ADHD 動態配置統合理論:可證偽命題與未來研究綱領
英文題名: An Integrated Dynamic Configuration Theory of ADHD: Falsifiable Hypotheses and a Future Research Program系列: ADHD 動態配置與認知拓撲系列,第 10 篇/封頂篇版本: v1.0日期: 2026-08-17作者: Neo.K(許筌崴)協作: GPT-5.6 Sol文件性質: 理論整合/認知科學命題/研究綱領/系列封頂文件文獻檢索截點: 2026-08-17
0. 醫學、診斷、藥理與證據邊界聲明
本文不是臨床研究、診斷工具、治療指南、藥理試驗、神經影像 biomarker 研究或醫療建議。
本文提出的「動態配置」「神經調節層」「配置熵」「認知拓撲」「臨床可見性」「連續配置空間」「情境匹配核」等概念均屬待驗證理論構件。除非本文明確指出某一背景結論來自既有研究,否則不能將本文新增數學形式視為已被神經科學、精神醫學或臨床心理學驗證的 ADHD 機制。
原作者並非醫學、精神醫學、藥理學、神經科學、遺傳學或臨床心理專業研究者。本文不提供新的臨床、人體、藥理、遺傳、神經影像、流行病學或心理實驗數據。本文所有實證性背景均依賴公開同行評審文獻與正式分類資料。
本文不使用原作者的個人 ADHD 診斷史、用藥經驗或其他個人經驗作為一般化實證證據。個人觀察最多可構成研究問題的來源,不構成群體層證明。
本文不得被用於:
自行診斷或排除 ADHD;
替他人判斷是否有 ADHD;
自行開始、停止、增加、減少或混合藥物;
以本文的 configuration score、entropy、graph metric 或 fit score 取代臨床評估;
以「局部優勢」否定功能損害或醫療需要;
以「連續光譜」主張所有人都有 ADHD;
以「biotype」主張目前已有可供個人臨床使用的腦影像分型。
本文最核心的證據規則是:
formal plausibility ≠ empirical validation ≠ clinical utility . \boxed{
\text{formal plausibility}
\neq
\text{empirical validation}
\neq
\text{clinical utility}.
} formal plausibility = empirical validation = clinical utility .
摘要
本系列以十篇論文重新檢視 ADHD 是否能被建模為一個多層、動態、情境依賴的配置系統,而不是僅以「注意力不足」作為完整機制敘述。
截至 2026 年,外部研究已提供數個彼此相容的重要背景。第一,ADHD symptoms、功能損害與遺傳 liability 具有明顯 dimensional evidence;2025 年大型 GWAS 支持 clinical ADHD 位於由 ADHD symptoms 索引的 continuous liability 高端。第二,ADHD 內部存在顯著 heterogeneity;2026 年 JAMA Psychiatry 的 normative morphometric-network study 在連續個體偏差上辨識出三個可外部驗證的 pediatric biotypes,顯示 dimensional variation 與 local clustering 可以同時存在。第三,刺激劑作用不能簡化為「增加一條注意力」;2025 年 Cell 研究將 stimulant-related connectivity effects 主要連結至 arousal、reward、salience 與 action-related systems,而 2026 年成人 ADHD dual-tracer PET 顯示 methylphenidate 同時影響 DAT 與 NET,但 transporter binding change 與 cognitive improvement 並非簡單一對一。第四,ADHD cognition 的平均值可能不足以描述動態差異;2025 年兒童研究發現 cognitive-control neural representations 的 temporal variability 與 spatial stability 差異,另有 methylphenidate 研究觀察到 whole-brain network flexibility 的穩定化。第五,adult ADHD literature 仍保留多項重大未解問題,包括 late-onset、emotional dysregulation、functional impairment、executive dysfunction、objective diagnostic measures 與長期 treatment effects。
基於上述背景,本篇將前九篇整合為「Integrated Dynamic Configuration Theory, IDCT-ADHD」。其最小核心不是:
ADHD = one mechanism , \text{ADHD}
=
\text{one mechanism}, ADHD = one mechanism ,
而是:
c i , t + 1 = Φ ( c i , t , q i , t , e i , t , s i , t e x t , k i , t , u i , t ; θ i ) + ε i , t . \boxed{
\mathbf c_{i,t+1}
=
\Phi
\left(
\mathbf c_{i,t},
\mathbf q_{i,t},
\mathbf e_{i,t},
\mathbf s_{i,t}^{\mathrm{ext}},
\mathbf k_{i,t},
\mathbf u_{i,t};
\boldsymbol\theta_i
\right)
+
\boldsymbol\varepsilon_{i,t}.
} c i , t + 1 = Φ ( c i , t , q i , t , e i , t , s i , t ext , k i , t , u i , t ; θ i ) + ε i , t .
其中 c i , t \mathbf c_{i,t} c i , t 是時間 t t t 的動態認知配置, q \mathbf q q 是任務/生活需求, e \mathbf e e 是環境, s e x t \mathbf s^{\mathrm{ext}} s ext 是外部支架, k \mathbf k k 是補償策略, u \mathbf u u 是即時輸入, θ i \boldsymbol\theta_i θ i 是較慢變個體參數。
客觀性能為:
p i , t = F ( c i , t , q i , t , e i , t ) . \mathbf p_{i,t}
=
F
\left(
\mathbf c_{i,t},
\mathbf q_{i,t},
\mathbf e_{i,t}
\right). p i , t = F ( c i , t , q i , t , e i , t ) .
主觀狀態:
χ i , t = H ( c i , t , e i , t , x i , t e x p e c t a n c y ) . \chi_{i,t}
=
H
\left(
\mathbf c_{i,t},
\mathbf e_{i,t},
\mathbf x_{i,t}^{\mathrm{expectancy}}
\right). χ i , t = H ( c i , t , e i , t , x i , t expectancy ) .
元認知估計:
p ^ i , t = M ( p i , t , χ i , t , f i , t f e e d b a c k ) . \widehat{\mathbf p}_{i,t}
=
M
\left(
\mathbf p_{i,t},
\chi_{i,t},
\mathbf f_{i,t}^{\mathrm{feedback}}
\right). p i , t = M ( p i , t , χ i , t , f i , t feedback ) .
功能損害:
i i , t = J ( p i , t , q i , t , e i , t , s i , t e x t , k i , t ) . \mathbf i_{i,t}
=
J
\left(
\mathbf p_{i,t},
\mathbf q_{i,t},
\mathbf e_{i,t},
\mathbf s_{i,t}^{\mathrm{ext}},
\mathbf k_{i,t}
\right). i i , t = J ( p i , t , q i , t , e i , t , s i , t ext , k i , t ) .
臨床診斷則保留為獨立決策層:
D i , t = Γ ( y i , t , i i , t , H i d e v , X i , t d i f f ) , \boxed{
\mathfrak D_{i,t}
=
\Gamma
\left(
\mathbf y_{i,t},
\mathbf i_{i,t},
H_i^{\mathrm{dev}},
X_{i,t}^{\mathrm{diff}}
\right),
} D i , t = Γ ( y i , t , i i , t , H i dev , X i , t diff ) ,
其中 y \mathbf y y 為可觀察表型, H d e v H^{\mathrm{dev}} H dev 為發展史, X d i f f X^{\mathrm{diff}} X diff 為跨情境與鑑別診斷證據。
因此本理論的核心不是把 ADHD 診斷替換成數學模型,而是提出:
Configuration ≠ Phenotype ≠ Impairment ≠ Diagnosis . \boxed{
\text{Configuration}
\neq
\text{Phenotype}
\neq
\text{Impairment}
\neq
\text{Diagnosis}.
} Configuration = Phenotype = Impairment = Diagnosis .
本篇建立三層證據帳本:(A)已有較強外部證據支持的背景;(B)有間接或局部支持、但仍需特定驗證的中層連結;(C)主要由本系列提出的新假說。配置熵、認知 graph topology、disengagement barrier、attention debt、reversal surface 等均被明確列入 C 層,不得與 continuous genetic liability、clinical heterogeneity 等 A 層證據等量齊觀。
本文提出十八項統一可證偽命題與六階段研究綱領。若高維動態模型不能在獨立樣本中超越 symptom score、executive-function model、單一 trait dimension 或現行 diagnosis 對功能損害、病程與 treatment-relevant outcomes 的預測,本理論應被簡化或放棄。
最終研究問題是:
Does a dynamic, multilevel, context-sensitive configuration model predict what simpler ADHD models systematically miss? \boxed{
\text{Does a dynamic, multilevel, context-sensitive configuration model
predict what simpler ADHD models systematically miss?}
} Does a dynamic, multilevel, context-sensitive configuration model predict what simpler ADHD models systematically miss?
關鍵詞: ADHD、dynamic configuration、neuromodulation、attention allocation、heterogeneity、continuous liability、biotype、metacognition、context dependence、person–environment fit、falsifiability
1. 本系列真正提出了什麼?
本系列不是提出:
ADHD 的真正本質已被找到。
更弱、也更科學的命題是:
是否存在一個可操作的中層動態模型,可以把神經調節、注意配置、狀態穩定性、認知路徑、主觀狀態、發展支架、功能損害與情境性能放進同一個可證偽框架?
因此:
IDCT-ADHD = research framework , \boxed{
\text{IDCT-ADHD}
=
\text{research framework},
} IDCT-ADHD = research framework ,
不是:
IDCT-ADHD = established medical theory . \boxed{
\text{IDCT-ADHD}
=
\text{established medical theory}.
} IDCT-ADHD = established medical theory .
2. 十篇系列的依賴結構
本系列依序建立:
Paper 1
Attention Deficit → Dynamic Configuration Conjecture . \text{Attention Deficit}
\rightarrow
\text{Dynamic Configuration Conjecture}. Attention Deficit → Dynamic Configuration Conjecture .
Paper 2
Neuromodulation ≠ Attention . \text{Neuromodulation}
\neq
\text{Attention}. Neuromodulation = Attention .
Paper 3
Salience ≠ Activation ≠ Allocation ≠ Observation ≠ Update ≠ Action . \text{Salience}
\neq
\text{Activation}
\neq
\text{Allocation}
\neq
\text{Observation}
\neq
\text{Update}
\neq
\text{Action}. Salience = Activation = Allocation = Observation = Update = Action .
Paper 4
Distractibility ⊥̸ Hyperfocus-like lock-in . \text{Distractibility}
\not\perp
\text{Hyperfocus-like lock-in}. Distractibility ⊥ Hyperfocus-like lock-in .
Paper 5
Associative breadth ≠ Path diversity ≠ Coherence ≠ Convergence . \text{Associative breadth}
\neq
\text{Path diversity}
\neq
\text{Coherence}
\neq
\text{Convergence}. Associative breadth = Path diversity = Coherence = Convergence .
Paper 6
Subjective state ≠ Metacognitive confidence ≠ Objective performance . \text{Subjective state}
\neq
\text{Metacognitive confidence}
\neq
\text{Objective performance}. Subjective state = Metacognitive confidence = Objective performance .
Paper 7
Configuration ≠ Impairment ≠ Visibility ≠ Diagnosis . \text{Configuration}
\neq
\text{Impairment}
\neq
\text{Visibility}
\neq
\text{Diagnosis}. Configuration = Impairment = Visibility = Diagnosis .
Paper 8
Continuous variation + Local clusters + Categorical clinical decisions . \text{Continuous variation}
+
\text{Local clusters}
+
\text{Categorical clinical decisions}. Continuous variation + Local clusters + Categorical clinical decisions .
Paper 9
Functional effect = F ( Configuration , Task , Environment ) . \text{Functional effect}
=
F
\left(
\text{Configuration},
\text{Task},
\text{Environment}
\right). Functional effect = F ( Configuration , Task , Environment ) .
Paper 10
將上述結構統一並建立:
measurement → prediction → falsification → possible translation . \boxed{
\text{measurement}
\rightarrow
\text{prediction}
\rightarrow
\text{falsification}
\rightarrow
\text{possible translation}.
} measurement → prediction → falsification → possible translation .
3. 統一符號:停止讓不同論文的符號互相碰撞
前九篇在局部模型中使用過若干重複符號。
封頂篇重新定義 canonical notation。
3.1 神經調節狀態
n t = neuromodulatory state . \mathbf n_t
=
\text{neuromodulatory state}. n t = neuromodulatory state .
候選包括:
n t = ( D t , N E t , … ) . \mathbf n_t
=
(D_t,NE_t,\ldots). n t = ( D t , N E t , … ) .
3.2 中介調節狀態
z t = ( z t a r o u s a l , z t r e w a r d , z t s a l i e n c e , z t v i g o r , z t g a t i n g , z t s t a b i l i t y ) . \mathbf z_t
=
\left(
z_t^{\mathrm{arousal}},
z_t^{\mathrm{reward}},
z_t^{\mathrm{salience}},
z_t^{\mathrm{vigor}},
z_t^{\mathrm{gating}},
z_t^{\mathrm{stability}}
\right). z t = ( z t arousal , z t reward , z t salience , z t vigor , z t gating , z t stability ) .
3.3 配置分布
π t = ( π 1 ( t ) , … , π n ( t ) ) , \boldsymbol\pi_t
=
\left(
\pi_1(t),\ldots,\pi_n(t)
\right), π t = ( π 1 ( t ) , … , π n ( t ) ) ,
其中:
∑ i π i ( t ) = 1. \sum_i\pi_i(t)=1. i ∑ π i ( t ) = 1.
3.4 配置動態
x t = ( H ^ π , L t , T t d w e l l , ν t s w i t c h , B t e x i t , R t g o a l ) . \mathbf x_t
=
\left(
\widehat{\mathcal H}_{\pi},
L_t,
T_t^{\mathrm{dwell}},
\nu_t^{\mathrm{switch}},
B_t^{\mathrm{exit}},
R_t^{\mathrm{goal}}
\right). x t = ( H π , L t , T t dwell , ν t switch , B t exit , R t goal ) .
3.5 認知關係圖
G t = ( V t , E t , W t ) . \mathcal G_t
=
\left(
V_t,E_t,W_t
\right). G t = ( V t , E t , W t ) .
3.6 表徵、更新與行動狀態
m t = ( m t o b s , m t u p d a t e , m t m e m o r y , m t g a t e ) . \mathbf m_t
=
\left(
m_t^{\mathrm{obs}},
m_t^{\mathrm{update}},
m_t^{\mathrm{memory}},
m_t^{\mathrm{gate}}
\right). m t = ( m t obs , m t update , m t memory , m t gate ) .
3.7 客觀性能
p t = ( p t a c c u r a c y , p t R T , p t R T V , p t m e m o r y , p t i n h i b i t i o n , p t r e a s o n i n g , p t t r a n s f e r , … ) . \mathbf p_t
=
\left(
p_t^{\mathrm{accuracy}},
p_t^{\mathrm{RT}},
p_t^{\mathrm{RTV}},
p_t^{\mathrm{memory}},
p_t^{\mathrm{inhibition}},
p_t^{\mathrm{reasoning}},
p_t^{\mathrm{transfer}},
\ldots
\right). p t = ( p t accuracy , p t RT , p t RTV , p t memory , p t inhibition , p t reasoning , p t transfer , … ) .
3.8 主觀狀態
χ t = subjective clarity/engagement state . \chi_t
=
\text{subjective clarity/engagement state}. χ t = subjective clarity / engagement state .
3.9 元認知估計
p ^ t = estimated performance . \widehat{\mathbf p}_t
=
\text{estimated performance}. p t = estimated performance .
3.10 生命/任務需求
q t = task and life demands . \mathbf q_t
=
\text{task and life demands}. q t = task and life demands .
3.11 環境
e t = environmental state . \mathbf e_t
=
\text{environmental state}. e t = environmental state .
3.12 外部支架
s t e x t = external scaffolding . \mathbf s_t^{\mathrm{ext}}
=
\text{external scaffolding}. s t ext = external scaffolding .
3.13 補償策略
k t = compensation . \mathbf k_t
=
\text{compensation}. k t = compensation .
3.14 功能損害
i t = functional impairment profile . \mathbf i_t
=
\text{functional impairment profile}. i t = functional impairment profile .
3.15 臨床可見性
v t = clinical visibility . v_t
=
\text{clinical visibility}. v t = clinical visibility .
3.16 臨床判定
D t ∈ { 0 , 1 } . \mathfrak D_t
\in
\{0,1\}. D t ∈ { 0 , 1 } .
此符號只是抽象表示「不符合/符合正式診斷決策」,不是臨床計算公式。
4. 全域配置狀態
定義個體 i i i 在時間 t t t 的全域 configuration:
c i , t = ( n i , t , z i , t , π i , t , x i , t , G i , t , m i , t ) . \boxed{
\mathbf c_{i,t}
=
\left(
\mathbf n_{i,t},
\mathbf z_{i,t},
\boldsymbol\pi_{i,t},
\mathbf x_{i,t},
\mathcal G_{i,t},
\mathbf m_{i,t}
\right).
} c i , t = ( n i , t , z i , t , π i , t , x i , t , G i , t , m i , t ) .
此表示故意沒有直接包含:
D . \mathfrak D. D .
因為診斷不是底層認知狀態本身。
5. 全域狀態更新
候選:
c i , t + 1 = Φ ( c i , t , q i , t , e i , t , s i , t e x t , k i , t , u i , t ; θ i ) + ε i , t . \boxed{
\mathbf c_{i,t+1}
=
\Phi
\left(
\mathbf c_{i,t},
\mathbf q_{i,t},
\mathbf e_{i,t},
\mathbf s_{i,t}^{\mathrm{ext}},
\mathbf k_{i,t},
\mathbf u_{i,t};
\boldsymbol\theta_i
\right)
+
\boldsymbol\varepsilon_{i,t}.
} c i , t + 1 = Φ ( c i , t , q i , t , e i , t , s i , t ext , k i , t , u i , t ; θ i ) + ε i , t .
其中:
u i , t \mathbf u_{i,t} u i , t :即時輸入;
θ i \boldsymbol\theta_i θ i :較慢變的個體參數;
ε i , t \boldsymbol\varepsilon_{i,t} ε i , t :未建模波動。
這是系列最重要的候選動力式。
6. 為什麼一定要保留時間?
若只測:
c ‾ i = 1 T ∑ t c i , t , \overline{\mathbf c}_i
=
\frac1T
\sum_t
\mathbf c_{i,t}, c i = T 1 t ∑ c i , t ,
可能遺失:
Var t ( c ) , \operatorname{Var}_t(\mathbf c), Var t ( c ) ,
P ( c t + 1 ∣ c t ) , P
\left(
\mathbf c_{t+1}
\mid
\mathbf c_t
\right), P ( c t + 1 ∣ c t ) ,
T d w e l l , T^{\mathrm{dwell}}, T dwell ,
transition asymmetry . \text{transition asymmetry}. transition asymmetry .
2025 年 cognitive-control neural-stability 研究與 network-flexibility stimulant study 都支持:
temporal organization itself can carry information . \boxed{
\text{temporal organization itself can carry information}.
} temporal organization itself can carry information .
但這仍不證明本文的具體配置變量正確。
7. 外部證據層 A:目前相對較強的背景事實
本文把以下內容列入:
Evidence Layer A . \boxed{
\text{Evidence Layer A}.
} Evidence Layer A .
其意義是:
有多項現代研究或大型研究支持相關背景,但仍不代表本文所有中介公式成立。
8. A1:ADHD 是臨床有效的神經發展診斷框架
截至 2026 年 ICD-11 最新 release,ADHD 仍位於 neurodevelopmental-disorder framework。
因此本系列不主張:
replace ADHD diagnosis . \boxed{
\text{replace ADHD diagnosis}.
} replace ADHD diagnosis .
9. A2:ADHD 具有顯著異質性
2025 World Psychiatry 成人 ADHD 綜述明確將 heterogeneity、functional impairment、executive dysfunction、late-onset、emotional dysregulation 等列為重要研究問題。
2026 JAMA Psychiatry 又以 data-driven methods 找到可重現 pediatric biotypes。
因此:
ADHD ≠ one uniform cognitive state . \boxed{
\text{ADHD}
\neq
\text{one uniform cognitive state}.
} ADHD = one uniform cognitive state .
10. A3:Symptoms 與 Genetic Liability 具有連續性
2025 Nature Genetics GWAS:
290 , 134 290,134 290 , 134
次 symptom measures,
來自:
70 , 953 70,953 70 , 953
名獨立個體,
支持 clinical ADHD 位於 ADHD symptom continuous liability 高端。
因此:
continuous liability \boxed{
\text{continuous liability}
} continuous liability
具有相當外部支持。
11. A4:連續性不排除 Local Clustering
2026 pediatric biotype study 的方法本身先建立 normative dimensional deviations,再進行 clustering。
因此:
dimension + cluster \boxed{
\text{dimension}
+
\text{cluster}
} dimension + cluster
不是邏輯矛盾。
12. A5:Stimulant 作用不是單純「Attention Channel Gain」
2025 Cell study 在大型 ABCD data 與 highly sampled drug-imaging validation 中,將 stimulant-related functional connectivity effects 主要連結至:
arousal;
reward;
salience;
action-related systems;
而非 canonical attention networks 的簡單增強。
因此:
stimulant effect ≠ one attention-network amplifier . \boxed{
\text{stimulant effect}
\neq
\text{one attention-network amplifier}.
} stimulant effect = one attention-network amplifier .
13. A6:DAT 與 NET 都與 Methylphenidate 有關
2026 issue 的成人 ADHD longitudinal dual-tracer PET 顯示 extended-release methylphenidate 同時改變:
D A T DAT D A T
與:
N E T NET N E T
binding。
但 transporter change 與 cognitive improvement 不是簡單一對一。
因此:
target engagement ≠ cognitive outcome . \boxed{
\text{target engagement}
\neq
\text{cognitive outcome}.
} target engagement = cognitive outcome .
14. A7:ADHD Cognitive/Neural Variability 值得獨立研究
2025 Nature Communications 兒童研究顯示:
temporal variability ↑ \text{temporal variability}\uparrow temporal variability ↑
與:
spatial stability ↓ \text{spatial stability}\downarrow spatial stability ↓
可以出現在 cognitive-control neural representations。
另有 2025 methylphenidate study 觀察到 whole-brain flexibility 降低及部分 behavioral variability 改善。
因此:
mean state ≠ dynamic stability . \boxed{
\text{mean state}
\neq
\text{dynamic stability}.
} mean state = dynamic stability .
15. A8:成人 ADHD 的晚診斷與晚起病不可直接等同
2025 adult ADHD review 與 2026 late-onset literature 仍將 adult-onset ADHD 視為未完全解決的 controversy。
因此:
adult diagnosis ≠ adult onset . \boxed{
\text{adult diagnosis}
\neq
\text{adult onset}.
} adult diagnosis = adult onset .
16. 外部證據層 B:有局部或間接支持的中層連結
以下列為:
Evidence Layer B . \boxed{
\text{Evidence Layer B}.
} Evidence Layer B .
它們不是純想像,但證據仍不足以支持本系列最強版本。
17. B1:Allocation Capacity 與 Allocation Stability 可分離
2025 working-memory prioritization 研究顯示,具有 ADHD symptoms 的成人仍可以依價值有效 prioritise information。
因此:
ADHD ⇏ universal allocation incapacity . \boxed{
\text{ADHD}
\not\Rightarrow
\text{universal allocation incapacity}.
} ADHD ⇒ universal allocation incapacity .
真正差異可能在:
maintenance;
switching;
interference;
update;
context coupling。
但這需要臨床與跨任務 replication。
18. B2:Hyperfocus 可被量化,但不是 ADHD 專屬核心
2024 AHQ-D validation 支持 dispositional hyperfocus 可以可靠測量。
但:
hyperfocus ≠ ADHD-specific defining symptom . \text{hyperfocus}
\neq
\text{ADHD-specific defining symptom}. hyperfocus = ADHD-specific defining symptom .
因此本系列只把 hyperfocus 作為 candidate state。
19. B3:Metacognitive Calibration 可能具有 Domain-Specific Difference
2026 college-student study 比較:
70 70 70
名正式診斷 ADHD 學生與:
70 70 70
名 matched controls,
在 verbal 與 non-verbal tasks 觀察到 performance 與 confidence calibration 差異。
但:
ADHD ≠ global metacognitive blindness . \boxed{
\text{ADHD}
\neq
\text{global metacognitive blindness}.
} ADHD = global metacognitive blindness .
其他認知域仍可保有相對完整 metacognition。
20. B4:Associative Breadth/Divergent Thinking 可能在部分 Profiles 不同
既有成人 ADHD semantic-activation 與 creativity literature 支持:
associative breadth \text{associative breadth} associative breadth
與:
divergent thinking \text{divergent thinking} divergent thinking
可能存在 profile-specific differences。
但 2026 strengths review 顯示整個 evidence base 高度異質。
因此:
networked cognition \boxed{
\text{networked cognition}
} networked cognition
仍是候選機制,不是既定 ADHD 特徵。
21. B5:Person–Environment Fit 對 Function 具有合理性
2024 employment systematic review 顯示工作環境、support、structure、autonomy 與 ADHD occupational experience 具有重要關係。
但多數 evidence 仍不足以證明:
specific trait → objective performance reversal . \text{specific trait}
\rightarrow
\text{objective performance reversal}. specific trait → objective performance reversal .
所以 performance reversal 仍需實驗。
22. 外部證據層 C:主要由本系列提出的新假說
以下全部列為:
Evidence Layer C . \boxed{
\text{Evidence Layer C}.
} Evidence Layer C .
除非未來實驗支持,不得稱為 ADHD 已知機制。
23. C1:Allocation Entropy
H ^ π = − ∑ i π i log π i log n . \widehat{\mathcal H}_{\pi}
=
-\frac{
\sum_i\pi_i\log\pi_i
}{
\log n
}. H π = − log n ∑ i π i log π i .
目前只是候選描述變量。
它不是:
EEG entropy;
thermodynamic entropy;
validated ADHD biomarker。
24. C2:Disengagement Barrier
B e x i t B^{\mathrm{exit}} B exit
表示從 dominant allocation state 退出的候選成本。
目前沒有成熟 ADHD standard measure。
25. C3:Attention Debt
D t a t t n D_t^{\mathrm{attn}} D t attn
表示長期未處理必要事件形成的累積負荷。
這是本系列新增機制。
26. C4:Cognitive Path Topology
G t = ( V t , E t , W t ) \mathcal G_t
=
(V_t,E_t,W_t) G t = ( V t , E t , W t )
及:
D P , C ‾ P , X P , E c o n v D_P,
\overline C_P,
X_P,
E_{\mathrm{conv}} D P , C P , X P , E conv
都是本系列中層模型。
不能用 brain network topology 直接證明。
27. C5:Configuration Phase Transition
「相變」只表示:
nonlinear regime shift . \text{nonlinear regime shift}. nonlinear regime shift .
它不是對真實腦熱力學相變的宣稱。
28. C6:Reversal Surface
R k = { ( T , E ) : ∂ P ∂ c k = 0 } . \mathcal R_k
=
\left\{
(T,E):
\frac{\partial P}{\partial c_k}=0
\right\}. R k = { ( T , E ) : ∂ c k ∂ P = 0 } .
這是 performance-reversal research 的新候選工具。
29. C7:Global Fit Kernel
K f i t = K ( c , q , e ) . K_{\mathrm{fit}}
=
K
\left(
\mathbf c,
\mathbf q,
\mathbf e
\right). K fit = K ( c , q , e ) .
尚未被 ADHD workplace research 直接驗證。
30. 統一 Neuromodulation Layer
第 2 篇的 canonical form:
z t = Ψ ( n t , b i , q t , e t ) . \mathbf z_t
=
\Psi
\left(
\mathbf n_t,
\mathbf b_i,
\mathbf q_t,
\mathbf e_t
\right). z t = Ψ ( n t , b i , q t , e t ) .
其中:
n t \mathbf n_t n t
不是注意力本身。
而:
z t \mathbf z_t z t
可能調節:
arousal;
reward;
salience;
vigor;
gating;
stability。
31. 統一 Allocation Layer
給定候選事件:
E t = { e 1 , … , e n } , \mathcal E_t
=
\{e_1,\ldots,e_n\}, E t = { e 1 , … , e n } ,
配置:
π t = Π ( z t , q t , m t , h t ) . \boldsymbol\pi_t
=
\Pi
\left(
\mathbf z_t,
\mathbf q_t,
\mathbf m_t,
\mathbf h_t
\right). π t = Π ( z t , q t , m t , h t ) .
其核心不是:
attention amount , \text{attention amount}, attention amount ,
而是:
who gets processing opportunity . \boxed{
\text{who gets processing opportunity}.
} who gets processing opportunity .
32. Processing Opportunity 命題
令:
o t ( e ) o_t(e) o t ( e )
為有效處理。
則:
P ( o t ( e ) = 1 ∣ π t ( e ) ) P
\left(
o_t(e)=1
\mid
\pi_t(e)
\right) P ( o t ( e ) = 1 ∣ π t ( e ) )
通常可以隨配置提高,但不必:
= 1. =1. = 1.
所以:
allocation ≠ successful processing . \boxed{
\text{allocation}
\neq
\text{successful processing}.
} allocation = successful processing .
33. Observation–Update–Action 分離
本文統一為:
m t o b s , m_t^{\mathrm{obs}}, m t obs ,
m t u p d a t e , m_t^{\mathrm{update}}, m t update ,
m t g a t e . m_t^{\mathrm{gate}}. m t gate .
因此可以:
m o b s > 0 m^{\mathrm{obs}}>0 m obs > 0
但:
m u p d a t e ≈ 0 , m^{\mathrm{update}}\approx0, m update ≈ 0 ,
也可以:
m u p d a t e > 0 m^{\mathrm{update}}>0 m update > 0
但:
m g a t e = 0. m^{\mathrm{gate}}=0. m gate = 0.
所以:
noticed ≠ updated ≠ acted . \boxed{
\text{noticed}
\neq
\text{updated}
\neq
\text{acted}.
} noticed = updated = acted .
34. 統一 Allocation Dynamics
x t = ( H ^ π , L t , T t d w e l l , ν t s w i t c h , B t e x i t , R t g o a l ) . \mathbf x_t
=
\left(
\widehat{\mathcal H}_{\pi},
L_t,
T_t^{\mathrm{dwell}},
\nu_t^{\mathrm{switch}},
B_t^{\mathrm{exit}},
R_t^{\mathrm{goal}}
\right). x t = ( H π , L t , T t dwell , ν t switch , B t exit , R t goal ) .
此層用來描述:
distribution;
dominance;
dwell;
switching;
disengagement;
goal-relative allocation。
35. Distractibility 的統一定義方向
分心不能單獨由 entropy 決定。
令:
R t g o a l = ∑ e ∈ G t g o a l π t ( e ) . R_t^{\mathrm{goal}}
=
\sum_{e\in\mathcal G_t^{\mathrm{goal}}}
\pi_t(e). R t goal = e ∈ G t goal ∑ π t ( e ) .
則 task-relative distraction:
D t t a s k = 1 − R t g o a l . D_t^{\mathrm{task}}
=
1-R_t^{\mathrm{goal}}. D t task = 1 − R t goal .
因此:
H ^ π ↓ \widehat{\mathcal H}_{\pi}\downarrow H π ↓
仍可能:
D t t a s k ↑ D_t^{\mathrm{task}}\uparrow D t task ↑
如果資源集中在錯誤目標。
36. Hyperfocus-like State 的統一候選
候選至少需要:
L t ↑ , L_t\uparrow, L t ↑ ,
T t d w e l l ↑ , T_t^{\mathrm{dwell}}\uparrow, T t dwell ↑ ,
B t e x i t ↑ . B_t^{\mathrm{exit}}\uparrow. B t exit ↑ .
因此:
low entropy alone ≠ hyperfocus . \boxed{
\text{low entropy alone}
\neq
\text{hyperfocus}.
} low entropy alone = hyperfocus .
37. 統一 Cognitive Topology Layer
G t = ( V t , E t , W t ) . \mathcal G_t
=
(V_t,E_t,W_t). G t = ( V t , E t , W t ) .
需要至少區分:
semantic , \text{semantic}, semantic ,
causal , \text{causal}, causal ,
temporal , \text{temporal}, temporal ,
analogical , \text{analogical}, analogical ,
goal . \text{goal}. goal .
多 association 不能直接叫推理。
38. Topology Quality
候選:
Θ t t o p o = ( B t a s s o c , R t a c t , D t P , C t P , X t P , E t c o n v ) . \Theta_t^{\mathrm{topo}}
=
\left(
B_t^{\mathrm{assoc}},
R_t^{\mathrm{act}},
D_t^{P},
C_t^{P},
X_t^{P},
E_t^{\mathrm{conv}}
\right). Θ t topo = ( B t assoc , R t act , D t P , C t P , X t P , E t conv ) .
有效多路徑 cognition 需要:
Expansion + Coherence + Cross-validation + Convergence . \boxed{
\text{Expansion}
+
\text{Coherence}
+
\text{Cross-validation}
+
\text{Convergence}.
} Expansion + Coherence + Cross-validation + Convergence .
39. 統一 Subjective–Metacognitive Layer
主觀狀態:
χ t . \chi_t. χ t .
元認知 performance estimate:
p ^ t . \widehat{\mathbf p}_t. p t .
客觀 performance:
p t . \mathbf p_t. p t .
因此:
χ t ≠ p ^ t ≠ p t . \boxed{
\chi_t
\neq
\widehat{\mathbf p}_t
\neq
\mathbf p_t.
} χ t = p t = p t .
40. Metacognitive Error
e t m e t a = p ^ t − p t . \mathbf e_t^{\mathrm{meta}}
=
\widehat{\mathbf p}_t
-
\mathbf p_t. e t meta = p t − p t .
這個誤差可以:
positive;
negative;
domain-specific;
state-dependent。
所以:
ADHD ≠ global overconfidence . \boxed{
\text{ADHD}
\neq
\text{global overconfidence}.
} ADHD = global overconfidence .
41. Metacognitive Feedback
p t → p ^ t → a t + 1 s t r a t e g y → p t + 1 . \mathbf p_t
\rightarrow
\widehat{\mathbf p}_t
\rightarrow
\mathbf a_{t+1}^{\mathrm{strategy}}
\rightarrow
\mathbf p_{t+1}. p t → p t → a t + 1 strategy → p t + 1 .
因此 metacognition 不只是 report,也可能參與下一步行動。
42. 統一 Life-History Layer
生命需求:
q t . \mathbf q_t. q t .
外部支架:
s t e x t . \mathbf s_t^{\mathrm{ext}}. s t ext .
補償:
k t . \mathbf k_t. k t .
有效需求—支持差:
l t = q t − r t − s t e x t − k t n e t , \boxed{
\mathbf l_t
=
\mathbf q_t
-
\mathbf r_t
-
\mathbf s_t^{\mathrm{ext}}
-
\mathbf k_t^{\mathrm{net}},
} l t = q t − r t − s t ext − k t net ,
其中 r t \mathbf r_t r t 表示可用功能資源。
43. Compensation Cost
k t n e t = k t − c t K . \mathbf k_t^{\mathrm{net}}
=
\mathbf k_t
-
\mathbf c_t^{K}. k t net = k t − c t K .
這避免:
successful output \text{successful output} successful output
被直接等同:
low burden . \text{low burden}. low burden .
44. Impairment Layer
i t = J ( p t , q t , e t , s t e x t , k t ) . \mathbf i_t
=
J
\left(
\mathbf p_t,
\mathbf q_t,
\mathbf e_t,
\mathbf s_t^{\mathrm{ext}},
\mathbf k_t
\right). i t = J ( p t , q t , e t , s t ext , k t ) .
因此:
symptom ≠ impairment . \boxed{
\text{symptom}
\neq
\text{impairment}.
} symptom = impairment .
45. Visibility Layer
v t = V ( y t , i t , Ω o , A t c a r e ) . v_t
=
V
\left(
\mathbf y_t,
\mathbf i_t,
\Omega_o,
A_t^{\mathrm{care}}
\right). v t = V ( y t , i t , Ω o , A t care ) .
其中:
Ω o \Omega_o Ω o :觀察者 sampling domain;
A c a r e A^{\mathrm{care}} A care :醫療可近性。
所以:
impairment ≠ visibility . \boxed{
\text{impairment}
\neq
\text{visibility}.
} impairment = visibility .
46. Clinical Decision Layer
D t = Γ ( y t , i t , H d e v , X t d i f f ) . \boxed{
\mathfrak D_t
=
\Gamma
\left(
\mathbf y_t,
\mathbf i_t,
H^{\mathrm{dev}},
X_t^{\mathrm{diff}}
\right).
} D t = Γ ( y t , i t , H dev , X t diff ) .
這一層故意不由:
c t \mathbf c_t c t
直接推出。
原因是正式 ADHD diagnosis 還必須處理:
developmental history;
cross-context evidence;
differential diagnosis;
functional significance。
47. Continuous Configuration Space
令:
c i , t ∈ Ω C . \mathbf c_{i,t}
\in
\Omega_C. c i , t ∈ Ω C .
群體:
ρ ( c ) . \rho(\mathbf c). ρ ( c ) .
clinical ADHD sample:
ρ D ( c ) . \rho_D(\mathbf c). ρ D ( c ) .
因此可以:
continuous density + local cluster structure . \boxed{
\text{continuous density}
+
\text{local cluster structure}.
} continuous density + local cluster structure .
48. Subthreshold 邊界
本文再次明確:
subthreshold traits ≠ hidden clinical ADHD . \boxed{
\text{subthreshold traits}
\neq
\text{hidden clinical ADHD}.
} subthreshold traits = hidden clinical ADHD .
可以:
D = 0 \mathfrak D=0 D = 0
但:
y ≠ 0 \mathbf y\neq0 y = 0
或:
i ≠ 0. \mathbf i\neq0. i = 0.
這表示可能需要支持,不表示特定 diagnosis 必然成立。
49. 統一 Context-Fit Layer
任務需求:
q T . \mathbf q_T. q T .
環境:
e . \mathbf e. e .
configuration:
c . \mathbf c. c .
fit:
K f i t = K ( c , q T , e ) . K_{\mathrm{fit}}
=
K
\left(
\mathbf c,
\mathbf q_T,
\mathbf e
\right). K fit = K ( c , q T , e ) .
performance:
p = F ( K f i t , c , q T , e ) . \mathbf p
=
F
\left(
K_{\mathrm{fit}},
\mathbf c,
\mathbf q_T,
\mathbf e
\right). p = F ( K fit , c , q T , e ) .
50. Performance Reversal
對 configuration dimension:
c k , c_k, c k ,
定義:
β k ( T , E ) = ∂ P ∂ c k . \beta_k(T,E)
=
\frac{
\partial P
}{
\partial c_k
}. β k ( T , E ) = ∂ c k ∂ P .
真正 reversal 需要:
β k ( T a , E a ) < 0 \beta_k(T_a,E_a)<0 β k ( T a , E a ) < 0
而:
β k ( T b , E b ) > 0. \beta_k(T_b,E_b)>0. β k ( T b , E b ) > 0.
這仍是 C 層候選命題,尚未被 ADHD literature 系統驗證。
51. 局部優勢與全域損害可同時成立
令:
A i ( T , E ) = P i ( T , E ) − P r e f ( T , E ) . A_i(T,E)
=
P_i(T,E)-P_{\mathrm{ref}}(T,E). A i ( T , E ) = P i ( T , E ) − P ref ( T , E ) .
可以:
A i ( T ∗ , E ∗ ) > 0 A_i(T^*,E^*)>0 A i ( T ∗ , E ∗ ) > 0
同時:
∥ i i ∥ > 0. \|\mathbf i_i\|>0. ∥ i i ∥ > 0.
所以:
local advantage ≠ absence of disorder-related impairment . \boxed{
\text{local advantage}
\neq
\text{absence of disorder-related impairment}.
} local advantage = absence of disorder-related impairment .
52. 一條統一因果候選鏈
將所有層整合:
n t → z t → π t → x t → G t → m t → p t → y t . \boxed{
\mathbf n_t
\rightarrow
\mathbf z_t
\rightarrow
\boldsymbol\pi_t
\rightarrow
\mathbf x_t
\rightarrow
\mathcal G_t
\rightarrow
\mathbf m_t
\rightarrow
\mathbf p_t
\rightarrow
\mathbf y_t.
} n t → z t → π t → x t → G t → m t → p t → y t .
但同時存在:
q t , e t , s t e x t , k t \boxed{
\mathbf q_t,
\mathbf e_t,
\mathbf s_t^{\mathrm{ext}},
\mathbf k_t
} q t , e t , s t ext , k t
對多層的調節。
53. 另一條主觀/元認知迴路
z t → χ t → p ^ t → a t + 1 s t r a t e g y → p t + 1 . \boxed{
\mathbf z_t
\rightarrow
\chi_t
\rightarrow
\widehat{\mathbf p}_t
\rightarrow
\mathbf a_{t+1}^{\mathrm{strategy}}
\rightarrow
\mathbf p_{t+1}.
} z t → χ t → p t → a t + 1 strategy → p t + 1 .
所以 subjectivity 不是附錄,而可能參與動態閉環。
54. 生命史/診斷鏈
p t + q t + e t + s t e x t + k t → i t → v t → assessment → D t . \boxed{
\mathbf p_t
+
\mathbf q_t
+
\mathbf e_t
+
\mathbf s_t^{\mathrm{ext}}
+
\mathbf k_t
\rightarrow
\mathbf i_t
\rightarrow
v_t
\rightarrow
\text{assessment}
\rightarrow
\mathfrak D_t.
} p t + q t + e t + s t ext + k t → i t → v t → assessment → D t .
這條鏈解釋:
diagnosis timing \text{diagnosis timing} diagnosis timing
為何不能被當作:
biological onset time . \text{biological onset time}. biological onset time .
55. IDCT-ADHD 十八項統一可證偽命題
U-H1:非單因子命題
不存在一個單一 scalar:
a a a
能穩定、完整解釋 ADHD-related symptom、impairment、task variation 與 treatment response。
如果單一 scalar model 在跨資料集持續勝過 IDCT,則本理論應簡化。
U-H2:動態增量命題
加入:
Var t ( c ) \operatorname{Var}_t(\mathbf c) Var t ( c )
與:
P ( c t + 1 ∣ c t ) P(\mathbf c_{t+1}\mid\mathbf c_t) P ( c t + 1 ∣ c t )
應在部分 outcomes 提供超越平均值的預測力。
U-H3:Allocation–Maintenance 分離命題
能初始 prioritise 不代表能長時間穩定維持配置。
U-H4:Observation–Update–Action 分離命題
noticed ≠ updated ≠ acted . \text{noticed}
\neq
\text{updated}
\neq
\text{acted}. noticed = updated = acted .
若三者無法可靠操作化區分,SAD 部分應被簡化。
U-H5:State-Multiplicity 命題
distractibility、mind wandering、adaptive focus 與 hyperfocus-like lock-in 不應被單一 attention amount 完整描述。
U-H6:Neuromodulatory Mediation 命題
部分 pharmacological effects 透過:
z t \mathbf z_t z t
狀態中介,而不是直接:
drug → P . \text{drug}
\rightarrow
P. drug → P .
U-H7:Subjective–Objective 分離命題
存在:
Δ χ > 0 \Delta\chi>0 Δ χ > 0
而:
Δ p ≈ 0 \Delta\mathbf p\approx0 Δ p ≈ 0
的可重現狀態。
U-H8:Metacognitive Feedback 命題
p ^ t − p t \widehat{\mathbf p}_t-\mathbf p_t p t − p t
能預測後續 strategy change。
U-H9:Associative Breadth 非充分命題
B a s s o c ↑ B^{\mathrm{assoc}}\uparrow B assoc ↑
只有在:
C P , X P , E c o n v C^P,
X^P,
E^{\mathrm{conv}} C P , X P , E conv
足夠時才可能提高有效推理。
U-H10:Demand–Support 命題
q t − s t e x t − k t \mathbf q_t
-
\mathbf s_t^{\mathrm{ext}}
-
\mathbf k_t q t − s t ext − k t
的變化應能解釋部分 within-person impairment change。
U-H11:Visibility 非等同命題
功能損害與 clinical visibility 可在不同時間尺度變化。
U-H12:Dimensional-plus-Cluster 命題
高維 ADHD-related space 同時容許 continuous variation 與 local clusters。
U-H13:Subthreshold Non-Identity 命題
below-threshold traits 不等於 clinical ADHD。
U-H14:Context-Fit 命題
K f i t K_{\mathrm{fit}} K fit
對 objective task performance 應有增量預測力。
U-H15:Performance-Reversal 命題
至少某些 configuration dimensions 在嚴格 matched tasks 中可能出現 effect-sign reversal。
U-H16:Local–Global 分離命題
局部 performance advantage 不等於整體 impairment 消失。
U-H17:Transdiagnostic Overlap 命題
部分 IDCT variables 應跨 ADHD、anxiety、sleep problems 等共享,因此不能被直接當 ADHD-specific biomarker。
U-H18:Out-of-Sample Superiority 命題
整套模型只有在:
P I D C T , o u t > P s i m p l e r , o u t P_{\mathrm{IDCT,out}}
>
P_{\mathrm{simpler,out}} P IDCT , out > P simpler , out
時才值得保留。
這是最重要的一條。
56. 什麼叫「更簡單模型」?
至少應比較:
M1:Binary Diagnosis Model
P = F ( D ) . P
=
F(\mathfrak D). P = F ( D ) .
M2:Symptom Total Model
P = F ( S t o t a l ) . P
=
F(S_{\mathrm{total}}). P = F ( S total ) .
M3:Executive Function Model
P = F ( E F ) . P
=
F(EF). P = F ( E F ) .
M4:Single Dimensional Liability Model
P = F ( z A D H D ) . P
=
F(z_{\mathrm{ADHD}}). P = F ( z ADHD ) .
M5:Static High-Dimensional Model
P = F ( c ) . P
=
F(\mathbf c). P = F ( c ) .
M6:Dynamic Configuration Model
P = F ( c t , Δ c t , T , E ) . P
=
F
\left(
\mathbf c_t,
\Delta\mathbf c_t,
T,
E
\right). P = F ( c t , Δ c t , T , E ) .
只有:
M 6 M6 M 6
在外部驗證中有穩定增量,動態理論才有存在價值。
57. Complexity Penalty
模型越複雜,越容易 overfit。
因此 model selection 必須包括:
prediction gain − λ complexity . \text{prediction gain}
-
\lambda
\text{complexity}. prediction gain − λ complexity .
可以使用:
held-out likelihood;
cross-validation;
information criteria;
preregistered primary metric。
而不是:
模型看起來比較完整。
58. 研究綱領 Phase A:Construct Validation
第一階段完全不需要 neuroimaging。
先建立:
allocation stability;
disengagement;
cognitive-path diversity;
subjective clarity;
compensation cost;
context fit;
的可靠 measurement。
要求:
Reliability > 0 \operatorname{Reliability}>0 Reliability > 0
且跨 session 可重現。
如果 construct 本身測不穩,後面全部停止。
59. Phase B:Within-Person Dynamic Experiments
同一人跨不同:
novelty;
reward;
structure;
interruption;
delay;
task type;
重複測量。
核心:
within-person transition \boxed{
\text{within-person transition}
} within-person transition
而不是只做:
ADHD mean − control mean . \text{ADHD mean}
-
\text{control mean}. ADHD mean − control mean .
60. Phase C:Clinical Replication
Phase A、B 的 construct 必須在:
clinically diagnosed ADHD;
matched controls;
relevant clinical comparison groups;
重現。
尤其需要:
sleep disorders;
anxiety;
depression;
autism;
作 differential comparison。
61. Phase D:Multimodal Mechanism
只有在行為 construct 成立後,再加入:
EEG;
fMRI;
PET;
pupillometry;
actigraphy;
digital phenotyping。
避免:
brain-first storytelling . \boxed{
\text{brain-first storytelling}.
} brain-first storytelling .
62. Phase E:Longitudinal Development
追蹤:
c t , q t , e t , s t e x t , k t , i t , v t . \mathbf c_t,
\mathbf q_t,
\mathbf e_t,
\mathbf s_t^{\mathrm{ext}},
\mathbf k_t,
\mathbf i_t,
v_t. c t , q t , e t , s t ext , k t , i t , v t .
才能研究:
late diagnosis;
compensation collapse;
developmental trajectories;
context transition。
63. Phase F:Prediction and Translation
最後才測:
P ( future impairment ∣ c ) , P
\left(
\text{future impairment}
\mid
\mathbf c
\right), P ( future impairment ∣ c ) ,
P ( treatment response ∣ c ) . P
\left(
\text{treatment response}
\mid
\mathbf c
\right). P ( treatment response ∣ c ) .
只有 out-of-sample predictive gain 足夠,才討論 precision support。
64. 不允許跳過的順序
本文提出:
Construct → Reliability → Replication → Prediction → Clinical utility . \boxed{
\text{Construct}
\rightarrow
\text{Reliability}
\rightarrow
\text{Replication}
\rightarrow
\text{Prediction}
\rightarrow
\text{Clinical utility}.
} Construct → Reliability → Replication → Prediction → Clinical utility .
不能:
nice equation → clinical tool . \text{nice equation}
\rightarrow
\text{clinical tool}. nice equation → clinical tool .
65. Measurement Principle 1:平均值與變異都要保存
至少測:
E [ X ] , \mathbb E[X], E [ X ] ,
Var ( X ) , \operatorname{Var}(X), Var ( X ) ,
Autocorr ( X t ) , \operatorname{Autocorr}(X_t), Autocorr ( X t ) ,
P ( X t + 1 ∣ X t ) . P(X_{t+1}\mid X_t). P ( X t + 1 ∣ X t ) .
ADHD dynamic research 不應只報平均 performance。
66. Measurement Principle 2:主觀與客觀不能互換
同步測:
χ t , \chi_t, χ t ,
p ^ t , \widehat{\mathbf p}_t, p t ,
p t . \mathbf p_t. p t .
主觀資料不是低級資料,但它測的是不同東西。
67. Measurement Principle 3:Task 必須明確參數化
不要只寫:
attention task。
應寫:
q T = ( sustained demand , novelty , reward delay , switching , interruption , … ) . \mathbf q_T
=
\left(
\text{sustained demand},
\text{novelty},
\text{reward delay},
\text{switching},
\text{interruption},
\ldots
\right). q T = ( sustained demand , novelty , reward delay , switching , interruption , … ) .
否則 context dependence 無法重現。
68. Measurement Principle 4:Environment 必須被記錄
至少保留:
noise;
social presence;
autonomy;
structure;
feedback;
time pressure;
support。
這些不能全部當 noise。
69. Measurement Principle 5:跨診斷對照
如果某變量:
X X X
在:
ADHD;
anxiety;
sleep deprivation;
都相同改變,
則:
X X X
可能是 general dysregulation marker,而非 ADHD-specific marker。
這不是失敗,而是分類修正。
70. Measurement Principle 6:Clinical Diagnosis 不可由研究變量循環定義
若先用:
X X X
定義 ADHD subgroup,
再說:
X X X
證明 ADHD subgroup 不同,
即 circularity。
因此 clinical diagnosis 與 experimental constructs 必須保持可追溯分離。
71. Measurement Principle 7:多評分者與觀察域
保存:
Y s e l f , Y p a r e n t , Y t e a c h e r , Y p a r t n e r . Y^{\mathrm{self}},
Y^{\mathrm{parent}},
Y^{\mathrm{teacher}},
Y^{\mathrm{partner}}. Y self , Y parent , Y teacher , Y partner .
不要只求一個平均值就丟失 disagreement。
disagreement 本身可能包含 context information。
72. Measurement Principle 8:Treatment State 必須記錄
至少區分:
stimulant-naïve;
currently medicated;
washout/not washout;
treatment history;
dose/formulation。
否則不同研究不應被直接比較。
73. 成人與兒童必須分開驗證
不能:
child ADHD mechanism ⇒ adult ADHD mechanism \text{child ADHD mechanism}
\Rightarrow
\text{adult ADHD mechanism} child ADHD mechanism ⇒ adult ADHD mechanism
自動成立。
需要:
developmental invariance test . \boxed{
\text{developmental invariance test}.
} developmental invariance test .
74. Cross-Cultural Validity
ADHD diagnosis、observer threshold、school demand 與 work structure 受文化與制度影響。
因此:
Φ Taiwan \Phi_{\text{Taiwan}} Φ Taiwan
不必:
= Φ US . =
\Phi_{\text{US}}. = Φ US .
研究需檢驗 measurement invariance。
75. Digital Phenotyping 的角色
2026 成人 ADHD digital-health review 顯示數位工具研究快速增加。
IDCT 可利用:
task timing;
switching;
calendar behavior;
phone interaction;
EMA;
做高頻狀態測量。
但:
digital trace ≠ diagnosis . \boxed{
\text{digital trace}
\neq
\text{diagnosis}.
} digital trace = diagnosis .
76. 生態瞬時評估
EMA 特別適合:
c t \mathbf c_t c t
與:
e t \mathbf e_t e t
的 repeated measurements。
例如每次記錄:
current task;
interest;
stress;
clarity;
distraction;
urge to switch;
time awareness;
competing obligations。
77. 模型必須保存零結果
如果某構念:
X X X
在 preregistered study:
Δ X ≈ 0 , \Delta X\approx0, Δ X ≈ 0 ,
不能只把它移出故事。
需要保留:
null evidence . \text{null evidence}. null evidence .
本理論的演化應為:
T 0 → Test → T 1 , T_0
\rightarrow
\text{Test}
\rightarrow
T_1, T 0 → Test → T 1 ,
而不是只累積正向發現。
78. Pre-registration Rule
所有關鍵 experiment 應提前固定:
primary hypothesis;
primary outcome;
exclusion criteria;
model comparison;
stopping rule;
multiplicity correction。
避免 model flexibility 讓任何結果都可以「解釋」。
79. Out-of-Sample Rule
任何新的 IDCT construct 要進入核心模型,至少要求:
P out > P baseline,out . P_{\text{out}}
>
P_{\text{baseline,out}}. P out > P baseline,out .
如果只在 training sample 漂亮:
discard or downgrade . \boxed{
\text{discard or downgrade}.
} discard or downgrade .
80. Falsification Level F0:Measurement Failure
如果:
Reliability ( X ) ≈ 0 , \operatorname{Reliability}(X)\approx0, Reliability ( X ) ≈ 0 ,
直接淘汰 X X X 。
不需要進入神經機制討論。
81. F1:Dissociation Failure
若:
Q , P ^ , P Q,
\widehat P,
P Q , P , P
實際上高度不可分,
則 SMOSH 簡化。
若:
S a l i e n c e , A c t i v a t i o n , A l l o c a t i o n Salience,
Activation,
Allocation S a l i e n ce , A c t i v a t i o n , A l l oc a t i o n
不可區分,
則 SAD 簡化。
82. F2:Dynamic Failure
若:
Var t ( X ) \operatorname{Var}_t(X) Var t ( X )
與 transition features 沒有增量價值,
則 dynamic component 應移除。
83. F3:Topology Failure
若 graph-derived:
D P , C P , X P , E c o n v D_P,
C_P,X_P,E_{\mathrm{conv}} D P , C P , X P , E conv
不優於普通 executive/divergent measures,
則 NCTH 淘汰。
84. F4:Context-Reversal Failure
若嚴格 matched-task crossover 中:
β k ( T , E ) \beta_k(T,E) β k ( T , E )
從不換號,
performance-reversal hypothesis 應大幅削弱。
85. F5:Longitudinal Failure
若:
Δ q , Δ s e x t , Δ k \Delta q,
\Delta s^{\mathrm{ext}},
\Delta k Δ q , Δ s ext , Δ k
無法預測 within-person impairment change,
DCVH 簡化。
86. F6:Clinical Increment Failure
如果整體 configuration model 在控制:
diagnosis;
symptom severity;
executive function;
IQ;
comorbidity;
後:
Δ R 2 ≈ 0 , \Delta R^2\approx0, Δ R 2 ≈ 0 ,
則 IDCT 不具有臨床研究增量價值。
87. 最強反證:簡單模型一直贏
如果跨多資料集:
P ( M s i m p l e , o u t ) > P ( M I D C T , o u t ) , P(M_{\mathrm{simple,out}})
>
P(M_{\mathrm{IDCT,out}}), P ( M simple , out ) > P ( M IDCT , out ) ,
則最合理結論不是:
資料還不夠理解高深理論。
而是:
use the simpler model . \boxed{
\text{use the simpler model}.
} use the simpler model .
88. IDCT 成功的最低條件
本理論不需要所有模組都成立。
最低成功條件是:
至少數個 dynamic constructs 可可靠測量;
它們跨情境可重現;
可預測 within-person performance changes;
在 external data 有增量;
能明確指出何時失效。
89. IDCT 的強成功條件
更強版本需要:
configuration → future impairment \boxed{
\text{configuration}
\rightarrow
\text{future impairment}
} configuration → future impairment
與:
configuration → treatment-relevant outcome \boxed{
\text{configuration}
\rightarrow
\text{treatment-relevant outcome}
} configuration → treatment-relevant outcome
具有可重複 external prediction。
在此之前不應宣稱 precision psychiatry。
90. Clinical Translation Guardrail
任何未來 clinical tool 至少需要:
prospective validation;
external replication;
calibration;
fairness;
harm analysis;
clinical utility analysis;
clinician oversight。
不能只報:
A U C . AUC. A U C .
91. Imaging Biomarker Guardrail
2026 biotype study 是重要研究進展。
但目前不能:
brain scan → routine ADHD diagnosis . \boxed{
\text{brain scan}
\rightarrow
\text{routine ADHD diagnosis}.
} brain scan → routine ADHD diagnosis .
研究 stratification 與 clinical biomarker 仍是不同階段。
92. Genetics Guardrail
2025 genetics 支持 continuous liability。
但:
polygenic score ≠ individual ADHD diagnosis . \boxed{
\text{polygenic score}
\neq
\text{individual ADHD diagnosis}.
} polygenic score = individual ADHD diagnosis .
IDCT 不使用 genetics 作單人 diagnostic oracle。
93. Strengths Guardrail
2026 strengths scoping review 支持 strengths field 值得研究。
但:
ADHD strength ≠ universal advantage . \boxed{
\text{ADHD strength}
\neq
\text{universal advantage}.
} ADHD strength = universal advantage .
每個 strength 都需要:
objective measure;
context;
comparator;
cost;
generalization。
94. Masking Guardrail
ADHD camouflaging/masking 目前仍需要更成熟 construct validation。
所以:
late diagnosis ≠ masking by default . \boxed{
\text{late diagnosis}
\neq
\text{masking by default}.
} late diagnosis = masking by default .
95. Transdiagnostic Guardrail
如果 IDCT 最後發現:
c \mathbf c c
中的很多變量同樣適用於 anxiety、autism、sleep disorder,
可能意味:
IDCT becomes a general cognitive-regulation framework . \boxed{
\text{IDCT becomes a general cognitive-regulation framework}.
} IDCT becomes a general cognitive-regulation framework .
這不必視為失敗。
但其 ADHD-specific claim 必須縮小。
96. 理論可能最後「離開 ADHD」
這是一個重要可能。
若:
Φ \Phi Φ
對不同 neurodevelopmental/psychiatric groups 都有效,
則最合理名稱可能從:
ADHD Dynamic Configuration Theory \text{ADHD Dynamic Configuration Theory} ADHD Dynamic Configuration Theory
變成:
General Dynamic Cognitive Configuration Theory . \text{General Dynamic Cognitive Configuration Theory}. General Dynamic Cognitive Configuration Theory .
科學模型不應被原始命名綁架。
97. 最小可實作研究原型
若只做第一個真正實驗,不需要十層全部上。
可以先測:
allocation stability × task novelty × reward timing . \boxed{
\text{allocation stability}
\times
\text{task novelty}
\times
\text{reward timing}.
} allocation stability × task novelty × reward timing .
每個人重複完成:
low novelty/delayed reward;
low novelty/immediate reward;
high novelty/delayed reward;
high novelty/immediate reward。
測:
a c c u r a c y , R T , R T V , s w i t c h e s , c o n f i d e n c e , s u b j e c t i v e c l a r i t y . accuracy,
RT,
RTV,
switches,
confidence,
subjective clarity. a cc u r a cy , R T , R T V , s w i t c h es , co n f i d e n ce , s u bj ec t i v ec l a r i t y .
這已經可以同時測 Paper 3、4、6、9 的一部分命題。
98. 第二個最小實驗:Sparse vs Scaffolded Completion
比較:
I s p a r s e I_{\mathrm{sparse}} I sparse
與:
I s c a f f o l d e d . I_{\mathrm{scaffolded}}. I scaffolded .
測:
a c c u r a c y , c o m p l e t i o n t i m e , p a t h d i v e r s i t y , f a l s e p a t h , c o n f i d e n c e . accuracy,
completion\ time,
path\ diversity,
false\ path,
confidence. a cc u r a cy , co m pl e t i o n t im e , p a t h d i v er s i t y , f a l se p a t h , co n f i d e n ce .
若 graph measures 沒有可靠增量,Paper 5 可以直接降級。
99. 第三個最小實驗:Life-Transition EMA
追蹤進入:
前後。
測:
q t , s t e x t , k t , i t , v t . q_t,
s_t^{\mathrm{ext}},
k_t,
i_t,
v_t. q t , s t ext , k t , i t , v t .
直接測 Paper 7,而不是依賴回憶。
100. 最後的研究順序
本文建議:
先行為 → 再動態 → 再跨診斷 → 再神經 → 再縱向 → 最後臨床預測 . \boxed{
\text{先行為}
\rightarrow
\text{再動態}
\rightarrow
\text{再跨診斷}
\rightarrow
\text{再神經}
\rightarrow
\text{再縱向}
\rightarrow
\text{最後臨床預測}.
} 先行為 → 再動態 → 再跨診斷 → 再神經 → 再縱向 → 最後臨床預測 .
而不是從:
brain scan \text{brain scan} brain scan
直接跳到:
new ADHD subtype . \text{new ADHD subtype}. new ADHD subtype .
101. 本系列最需要避免的十五種錯誤
把數學形式當證據;
把 ADHD 當單一機制;
把 dopamine 當 attention;
把 subjective enhancement 當 performance enhancement;
把 hyperfocus 當 ADHD 專屬核心;
把 creativity 當 ADHD 超能力;
把 brain network 當 cognitive graph;
把 adult diagnosis 當 adult onset;
把 compensation 當 absence of impairment;
把 masking 當已成熟 ADHD construct;
把 continuous traits 當 everyone-has-ADHD;
把 subthreshold 當 hidden diagnosis;
把 biotype 當新臨床 subtype;
把 local advantage 當 global advantage;
把 complex model 當 better model。
102. 十篇系列的最小共同命題
如果把所有內容壓到最小,只剩:
ADHD-related functioning is likely to be heterogeneous, dynamic, multidimensional, and context-sensitive . \boxed{
\text{ADHD-related functioning is likely to be
heterogeneous, dynamic, multidimensional,
and context-sensitive}.
} ADHD-related functioning is likely to be heterogeneous, dynamic, multidimensional, and context-sensitive .
這句本身與現代 literature 相容。
但本系列真正需要被驗證的是更強版本:
specific dynamic configuration variables provide incremental predictive value . \boxed{
\text{specific dynamic configuration variables
provide incremental predictive value}.
} specific dynamic configuration variables provide incremental predictive value .
103. 封頂總式
最終候選系統:
c t + 1 = Φ ( c t , q t , e t , s t e x t , k t , u t ) + ε t , p t = F ( c t , q t , e t ) , χ t = H ( c t , e t , x t e x p e c t a n c y ) , p ^ t = M ( p t , χ t , f t f e e d b a c k ) , i t = J ( p t , q t , e t , s t e x t , k t ) , v t = V ( y t , i t , Ω o , A t c a r e ) , D t = Γ ( y t , i t , H d e v , X t d i f f ) . \boxed{
\begin{aligned}
\mathbf c_{t+1}
&=
\Phi
\left(
\mathbf c_t,
\mathbf q_t,
\mathbf e_t,
\mathbf s_t^{\mathrm{ext}},
\mathbf k_t,
\mathbf u_t
\right)
+
\boldsymbol\varepsilon_t,
\\
\mathbf p_t
&=
F
\left(
\mathbf c_t,
\mathbf q_t,
\mathbf e_t
\right),
\\
\chi_t
&=
H
\left(
\mathbf c_t,
\mathbf e_t,
\mathbf x_t^{\mathrm{expectancy}}
\right),
\\
\widehat{\mathbf p}_t
&=
M
\left(
\mathbf p_t,
\chi_t,
\mathbf f_t^{\mathrm{feedback}}
\right),
\\
\mathbf i_t
&=
J
\left(
\mathbf p_t,
\mathbf q_t,
\mathbf e_t,
\mathbf s_t^{\mathrm{ext}},
\mathbf k_t
\right),
\\
v_t
&=
V
\left(
\mathbf y_t,
\mathbf i_t,
\Omega_o,
A_t^{\mathrm{care}}
\right),
\\
\mathfrak D_t
&=
\Gamma
\left(
\mathbf y_t,
\mathbf i_t,
H^{\mathrm{dev}},
X_t^{\mathrm{diff}}
\right).
\end{aligned}
} c t + 1 p t χ t p t i t v t D t = Φ ( c t , q t , e t , s t ext , k t , u t ) + ε t , = F ( c t , q t , e t ) , = H ( c t , e t , x t expectancy ) , = M ( p t , χ t , f t feedback ) , = J ( p t , q t , e t , s t ext , k t ) , = V ( y t , i t , Ω o , A t care ) , = Γ ( y t , i t , H dev , X t diff ) .
這不是 ADHD 的已證實方程。
它是整個系列的 research-program compression。
104. 這套模型如果是錯的,應該怎麼死?
它不應該透過不斷增加自由參數來逃避反證。
最清楚的死亡條件是:
simple models predict just as well or better . \boxed{
\text{simple models predict just as well or better}.
} simple models predict just as well or better .
其次是:
new constructs cannot be measured reliably . \boxed{
\text{new constructs cannot be measured reliably}.
} new constructs cannot be measured reliably .
再其次:
dynamic/contextual predictions fail to replicate . \boxed{
\text{dynamic/contextual predictions fail to replicate}.
} dynamic / contextual predictions fail to replicate .
如果這三件事發生,IDCT 應被歸檔,而不是繼續擴張。
105. 這套模型如果是真的,最先會看到什麼?
最先不會是:
發現 ADHD 新腦區。
而更可能是:
同一人跨任務 performance variance 很有結構;
配置穩定性比單次平均值更能預測錯誤;
subjective clarity 與 objective performance 可穩定脫鉤;
context fit 可預測 within-person performance change;
某些 profile 在 matched tasks 出現可重現 sign reversal;
這些變量在 external cohorts 有增量價值。
106. 系列的理論地位
本系列最合理的目前地位:
pre-empirical integrative computational framework . \boxed{
\text{pre-empirical integrative computational framework}.
} pre-empirical integrative computational framework .
不是:
medical theory confirmed , \text{medical theory confirmed}, medical theory confirmed ,
也不是:
clinical diagnostic framework . \text{clinical diagnostic framework}. clinical diagnostic framework .
107. 系列封頂後的下一步不是再寫更多理論篇
若此系列要繼續,優先順序應從:
theory generation \text{theory generation} theory generation
切換為:
measurement design → pilot protocol → preregistered validation . \boxed{
\text{measurement design}
\rightarrow
\text{pilot protocol}
\rightarrow
\text{preregistered validation}.
} measurement design → pilot protocol → preregistered validation .
也就是停止繼續堆疊新名詞。
108. 結論
本系列從一個很簡單的懷疑開始:
ADHD ≠ just too little attention . \text{ADHD}
\neq
\text{just too little attention}. ADHD = just too little attention .
十篇之後,這個懷疑被拆成一套更嚴格、也更容易被打假的研究架構。
目前的外部研究支持:
ADHD 是高度異質的神經發展障礙;
symptoms 與 genetic liability 具有明顯連續性;
continuous variation 與 local biotypes 可以共存;
stimulant mechanisms 涉及 dopamine、norepinephrine、arousal、reward 與 network dynamics,而非單一 attention gain;
trial-to-trial neural stability 與 network flexibility 值得作為獨立變量;
adult ADHD diagnosis、impairment、late onset、emotional dysregulation、objective measures 仍有大量重要未解問題。
但這些證據尚未證明 :
allocation entropy;
attention debt;
cognitive graph topology;
disengagement barrier;
reversal surface;
global fit kernel;
是 ADHD 的真實機制。
這些都是本系列交給未來實驗的東西。
因此封頂篇不以:
我們已經解釋 ADHD。
作結。
而以:
Can these constructs survive measurement, replication, prediction, and falsification? \boxed{
\text{Can these constructs survive measurement,
replication, prediction, and falsification?}
} Can these constructs survive measurement, replication, prediction, and falsification?
作結。
如果不能:
discard them . \boxed{
\text{discard them}.
} discard them .
如果能:
keep only the parts that survive . \boxed{
\text{keep only the parts that survive}.
} keep only the parts that survive .
這才是本系列最終的科學立場。
參考文獻
World Health Organization. Clinical descriptions and diagnostic requirements for ICD-11 mental, behavioural and neurodevelopmental disorders (CDDR). WHO. Current ICD-11 framework accessed 2026-08-17.
World Health Organization. ICD-11 2026 Release. WHO, 16 February 2026.
Cortese, S., Bellgrove, M. A., Brikell, I., Franke, B., Goodman, D. W., Hartman, C. A., et al. Attention-deficit/hyperactivity disorder (ADHD) in adults: evidence base, uncertainties and controversies. World Psychiatry . 2025;24(3):347–371. DOI: 10.1002/wps.21374.
Pan, N., Long, Y., Qin, K., et al. Mapping ADHD Heterogeneity and Biotypes by Topological Deviations in Morphometric Similarity Networks. JAMA Psychiatry . 2026;83(5):478–490. DOI: 10.1001/jamapsychiatry.2026.0001.
van der Laan, C. M., Ip, H. F., Schipper, M., et al. Genome-wide association meta-analysis of childhood ADHD symptoms and diagnosis identifies new loci and potential effector genes. Nature Genetics . 2025;57:2427–2435. DOI: 10.1038/s41588-025-02295-y.
Kay, B. P., Wheelock, M. D., Siegel, J. S., et al. Stimulant medications affect arousal and reward, not attention networks. Cell . 2025. DOI: 10.1016/j.cell.2025.11.039.
Oya, M., Matsuoka, K., Kubota, M., et al. Effects of Extended-Release Methylphenidate on Dopamine and Norepinephrine Transporters in Adults With Attention-Deficit/Hyperactivity Disorder: A Longitudinal Dual-Tracer PET Study. Psychiatry and Clinical Neurosciences . 2026;80(1):48–54. DOI: 10.1111/pcn.13911.
Gao, Z., et al. Reduced temporal and spatial stability of neural activity patterns predict cognitive control deficits in children with ADHD. Nature Communications . 2025;16:2346. DOI: 10.1038/s41467-025-57685-x.
Nugiel, T., et al. Methylphenidate stabilizes dynamic brain network organization during tasks probing attention and reward processing in stimulant-naïve children with ADHD. Translational Psychiatry . 2025;15:488. DOI: 10.1038/s41398-025-03694-9.
Atkinson, A. L., Pinheiro Sanchez, B., Warburton, M., Allmark, H., & Allen, R. J. The Ability to Direct Attention in Working Memory Is Not Impaired in Adults With Symptoms of ADHD. Journal of Attention Disorders . 2025;29(9):684–705. DOI: 10.1177/10870547251330039.
Hupfeld, K. E., Osborne, J. B., Tran, Q. T., Hyatt, H. W., Abagis, T. R., et al. Validation of the dispositional adult hyperfocus questionnaire (AHQ-D). Scientific Reports . 2024;14:19460. DOI: 10.1038/s41598-024-70028-y.
Elsholz, L., et al. ADHD symptomatology and metacognitive monitoring insights for college student support. Current Psychology . 2026. DOI: 10.1007/s12144-026-09164-9.
Kang, S., Fu, Z., Li, Q., Yang, L., & Cao, Q. Adult-diagnosed and childhood-diagnosed attention deficit/hyperactivity disorder: cognitive and environmental contributions to symptom severity across different age of diagnosis. Frontiers in Psychiatry . 2026;17:1782999. DOI: 10.3389/fpsyt.2026.1782999.
Bayard, S., Madiouni, C., Radiguer, F., Roulin, M., & Henrard, S. Late-Onset ADHD symptoms in the general population: A scoping review of longitudinal trajectories in population-based cohorts. The European Journal of Psychiatry . 2026;40(1):100337. DOI: 10.1016/j.ejpsy.2025.100337.
Rafael, R. B., Jia, H., Rouel, M., Wootton, B. M., & Mitchison, D. Attention Deficit/Hyperactivity Disorder (ADHD)-Related Strengths in Adults: A Scoping Review. Journal of Attention Disorders . 2026. DOI: 10.1177/10870547261425737.
Arildskov, T. W., Thomsen, P. H., Sonuga-Barke, E. J. S., Lambek, R., Østergaard, S. D., & Virring, A. Is Attention-Deficit/Hyperactivity Disorder (ADHD) a Dimension or a Category? What Does the Relationship Between ADHD Traits and Psychosocial Quality of Life Tell Us? Journal of Attention Disorders . 2024;28(7):1035–1044. DOI: 10.1177/10870547231222228.
Faraone, S. V., et al. Attention-deficit/hyperactivity disorder. Nature Reviews Disease Primers . 2024;10:11. DOI: 10.1038/s41572-024-00495-0.
Schork, A. J., et al. Polygenic profiles define aspects of clinical heterogeneity in attention deficit hyperactivity disorder. Nature Genetics . 2024;56:234–244. DOI: 10.1038/s41588-023-01593-7.
文獻使用聲明
本文僅使用上述研究建立截至 2026-08-17 的外部實證邊界。
本文提出的 IDCT-ADHD、全域 configuration state c t \mathbf c_t c t 、allocation entropy、disengagement barrier、attention debt、cognitive path topology、global fit kernel、reversal surface 與十八項統一命題,均為本系列理論構件,不應被誤認為上述研究作者的原始結論。
本篇也不把兒童、成人、healthy-control pharmacology、genetics、neuroimaging、self-report、clinical review 與 workplace research 視為可直接相加的單一證據池。不同研究只支持不同層級的局部背景。
系列封頂狀態
系列狀態: 10/10 完成本篇狀態: v1.0 封頂理論稿新增原始臨床/人體數據: 無醫學用途: 無下一階段: Measurement Design/Pilot Protocol/Preregistered Validation