Evidence level: C
修炼圈常说「入门即不退」——好像练过某个坎之后,功夫就永远不会掉。运动科学里对应的问题是:保留百分比随累计训练小时数的曲线,是平滑爬坡,还是存在一个拐点、过了它保留底陡然抬升?
正文(约1250字)
传统武学与养生修炼里有个流传很广的说法:基本功练到某个程度,「老本」就打下了——哪怕多年不练,捡起来也比新人快得多,甚至不会退步。用现代语言翻译,这是在主张「保留百分比—累计剂量」曲线不是线性积累,而是分段函数:低于某个累计训练小时数,停练后保留底约等于零;超过之后,保留底陡然抬升,出现「门槛」或「拐点」。竞争模型则是简单的线性(或对数)积累:练多少留多少,没有魔法门槛。
两种模型谁对?先看支持「门槛」一方的证据。一项纵向追踪研究让练习者练双手协调任务两个月,之后分别测 6 个月和 8 年的保留——结果惊人:不仅任务本身保留得很好,连个人的动作「风格」都原样重现,只有最精细的跨手串扰参数在 8 年后消失(PMID 24032015)。这个「学过就基本忘不掉」的现象,确实像暗示存在某种一过线就锁定的机制。
但再看反方。对慢性卒中康复 DOSE 剂量试验的再分析用分段线性混合模型专门检验了剂量与保留的关系,结论出人意料:剂量越大,练习后 6 个月的保留反而越差——大剂量组的衰减比小剂量组更快,作者提出「技能水平越高、维持成本越高」来解释(PMID 32830617)。如果保留底真有单调门槛,不该出现这种负向剂量效应。过度学习文献同样给出混合信号:稳定仪任务上达到标准后再加练 100%,快学习者与慢学习者的绝对保留并无显著差异(PMID 23952596)——多练的那段剂量并没有稳定转化为更高的保留底。
把两边摆在一起,现有证据更像这样一幅图:粗粒度的技能骨架一旦习得,保留确实非常顽固(24032015 的 8 年证据);但保留底的量并不随累计剂量平滑累积,甚至高剂量会因技能结构更复杂而衰减更快(32830617)。也就是说,「入门即不退」在「会不会忘」这个二元问题上部分成立,在「留多少」这个连续问题上不成立;而「线性积累」模型在两个问题上都过于粗糙——它既解释不了 8 年不忘,也解释不了负向剂量效应。
要真正裁决,需要的实验设计很明确:按累计训练小时数分多个剂量层(如 5/15/30/60/120 小时),统一停练协议与停练时长,统一保留测量(保留百分比=停练后成绩相对首日标准的保持比例),然后对保留百分比—剂量曲线做分段回归(segmented regression),报告拐点位置及其置信区间;再与传统线性/对数模型做 BIC 比较。若拐点置信区间窄且远离零,门槛模型胜;若曲线单调无内点,线性模型胜;若出现倒 U 形,则需要第三种模型——「最优剂量」模型。
测量本身也有陷阱。看「保留多少」至少有两种口径:一是停练后直接测成绩保持了几成,二是测「省力率」——复练时比初学快多少。同一个被试,前者可能掉到六成,后者却接近百分之百,两种口径给出的「保留底」完全不同。若口径不统一,拐点检测会把两种混合信号搅在一起。所以上述裁决实验必须预先锁定口径:主指标用保留百分比,省力率作次指标分开建模,两套结论并排报告,不允许哪边显著就只报哪边。
有趣的是,这套「剂量—保留—门槛」的裁决框架,与《易经》乾卦「见龙在田,利见大人」的进阶结构有形式上的呼应:传统工夫论从不主张匀速精进,而是强调阶段性台阶——每一爻是一次相变,跨过之后才谈得上稳固。这个类比当然不能当证据,但它提示了一个可检验的差异化预测:如果保留底真有拐点,拐点两侧被试的再学习曲线(复练快于初练的程度)应该呈现质的差异,而不是量的差异。这正是本篇留给后续研究的接口。
证据与局限
已验证: 长期运动记忆可以跨年保留,且保留内容包含个体化动作特征(PMID 24032015,纵向 8 年追踪);剂量—保留关系可用分段线性模型形式化并检验(PMID 32830617)。
部分验证: 过度学习在部分任务上提升短期保留,但对不同学习者水平的长期保留增益不一致(PMID 23952596);高剂量伴随更快停练后衰减(PMID 32830617)。
待验证: 保留百分比—累计剂量曲线是否存在明确拐点及其置信区间;拐点两侧再学习曲线是否呈质变;拐点位置是否随任务复杂度移动。该主题尚未进入 L1 对应网络,本文证据等级按规则封顶为 C。
相关阅读
L1 网络状态:EDGE_TABLE_science-yixue.md 与 CORRESPONDENCE_NETWORK.md 中检索「停练/保留底/保留百分比/累计剂量/阈值效应」均无对应边(唯一命中为屏气窒息断点,主题无关)——该边尚未建立。
证据等级: C
参考文献: PMID 24032015, PMID 32830617, PMID 23952596
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。
Threshold Effect of Retaining the Bottom in Training: When Does Kung Fu Truly Enter the Bones?
Evidence level: C
Threshold Effect Verdict on Retaining Base in Practice: When Does Training Truly “Sink into the Bones”?
In traditional Chinese metaphysics, retaining base (a foundational level of accumulated energy or structural integrity) refers to the concept that once a certain threshold is reached, the system maintains its core stability without regression. In evidence-based terms, this translates to: does the percentage of retained capacity rise smoothly with cumulative training hours, or is there a distinct inflection point beyond which the base level increases abruptly?
Body (approx. 1250 words)
In traditional martial arts and health cultivation, a widely circulated claim holds that once basic skills reach a certain threshold, the “old foundation” is established—even after years of inactivity, resuming practice yields faster progress than for novices, and regression is avoided. Translated into modern terms, this asserts that the “retention percentage versus cumulative dose” curve is not linearly accumulated but rather a piecewise function: below a certain cumulative training hour count, the retention floor remains near zero after stopping; beyond it, the floor abruptly rises, creating a “threshold” or “inflection point.” In contrast, competitive models assume simple linear (or logarithmic) accumulation: retain what you train, with no symbolic threshold.
Which model is correct? First, consider evidence supporting the “threshold.” A longitudinal tracking study had practitioners train a hand-coordination task for two months, then measured retention at 6 and 8 years—results were striking: not only was the task itself well retained, but individual movement “styles” reappeared intact; only the most refined cross-hand interference parameters vanished after 8 years (PMID 24032015). This “once learned, hardly forgotten” phenomenon indeed hints at some mechanism that locks in once crossed.
Now consider the counter-evidence. A re-analysis of the DOSE dose trial for chronic stroke rehabilitation, using segmented linear mixed models to specifically test dose versus retention, yielded an unexpected conclusion: higher doses led to worse 6-month retention after practice—the high-dose group decayed faster than the low-dose group. Authors attributed this to “higher skill level, higher maintenance cost” (PMID 32830617). If the retention floor truly had a monotonic threshold, such negative dose effects should not occur. Overlearning literature also provides mixed signals: on stabilometer tasks, after reaching standard, adding 100% more practice showed no significant difference in absolute retention between fast and slow learners (PMID 23952596)—the extra dose did not stably convert into higher retention floor.
Putting both sides together, current evidence resembles this picture: once coarse-grained skill skeletons are acquired, retention is indeed very stubborn (the 8-year evidence from PMID 24032015); but the amount of retention floor does not smoothly accumulate with cumulative dose, and high doses may even decay faster due to more complex skill structures (PMID 32830617). Thus, “entry means no regression” holds partially on the binary question of “whether to forget,” but fails on the continuous question of “how much to retain”; meanwhile, the “linear accumulation” model is too coarse on both questions—it cannot explain neither the 8-year non-forgetting nor the negative dose effects.
To truly adjudicate, the experimental design must be clear: divide into multiple dose layers by cumulative training hours (e.g., 5/15/30/60/120 hours), standardize stopping protocols and duration, unify retention measurement (retention percentage = proportion of post-stop performance relative to day-one standard), then perform segmented regression on the retention percentage versus dose curve, reporting inflection point location and confidence intervals; compare with traditional linear/logarithmic models via BIC. If the inflection point’s confidence interval is narrow and far from zero, the threshold model wins; if the curve is monotonic with no interior point, the linear model wins; if an inverted U-shape emerges, a third model—”optimal dose”—is needed.
Measurement itself has pitfalls. Assessing “how much retained” can follow two definitions: first, directly measure how much performance is kept after stopping; second, measure “effort-saving rate”—how much faster re-practice is compared to initial learning. For the same subject, the former might drop to 60%, while the latter approaches nearly 100%; these two definitions yield vastly different “retention floors.” Without unified definition, inflection detection mixes these dual signals. Therefore, the adjudication experiment must pre-lock definitions: primary metric uses retention percentage, effort-saving rate serves as secondary metric modeled separately, both conclusions reported side-by-side, not selectively reporting whichever is significant.
Interestingly, this “dose-retention-threshold” adjudication framework formally echoes the progression structure of the Yijing (I-Ching) Qian hexagram’s “seeing dragon in the field, beneficial to meet the great man”: traditional cultivation theory never advocates uniform steady progress but emphasizes staged steps—each line represents a phase transition, stability only discussed after crossing. This analogy cannot serve as evidence, yet it suggests a testable differentiated prediction: if the retention floor truly has an inflection point, relearning curves (degree of faster re-practice versus initial practice) for subjects on either side of the inflection should show qualitative differences, not merely quantitative ones. This is precisely the interface left for future research in this paper.
Evidence and Limitations
Verified: Long-term motor memory can persist across years, retaining individualized movement characteristics (PMID 24032015, longitudinal 8-year tracking); the dose-retention relationship can be formalized and tested using segmented linear models (PMID 32830617).
Partially Verified: Overlearning enhances short-term retention on certain tasks, but the benefit for long-term retention varies across learner proficiency levels (PMID 23952596); high dosage is accompanied by faster decay after cessation of practice (PMID 32830617).
Pending Verification: Whether the retention percentage–cumulative dose curve exhibits a distinct inflection point and its confidence interval; whether relearning curves on either side of the inflection point undergo qualitative change; and whether the location of the inflection point shifts with task complexity. This topic has not yet entered the L1 corresponding network; consequently, the evidence level for this section is capped at C according to the rules.
Related Reading
- Test of independence between the component “rehearsal accelerates initial practice” and “decay is slower than recovery”
- Asymmetry index of τ_decay/τ_recover and its regulation by “training duration × dosage”
- Research pool: Bio-frequency code
L1 Network status: No corresponding edges were found in EDGE_TABLE_science-yixue.md and CORRESPONDENCE_NETWORK.md for “cessation of practice / retention of baseline / retention percentage / cumulative dosage / threshold effect” (the only match was the breath-holding asphyxia breakpoint, which is irrelevant to the topic) — this edge has not yet been established.
Evidence level: C
References: PMID 24032015, PMID 32830617, PMID 23952596
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.