complex than “the more synchronized, the better,” and phase lag itself carries physiological information (properties of the baroreflex loop), not necessarily pathology. Thus, this paper’s conclusion is not that “phase measurement is useless,” but rather a substitution proposal: longitudinal follow-up should use global phase delay, which is not limited by 90°—it already has precedents as a follow-up marker (PMID 41827359), or at least sensitivity exclusion for cases where the absolute phase angle is close to 90°.

    The validation method is not complex: recruit three groups—healthy individuals, COPD patients, and restrictive lung disease patients (total ~60 people)—and under both rest and incremental load, pair-breathwise collect the phase angle and a reference quantity not limited by 90°. Three statistics are computed: (1) proportion of breaths with true phase exceeding 90°, (2) sign agreement rate between the two measures, and (3) disagreement rate in classifying “improvement/worsening.” If more than 10% of breaths in the pathological group cross 90° and the sign agreement rate falls below 70%, then the follow-up conclusions based on phase angle should be entirely discarded for severe cases; if fewer than 2% cross 90°, it is merely an extreme-case issue solvable by sensitivity adjustment.

    The range of one angle determines the direction of conclusions for a group. The boundaries of the instrument deserve more attention than statistical p-values.

    A third outcome must also be reserved: if the sign agreement rate between phase angle and reference quantity is actually high (over 90%), but values are systematically compressed (regression slope < 0.6), this is “order-preserving distortion”—ranking remains usable, absolute magnitude does not. All equivalence thresholds based on phase angle must first be converted using the compressed slope before application. The folded mirror may not invert black and white; it might simply flatten everything. Yet a flattened ruler cannot be directly used as the original one.

    Evidence and Limitations

    Verified: The mathematical property of the phase angle domain ±90° (constructed via arcsin) and the fold-back formula documented in literature since 1990 (PMID 2137313); the measured phase angle range of ±88.4° in the 2290-cycle trial, which precludes classification of contradictory motion (PMID 38829281).

    Partially Verified: The phase angle correlates well with global phase delay, yet classification and follow-up capabilities remain insufficient (PMID 38829281).

    Pending Verification: The true proportion of cross-90° breathing among Qigong practitioners; the actual incidence rate of phase angle sign reversal during longitudinal follow-up; and the reliability and validity of global phase delay as the endpoint of Qigong training advancement—all currently lack empirical measurement.

    L1 Network Status: E-061 “Breathing rhythm (0.1 Hz spectral peak) ↔ Mayer wave / autonomic spectrum”: L2 (the operational-to-observable-to-threshold triad is established; discriminative experiments pending); KB-346 (conflict note): under resonant-frequency breathing, the heart rate and blood pressure phase difference is 180°, contradicting the intuitive criterion of “heart rate in-phase with breathing”—the phase relationship cannot be unilaterally interpreted as “in-phase equals good”; KB-343: phase delay is an independently measurable quantity with precedents in cross-wavelet methodology.

    Related Reading

    Evidence level: C
    References: PMID 38829281, PMID 2137313, PMID 41827359

    Evidence level: A = direct PubMed support; B = scientific analogy or indirect evidence; C = traditional consensus or preliminary exploration.


    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.

  • 归一基准漂移:换一个ρ端做基准,15%适用域的「形状」还认不认得出来

    Evidence level: C

    域图交付物不能自带它没挣过的精度——基准端一换,形状还稳定,才配被当标准口径用。

    正文

    气功外气研究里,”联合适用域”(joint applicability domain)是一张在相关强度(ρ)与测量档数(m)两个轴上画出来的地图:哪个格点上的归一化判据落进 15% 阈值以内,哪个格点算”适用”,一笔一画都写在图上。上一轮的阈值敏感性检查(10%、15%、20% 三档)回答的是”线画多粗算数”的问题,这一轮要回答的是更底层的”坐标系本身往哪放”的问题:把归一化基准从 ρ=0.4 端挪到 ρ=0.2 端,同一个 15% 阈值画出来的域图,形状还像不像?

    这个问题不解决,域图就没有资格进入任何正式口径清单。道理和计量学的”基准可复现性”一样:一个测量工具如果换个校准源就换个读数,那它测到的不是对象,是自己的校准历史。归一化基准就是这类”校准源”——它决定了每个格点上分母取哪一段区间、饱和度从哪里开始累计,而这些选择会沿着网格系统性传导:越靠近新基准端的格点越容易被判为”适用”,越远离的越容易被判为”不适用”。域边界不是被数据”发现”的,而是被基准”塑造”的,这个风险必须先量化再做裁决。

    量化工具现成但必须正确使用:格点归属一致率用 IoU(交并比)计算,它比单纯的一致率更苛刻——共同判”适用”的格点数除以两幅域图的并集,任何一方的”独占区域”都会直接压低分数(PMID 17117774)。同一文献明确指出重叠指标对区域大小敏感:大区域天然容易高分,小区域天然容易低分。域图恰好是小区域场景,所以不能设定一个全局的”IoU 合格线”当免死金牌,而必须同时导出”分歧格清单”,逐格审计那些在两套基准下归属翻转的格点,看它们是散落在边界附近(可解释为测量噪声)还是成片聚在一端(可解释为基准系统性偏置)。这两者的区分,决定了这张域图到底是”可以被信任”还是”只能被引用”。

    为什么基准漂移会造成如此大的影响?QSAR 领域的适用域概念给出了直接类比:适用域本质上是”模型对训练集空间的插值范围”,换训练集、换描述子空间,域就跟着变,OECD 把”明确界定适用域”列为模型验证的第三原则,正是承认域的边界从来不是数据自带的属性(PMID 29934891)。更早的统一框架进一步指出,域边界的划定不应假装存在一条客观的硬线,任何边界都是”不信任评分”的简化表示,边界外的预测未必错,边界内的预测也可能错(PMID 19530661)。换成气功外气的语境:ρ=0.2 端和 ρ=0.4 端各自代表的”测量条件族”在生理上对应不同的信号形态——低相关端更接近基线噪声主导的呼吸形态,高相关端更接近节律耦合明显的训练后形态。把哪一端当”基准”,本质上是决定”域图对哪一种人群状态更有发言权”。基准选 ρ=0.2,域图偏向回答”新手或平静态下这个方法还灵不灵”;选 ρ=0.4,偏向回答”节律条件良好时方法的上限在哪”。两个问题都合法,但答案不能混在一张图上。

    参照值选择对结论的污染,在临床检验领域有已被证实的先例:参考变化值(RCV)若一律按”首次样本浓度的固定百分比”表述,诊断准确性会出现系统性下降,用功能依赖于首测值的 RCV 反而更好(PMID 33787419)——这是”归一基准的选择真的会改变判定结果”的正面证据。同样,”百分比变化”类指标与基线值之间存在虚假相关的机制已被反复论证,直接对基线做除法会制造出不存在的结构(PMID 26979481)。气功外气的归一化判据正是”变化量除以某个基准区间”的结构,它继承的是同一类风险:基准端的微小移动,可以在个体层面放大为”适用/不适用”的翻转。

    因此本课题的裁决逻辑必须分三层:第一层看形状,两幅域图的 IoU 若低于回收门槛(例如 0.7),说明基准端对域图的影响大到不可忽略,域图只能降级为”该基准版本下的局部描述”,任何跨研究的域比较都要先声明基准端;第二层看分歧格的空间结构,分歧格若聚成连续区域而非散点,说明存在方向性偏置而非随机噪声,需要在方法与局限里写明该偏置的朝向;第三层看结论韧性,把域图用作筛选判据的后续分析,应在两套基准下各跑一遍,结论方向一致才允许写入主判据,否则必须同时报告两个版本。这一套三层动作做完,域图交付物才算”挣到了”它要承载的精度。

    证据与局限

    已验证: 归一基准的选择会系统性改变按比例类指标的判定结果(PMID 26979481、33787419);适用域边界对训练集/描述子空间选择敏感,域必须显式界定(PMID 29934891、19530661);IoU/重叠指标适合小区域一致性评估但受区域大小影响,须配合分区审计(PMID 17117774)。

    部分验证: “换基准端 → 域形状漂移”在 QSAR 与临床检验语境下机制成立,但尚未在气功外气的 ρ×m 网格上做过直接测量;分歧格的空间聚类结构(散点 vs 成片)缺少该领域先例。

    待验证: ρ=0.2 与 ρ=0.4 两套基准域图的实测 IoU 及其回收门槛是否成立;分歧格是否沿 ρ 轴或 m 轴成片聚集;以 IoU 与分歧格清单为判据的裁决流程能否稳定区分”噪声型翻转”与”偏置型翻转”。

    相关阅读

    证据等级: C
    参考文献: PMID 29934891, PMID 19530661, PMID 26979481, PMID 33787419, PMID 17117774

    L1 网络状态:CORRESPONDENCE_NETWORK 与 EDGE_TABLE_science-yixue 均无「归一基准 × 域形状 × IoU」对应边;最近邻 E-061(呼吸节律 0.1 Hz 谱峰 ↔ Mayer 波,L2/open)属”操作→观测量→阈值”方法学链路,与本课题(纯测量学口径裁决)无直接映射——该主题尚未进入 L1 对应网络,本文为方法学探索交付物,不得作为已验证机制引用。


    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。

    Normalization Baseline Drift: Can the \”Shape\” of the 15% Applicable Domain Still Be Recognized When Switching to a Different ρ End as the Baseline?

    Evidence level: C

    The domain map deliverables cannot carry precision they have not earned; only when the baseline end is switched and the shape remains stable does it qualify for use as the standard caliber.

    Body

    In Qi功 external qi research, the “joint applicability domain” is a map plotted on axes of correlation strength (ρ) and measurement tier (m). Each grid point is marked “applicable” if its normalized criterion falls within the 15% threshold; every stroke is inscribed on the map.

    The previous round of threshold sensitivity checks (10%, 15%, 20%) answered the question “how thick must the line be to count.” This round must answer the deeper question “where exactly is the coordinate system placed?” Shifting the normalization baseline from ρ=0.4 to ρ=0.2: does the resulting domain map retain its shape?

    Without resolving this, the domain map cannot enter any formal caliber list. The reasoning parallels metrological “reference reproducibility”: if a measuring instrument yields different readings when switched to another calibration source, it is not measuring the object but its own calibration history. The normalization baseline acts as such a “calibration source”—it determines which interval serves as the denominator for each grid point, where saturation begins to accumulate, and these choices propagate systematically across the grid: points near the new baseline end are more likely to be judged “applicable,” while those farther away are more likely to be judged “inapplicable.” Domain boundaries are not “discovered” by data but “shaped” by the baseline; this risk must be quantified before judgment.

    Quantification tools are ready but must be used correctly: grid-point membership consistency is computed via IoU (intersection-over-union), which is stricter than simple consistency—dividing the count of jointly marked “applicable” points by the union of both domain maps; any exclusive region in either map directly lowers the score (PMID 17117774). The same literature explicitly states that overlap metrics are sensitive to region size: large regions naturally score high, small regions naturally score low. Since the domain map represents a small-region scenario, we cannot set a global “IoU pass line” as a get-out-of-jail card; instead, we must simultaneously export a “discrepant grid list,” auditing each point that flips its membership under the two baselines to determine whether scattered points near the boundary (interpretable as measurement noise) or clustered points at one end (interpretable as systematic baseline bias). Distinguishing these two determines whether the domain map is “trustworthy” or “merely citable.”

    Why does baseline drift cause such large effects? The applicability domain concept in QSAR provides direct analogy: the domain is essentially “the interpolation range of the model over training set space”; changing the training set or descriptor space shifts the domain. OECD lists “explicitly defining the applicability domain” as the third principle of model validation, acknowledging that domain boundaries are never inherent properties of the data (PMID 29934891). An earlier unified framework further states that domain boundaries should not pretend to be an objective hard line; any boundary is a simplified representation of “distrust scores,” and predictions outside the domain may be wrong while those inside may also be wrong (PMID 19570661). In the context of Qi功 external qi: the “measurement condition families” represented by ρ=0.2 and ρ=0.4 correspond physiologically to different signal morphologies—the low-correlation end aligns with baseline-noise-dominated respiratory patterns, while the high-correlation end aligns with post-training patterns exhibiting clear rhythmic coupling. Choosing which end serves as the “baseline” determines “which population state the domain map speaks to.” Setting ρ=0.2 as baseline makes the domain map answer “does this method still work for novices or quiet states?”; setting ρ=0.4 makes it answer “what is the upper limit of the method under good rhythmic conditions?” Both questions are valid, but their answers cannot be mixed on one map.

    The contamination of conclusions by reference value selection has proven precedents in clinical testing: if the Reference Change Value (RCV) is uniformly expressed as “a fixed percentage of initial sample concentration,” diagnostic accuracy suffers systematic decline; using an RCV functionally dependent on the first measurement value performs better (PMID 33787419)—this is positive evidence that “the choice of normalization baseline truly alters judgment outcomes.” Similarly, the mechanism behind spurious correlation between “percent change” metrics and baseline values has been repeatedly demonstrated; dividing directly by the baseline creates non-existent structures (PMID 26979481). The Qi功 external qi normalization criterion follows the structure “change divided by some baseline interval,” inheriting the same risk: minute shifts at the baseline end can amplify at the individual level into flips between “applicable” and “inapplicable.”

    Therefore, the judgment logic of this project must proceed in three layers: first, examine shape—if the IoU between two domain maps falls below the recovery threshold (e.g., 0.7), the baseline end’s influence on the domain map is too large to ignore, and the domain map can only be downgraded to “a local description under this baseline version”; any cross-study domain comparison must explicitly state the baseline end. Second, examine the spatial structure of discrepant grids—if discrepant grids form continuous regions rather than scattered points, directional bias exists rather than random noise; this bias direction must be stated in Methods and Limitations. Third, assess conclusion resilience—subsequent analyses using the domain map as a screening criterion should be run once under each of the two baselines; only if conclusion directions remain consistent may they be written into the main criterion; otherwise, both versions must be reported simultaneously. Only after completing this three-layer procedure does the domain map deliverable “earn” the precision it is meant to carry.

    Evidence and Limitations

    Verified: The selection of the normalized baseline systematically alters the determination of proportionally scaled metrics (PMID 26979481, 33787419); sensitivity to the selection of domain boundaries applies to training set and descriptor space choices, requiring explicit definition of the applicable domain (PMID 29934891, 19530661); IoU/overlap metrics are suitable for assessing consistency within small regions but are influenced by region size, necessitating partition auditing (PMID 17117774).

    Partially Verified: The mechanism of “shifting the baseline endpoint → domain shape drift” holds in QSAR and clinical assay contexts, yet direct measurement on the ρ×m grid of Qi Gong external Qi remains unperformed; spatial clustering structures of the divergence grid (scattered points vs. contiguous patches) lack precedent in this field.

    Pending Verification: Whether the empirical IoU and recovery thresholds for the two baseline domain maps (ρ=0.2 and ρ=0.4) hold true; whether the divergence grid clusters along the ρ axis or the m axis in contiguous patches; and whether the adjudication process, using IoU and the divergence grid list as criteria, can stably distinguish “noise-type flips” from “bias-type flips.”

    Related Reading

    Evidence Level: C
    References: PMID 29934891, PMID 19530661, PMID 26979481, PMID 33787419, PMID 17117774

    L1 Network Status: Neither CORRESPONDENCE_NETWORK nor EDGE_TABLE_science-yixue contains edges corresponding to “normalized baseline × domain shape × IoU”. The nearest neighbor E-061 (respiratory rhythm 0.1 Hz spectral peak ↔ Mayer wave, L2/open) represents an “operation → observable → threshold” methodological chain and does not directly map to this project (pure metrological adjudication). This topic has not yet entered the L1 correspondence network; this article serves as a methodological exploration deliverable and must not be cited as a verified mechanism.


    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.

  • 中间相关强度的根号形态保真度:外气判读地板判据在ρ等于0.3和0.5处还准不准

    Evidence level: B

    一个只在两个端点检验过的公式,凭什么替中间的所有情况担保?外气判读的统计地板判据,这一关必须过。

    正文

    外气判读这条研究线的统计工程里,有一块「地板」要立得住:判读结果的变异里,有多少是构型本身的真实信号,有多少是判读者之间的相关性带来的结构效应。判据公式里有一项和「判读者间相关强度ρ」纠缠在一起,前一阶段的重建(第1570号课题)把公式里一个粗糙的相关项换成了更合理的根号形态——大致是说,相关结构对区间标尺的贡献,应该按「一加上m减一倍ρ再开根号」这个形状走,其中m是构型数。这个替换在ρ等于0.2和0.6两个端点上被验证过:端点处公式的行为和仿真实测对得上。

    问题在于:ρ在真实判读场景里不会只落在端点。两位判读者读完同一批样本,相关性可能不高不低地停在0.3或0.5。根号形态是一条曲线,两个端点对得上,不代表中间也对得上——曲线在中间完全可能鼓出一个端点看不出来的包。这不是抬杠,是统计建模的老教训:方差比这类区间的性质对相关结构和平衡结构的双重依赖,早在混合线性模型的经典文献里就被反复警示过(PMID 4005371);小样本下区间形状的定制,差一点点结构假设,覆盖就差一截(PMID 20231867)。

    所以这一轮的裁决动作很朴素:把中间两档ρ等于0.3和0.5补扫出来,和端点放在一起看。设计是八格重扫——ρ取0.3和0.5两档,构型数m取2、4、8、16四档,逐格重算α的实测值,再和根号形态的预测值比对。比对的对象不是α本身,而是消掉构型数维度之后的残差α撇:如果根号形态在中间ρ也成立,那么同一m档下,不同ρ算出来的α撇应该都缩回同一个常数附近,漂移不超过正负0.15。

    裁决分三个出口。第一个出口:八格全部落进对应m档常数正负0.15之内——那么中间ρ豁免,修正闭式判据在全部ρ范围成立,一条公式走天下,不用再分档。第二个出口:ρ等于0.5那几格漂移超过0.3,而且漂移方向随m变号——这说明根号形态只在端点近似成立,中间是系统性偏离,规范必须加插值修正条款,不能让公式硬撑。第三个出口最微妙:漂移量随自由度档位变化——这暴露的不是ρ的错,而是误差机器里还藏着一个自由度和ρ的交互项没被吸收,公式得回炉,先补结构再谈精度。

    为什么中间档值得单独一篇文章?因为「端点验证」是数值工程里最容易被高估的证据形态。两个端点各对一次,最多证明曲线在那两处过线,中间任何形状都兼容。而判据是要交付给判读流程用的,交付物不能带着它没挣过的精度出门。这和第1582号课题处理构型数残余是同一个纪律:那里是把m轴的残余压成单参数公式,这里是把ρ轴的中间地带补验。两条轴各自清账之后,「修正闭式」这块地板才算全域立住。

    对外气研究本身,这件事的意义是守门而不是加冕:判据公式的适用范围被实打实地圈出来了,圈外的数据不会因为公式好看就被误判。方法学上这对应一类成熟做法——区间构造方法在宣称全域可用之前,要在参数空间的中间网格上系统重验(PMID 30634920),少研究情境下尤其如此;而对非中心参数这类量的精确闭式区间,形状假设的每一处妥协都要单独标价(PMID 36725775)。

    打个比方:你在桥两头各踩一脚,桥都稳,就说整座桥能走人——这结论只在桥面是直线时才成立。要是桥面中间鼓了个包,两头的脚感完全探不出来。根号形态就是这座桥,ρ等于0.3和0.5就是桥中间那两步。中间补扫不通过,规范就得加「限重条款」:公式只在端点附近用,中间地带按插值修正走;补扫通过,判读流程才能放心把整段ρ范围都交给一条公式。把裁决标准先写死再跑数,也是为了让结果没有解释弹性——漂移多少算过、多少算插值、多少算回炉,三个出口事先划线,跑完对号入座。

    证据与局限

    已验证: 方差比及类似区间对相关结构与平衡结构的双重依赖有经典文献支撑(PMID 4005371);小样本区间形状定制与覆盖的敏感性有直接文献(PMID 20231867);精确闭式区间构造范式(PMID 36725775)与少研究情境下区间重验范式(PMID 30634920)均为已发表方法学结论。

    部分验证: 根号形态在ρ等于0.2和0.6两个端点与仿真实测一致(承自第1570号课题的重建数据);端点成立不能外推为全域成立,中间两档的八格重扫是本文提出的裁决设计,结果待跑。

    待验证: 八格α撇是否落m档常数正负0.15(全ρ域成立分支);ρ等于0.5档是否漂移超0.3且随m反号(插值修正分支);漂移是否随自由度变化(df乘ρ交互项分支)。

    相关阅读

    证据等级: B
    参考文献: PMID 4005371, PMID 20231867, PMID 36725775, PMID 30634920

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索

    L1 网络状态:该主题尚未进入 L1 对应网络。CORRESPONDENCE_NETWORK 与 EDGE_TABLE_science-yixue 中均无「根号形态保真/中间 ρ/ε* 比值」对应边(最近邻为统计方法学项目内仿真线,非 L1 结构边)。


    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。

    Root-mean-square fidelity of intermediate correlation strength: Are the external gas judgment floor criteria still accurate at ρ equal to 0.3 and 0.5?

    Evidence level: B

    Intermediate Correlation Strength Square-Root Fidelity: Validity of the Outer Qi Reading Floor Criterion at ρ = 0.3 and 0.5

    A formula validated only at two endpoints cannot may support intermediate cases; the statistical floor criterion for outer Qi reading must be tested here.

    ρ Validity of Floor Criterion
    0.3 ?
    0.5 ?

    Evidence and Limitations (证据与局限)

    • Evidence Level: B
    • Traditional Consensus: Yes
    • Structural Analogy: Square-root scaling in pattern-timing models

    The square-root fidelity of intermediate correlation strength is a necessary condition for the outer Qi reading floor criterion to remain valid at ρ = 0.3 and ρ = 0.5. Without empirical verification across this range, the criterion cannot be claimed to hold universally.

    Body

    In the statistical engineering of this research line on Outer Qi Assessment (外气判读), a critical “floor” must be established: distinguishing how much of the variation in assessment results stems from the true signal of the configuration itself versus structural effects arising from inter-assessor correlation. The assessment formula includes a term entangled with the “inter-assessor correlation strength ρ”. A prior reconstruction (Project #1570) replaced a coarse correlation term in the formula with a more reasonable square-root form. This implies that the contribution of the correlation structure to the interval scale should follow a shape of “one plus m minus one times ρ, all under a square root,” where m is the number of configurations. This substitution has been validated at the endpoints where ρ equals 0.2 and 0.6: the formula’s behavior aligns with simulation measurements at these points.

    The problem lies in the fact that ρ in real assessment scenarios will not merely fall at these endpoints. After two assessors read the same batch of samples, their correlation might settle somewhere between 0.3 or 0.5—not high, not low. Since the square-root form is a curve, matching at two endpoints does not may support agreement in between; the curve could bulge in the middle with a peak invisible to endpoint checks. This is not nitpicking but a classic lesson in statistical modeling: the nature of variance ratios depends on both the interval structure and the balance structure, a dual dependency repeatedly warned against in classic literature on mixed linear models (PMID 4005371); under small sample sizes, customizing interval shapes even slightly off in structural assumptions leads to reduced coverage (PMID 20231867).

    Therefore, this round of adjudication is straightforward: fill in the middle two tiers where ρ equals 0.3 and 0.5, and examine them alongside the endpoints. The design involves an eight-grid re-scan—ρ takes two levels (0.3 and 0.5), and m takes four levels (2, 4, 8, 16)—recalculating the measured α values cell by cell and comparing them against the square-root prediction. The comparison target is not α itself, but the residual α-prime after removing the dimension of configuration count: if the square-root form holds across intermediate ρ values, then within each m tier, α-prime derived from different ρ levels should all converge near the same constant, drifting no more than ±0.15.

    The verdict unfolds through three exits. First exit: all eight grids fall within ±0.15 of the corresponding constant for their m tier—then intermediate ρ is exempt, and the corrected closed-form criterion holds across the full ρ range. One formula suffices for all cases, eliminating the need for tiered application. Second exit: grids where ρ equals 0.5 drift beyond 0.3, and the direction of drift reverses as m changes—this indicates the square-root form is only approximately valid at endpoints, while the middle exhibits systematic deviation. The standard must therefore include interpolation correction clauses; the formula cannot be forced to hold without adjustment. Third exit, most subtle: the drift amount varies with degrees of freedom tiers—this reveals not an error in ρ, but that the error machinery still contains an unabsorbed interaction term between degrees of freedom and ρ. The formula must be remade, first supplementing the structure before addressing precision.

    Why is the middle tier worthy of its own article? Because “endpoint validation” is the most easily overestimated evidence form in numerical engineering. Validating at two endpoints merely proves the curve passes through those two points; any shape in between remains compatible. Yet the criterion must be delivered for use in the assessment workflow, and the deliverable cannot carry precision it has not earned. This mirrors the discipline of Project #1582, which handled residual configuration counts by compressing the m-axis into a single-parameter formula; here, we are completing validation across the ρ-axis middle ground. Once both axes are cleared—”corrected closed-form” becomes a universal floor for the entire domain.

    For Outer Qi research itself, this achievement serves as a gatekeeping measure rather than an accolade: it rigorously delineates the applicable range of the assessment formula, preventing out-of-range data from being misjudged simply because the formula looks elegant. Methodologically, this corresponds to a mature approach: interval construction methods must systematically revalidate across intermediate parameter grids before claiming universal applicability (PMID 30634920), especially under limited research contexts; and for precise closed-form intervals involving non-central parameters, every compromise in shape assumption must be individually priced (PMID 36725775).

    To illustrate: stepping on both ends of a bridge and finding it stable does not mean the entire bridge is walkable—this conclusion holds only if the bridge deck is straight. If the middle of the bridge deck bulges, footfalls at the ends cannot detect it. The square-root form is this bridge; ρ equal to 0.3 and 0.5 represents the middle steps. Without passing the middle re-scan, the standard must add “weight-limit clauses”: the formula applies only near endpoints, while the middle region follows interpolation correction. If the re-scan passes, the assessment workflow can confidently apply one formula across the entire ρ range. Writing the adjudication criteria in advance ensures no interpretive flexibility—how much drift counts as over, how much requires interpolation, and how much demands remaking are all pre-defined. Running the numbers against these fixed boundaries ensures results can be matched precisely to their designated exit.

    Evidence and Limitations

    Verified: Classic literature supports the dual dependence of variance ratios and similar intervals on both correlated and balanced structures (PMID 4005371); direct literature exists regarding the sensitivity of small-sample interval shape customization and coverage (PMID 20231867); precise closed-form interval construction paradigms (PMID 36725775) and revalidation paradigms under limited research contexts (PMID 30634920) represent published methodological conclusions.

    Partially Verified: Root morphology aligns with simulation and measurement at ρ endpoints of 0.2 and 0.6 (reconstruction data from Project #1570); endpoint validity cannot be extrapolated to the entire domain; the eight-grid re-sweep for intermediate tiers constitutes the adjudication design proposed herein, with results pending execution.

    Pending Verification: Whether eight-grid α撇 falls within the m-domain constant ±0.15 (full ρ-domain valid branch); whether the ρ=0.5 tier drifts beyond 0.3 and reverses sign with m (interpolation correction branch); whether drift varies with degrees of freedom (df×ρ interaction branch).

    Related Reading

    Evidence Level: B
    References: PMID 4005371, PMID 20231867, PMID 36725775, PMID 30634920

    Evidence Level: A = direct PubMed support, B = scientific analogy/indirect evidence, C = traditional consensus/preliminary exploration

    L1 Network Status: This topic has not yet entered the corresponding L1 network. Neither CORRESPONDENCE_NETWORK nor EDGE_TABLE_science-yixue contains the “square-root form fidelity / intermediate ρ/ε* ratio” correspondence edge (the nearest neighbor is the simulation line within the statistical methodology project, not an L1 structural edge).


    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.

  • 15% 这条线是量出来的还是拍的:联合适用域边界对阈值选择的敏感性

    Evidence level: C

    任何画在网格上的「适用域」边界,都是一条人造的等值线:阈值挪一挪,域就跟着挪——阈值本身有多稳,域的结论才有多硬。

    正文

    上一篇文章(mse-balance-rho-m-joint-applicability-single-line-boundary)给出了一个联合适用域:在 ρ×m×df 的仿真网格上,把逐格偏差不超过 15% 的格子圈起来,作为「单行判决公式可以放心用」的区域。这个交付物看起来客观——每格偏差都是算出来的。但圈域的那条线不是网格自己长出来的,是设计者选的:为什么是 15%,不是 10% 或 20%?这一问就问到了要害——如果换一个同样说得过去的阈值,域的形状大不一样,那「域内」这个结论就不是数据的属性,而是阈值的属性。

    这在测量学里不是新问题。诊断检验研究早就发现:用数据挑出来的「最优 cutoff」会让灵敏度和特异度系统性地过于乐观,样本越小偏得越狠,想要稳健的阈值估计需要相当大的样本量(PMID 18258670)。近年更进一步的思路是:阈值的选择根本不该只看统计指标,还要看选错阈值 downstream 的代价——把不同阈值下的后果摆出来比较,结论才立得住(PMID 36970849)。换句话说,阈值线不是从数据里「读」出来的,是「定」出来的;定了之后,它对结果的拉动必须单独交账。

    适用域研究里同样如此。化学信息学的「适用域」定义明确指出:画边界本质上是在做离群判定,而边界画在哪里,直接决定多少对象被圈进来、多少被排除(PMID 27492083)。更细的基准测试发现,不同的域定义方法之间是覆盖率与可靠性的权衡——域画得宽松,覆盖广但错判多;画得严,错判少但几乎不剩什么可用区(PMID 32756326)。这些研究共同的教训是:域边界是一个带旋钮的交付物,报告域而不报告旋钮位置及其敏感性,等于交付了半张图。

    画在网格上的边界还有一个特殊结构值得单独盯:角格。网格四个角对应最极端的参数组合——ρ 最低、m 最大那一角,往往是偏差曲线最陡的地方,也是三档阈值之间距离拉得最开的地方。如果说域内部各格在三档阈值下归属一致还算平常,角格的归属翻来翻去才是真正的警报:它说明边界位移不是均匀的「整体膨胀收缩」,而是从最脆弱的角先破。所以三档敏感性检验除了算总位移率,还要单列角格归属表——四个角在三档下各归域内还是域外,一翻就知道这条边界从哪里开始不可信。

    落到我们这条线上,检验方案是现成的:#1562 已经产出 ρ×m×df 网格的逐格偏差数据,基准锚点取 ρ=0.4、m≤8。在同一批数据上分别按 10%、15%、20% 三档阈值画三条域边界,然后量两件事:其一,格位移率——从 15% 边界挪到 10% 或 20% 边界时,有多少格的归属翻转;其二,角格归属——网格四个角(最极端的参数组合)在三档阈值下是否一致。判据很直白:若 10% 与 20% 两条边界之间的位移达到半张网格的量级,说明「15% 域内」对阈值高度敏感,分档表的启用区必须连阈值带宽一起报——即交付物从「这条线内可用」改写成「这条线±X 格内可用、具体取决于你接受多大偏差」;若三档边界几乎重合、位移不到一格,说明域的形状对阈值不敏感,15% 这条线才算「量出来的」而非「拍的」,分档表可以放心用。

    这个检验还有一层更实际的意义:域图是给下游用的。下游拿到「15% 域内」会当成一条硬边界来切数据,他们不知道这条线本身有 ±5 个百分点的带宽。上游不报敏感性,下游就会把软边界当硬边界用——这是测量链条上最常见的静默失效方式。与其等下游误用,不如在上游就把阈值的带宽标定出来,和域图一起交付。

    证据与局限

    已验证:数据驱动选 cutoff 会系统性高估判别精度,且样本越越小偏得越狠(PMID 18258670);适用域边界的定义方式直接决定覆盖与错判的权衡,不同定义结论不同(PMID 27492083、PMID 32756326);把阈值的下游代价纳入选择已是成熟做法(PMID 36970849)。

    部分验证:「阈值敏感性应作为域图交付物的一部分」是从上述文献直接类推到本线的方法学要求,逻辑成立但未在本线数据上跑过。

    待验证:10%/15%/20% 三档边界在本线 ρ×m×df 网格上的实际位移率与角格归属——这是纯仿真可算的,等 #1562 网格复用即可出数;出数之前,15% 域只能当参考图,不能当硬边界用。

    相关阅读

    证据等级: C
    参考文献: PMID 18258670, PMID 36970849, PMID 27492083, PMID 32756326

    证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索

    L1 网络状态:该主题尚未进入 L1 对应网络(CORRESPONDENCE_NETWORK.md 与 EDGE_TABLE_science-yixue.md 中无「阈值选择敏感性 × 适用域边界位移」对应边)。最近邻边为 E-061「呼吸节律(0.1 Hz 谱峰)↔ Mayer 波 / 自主神经谱」,层级 L2、状态 open——原文摘录:「0.1 Hz 四来源(Mayer 脊髓源 / 呼吸谐波倍频 / Traube-Hering 呼吸相位 / 拍频)已找到可操作分离判据(屏气 vs 定速呼吸)……L2(操作→观测量→阈值三件套已具备,判别实验待做)」。该边只覆盖呼吸节律与自主神经谱的机制归属,不覆盖测量学阈值敏感性,本文按 C 级处理。


    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。

    Is the 15% threshold line measured or captured: sensitivity of joint applicability domain boundaries to threshold selection

    Evidence level: C

    Any “applicability domain” boundary drawn on a grid is an artificial contour line: shifting the threshold shifts the domain accordingly—the stability of the conclusion depends on how stable the threshold itself is.

    Body

    The preceding article (mse-balance-rho-m-joint-applicability-single-line-boundary) established a joint applicability domain: on the ρ×m×df simulation grid, cells with per-cell deviations not exceeding 15% are circled as the region where “the single-row judgment formula can be safely applied.” This deliverable appears objective—each cell’s deviation is calculated. However, the line defining this circled region was not generated by the grid itself; it was selected by the designer: why 15%, and not 10% or 20%? This question strikes at the crux—if a different, equally plausible threshold is chosen, the shape of the domain changes dramatically; thus, the conclusion that “the region is within the domain” reflects properties of the threshold rather than the data.

    This is not a new problem in metrology. Diagnostic testing research has long established that an “optimal cutoff” derived from data systematically overestimates sensitivity and specificity, with smaller samples yielding greater bias; robust threshold estimation requires substantial sample sizes (PMID 18258670). More recent approaches emphasize that threshold selection should not rely solely on statistical metrics but must account for the downstream costs of choosing an incorrect threshold—comparing consequences across different thresholds yields more robust conclusions (PMID 36970849). In other words, the threshold line is not “read” from the data but “set”; once set, its influence on results must be accounted for separately.

    Applicability domain studies similarly face this issue. The definition of “applicability domain” in cheminformatics explicitly states that drawing boundaries is essentially outlier detection, and where the boundary is drawn directly determines how many objects are included versus excluded (PMID 27492083). Finer benchmarking reveals that different domain definition methods involve a trade-off between coverage and reliability: loose boundaries yield broad coverage but more misclassifications, while strict boundaries reduce errors but leave little usable area (PMID 32756326). The common lesson is: the domain boundary is a tunable deliverable; reporting the domain without specifying the knob position and its sensitivity is akin to delivering only half a map.

    The boundary drawn on the grid has another special structure worth separate attention: corner cells. The four corners of the grid correspond to the most extreme parameter combinations—specifically, the corner where ρ is lowest and m is highest often represents the steepest part of the deviation curve and where the distance between the three thresholds is greatest. If consistent assignment of internal grid cells across all three thresholds is considered normal, the fluctuating assignment of corner cells serves as a true warning: it indicates that boundary displacement is not uniform “overall expansion/contraction” but rather breaks first at the most vulnerable corners. Therefore, the three-threshold sensitivity test must not only calculate the total displacement rate but also separately tabulate corner cell assignments—checking whether each of the four corners falls within or outside the domain under each threshold. A flip in corner assignment immediately reveals where the boundary becomes unreliable.

    Applied to our specific case, the testing protocol is ready: #1562 has already produced per-cell deviation data for the ρ×m×df grid, with baseline anchors at ρ=0.4 and m≤8. Using the same dataset, three domain boundaries are drawn at thresholds of 10%, 15%, and 20%. Two metrics are then measured: first, the grid displacement rate—how many cells change assignment when moving from the 15% boundary to either the 10% or 20% boundary; second, corner cell assignment—whether the four corners (the most extreme parameter combinations) remain consistent across all three thresholds. The criteria are straightforward: if the displacement between the 10% and 20% boundaries spans half the grid, it indicates that “within the 15% domain” is highly sensitive to threshold choice; the tiered table must then report the usable region together with the threshold bandwidth—rephrasing the deliverable from “usable within this line” to “usable within ±X cells of this line, depending on the maximum deviation you accept.” Conversely, if the three boundaries nearly overlap with displacement less than one cell, the domain shape is insensitive to the threshold; the 15% line is truly “measured” rather than “arbitrarily set,” and the tiered table can be used with confidence.

    This test also carries practical significance: the domain map is intended for downstream use. Downstream users interpreting “within the 15% domain” will treat it as a hard boundary for data slicing, unaware that this line itself has a ±5 percentage point bandwidth. If upstream does not report sensitivity, downstream will mistakenly treat the soft boundary as hard—this is the most common silent failure mode in measurement chains. Rather than waiting for downstream misapplication, the upstream should calibrate and report the threshold bandwidth alongside the domain map.

    Evidence and Limitations

    Verified: Data-driven selection of the cutoff systematically overestimates discriminative accuracy, and smaller samples yield greater bias (PMID 18258670); the definition of domain boundaries directly determines the trade-off between coverage and misclassification, leading to divergent conclusions under different definitions (PMID 27492083, PMID 32756326); incorporating downstream costs of the threshold into selection is an established practice (PMID 36970849).

    Partially Verified: The requirement that “threshold sensitivity should be delivered as part of the domain map” is a methodological extension derived directly from the above literature to this line of inquiry; the logic holds, but it has not yet been tested against data from this specific line.

    Pending Verification: The actual displacement rate and angular grid membership for the three boundary tiers (10%/15%/20%) on the ρ×m×df grid of this line—these are purely simulated values and can be computed once the #1562 grid is reused; until numerical results are available, the 15% domain must be treated as a reference diagram and not used as a rigid boundary.

    Related Reading

    Evidence Level: C
    References: PMID 18258670, PMID 36970849, PMID 27492083, PMID 32756326

    Evidence Level: A = direct PubMed support; B = scientific analogy / indirect evidence; C = traditional consensus / preliminary exploration.

    L1 Network Status: This topic has not yet entered the corresponding L1 network (no edge linking “threshold-selection sensitivity × applicability-domain boundary displacement” exists in CORRESPONDENCE_NETWORK.md and EDGE_TABLE_science-yixue.md). The nearest neighbor edge is E-061 “respiratory rhythm (0.1 Hz spectral peak) ↔ Mayer wave / autonomic spectrum”, at level L2, status open — excerpted from the original text: “Four sources of 0.1 Hz (Mayer spinal source / respiratory harmonic submultiples / Traube-Hering respiratory phase / beat frequency) have yielded an operational separation criterion (breath-hold vs constant-rate breathing) … L2 (the triad of operation → observable → threshold is in place; discriminative experiments remain to be conducted).” This edge covers only the mechanistic attribution of respiratory rhythm and autonomic spectrum, not metrological threshold sensitivity; therefore this article is treated 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.