ESAT 實務篇・中篇｜語意落點如何形成：SCA × EAX × CAX × TAX 的四軸語意偏向

本篇是《ESAT 實務篇》系列的第二篇。

上篇處理 BF／AF 與五大模組，說明語意如何被拆解；

本篇則聚焦在四大軸（SCA／EAX／CAX／TAX），
說明這些被拆解出的語意片段，如何在軸上形成落點與偏向。

下篇才會進一步處理 Do-Path／Pay-Path 的結果描述，
因此本篇刻意停留在「語意落點層級」，不直接進入責任結論。

一｜五大模組如何影響四大軸？（語意層級，而非運算層級）

在上篇中，我們已經看到 ESAT 五大模組，如何把人類的自然語句，萃取成「可被閱讀、可被比對的工程語意片段」。

但這些語意片段本身還不是「落點」，更不能直接等同於工程語意責任的歸屬。語意責任在 ESAT 中是一個多維度的結果，必須經由多軸綜合參考，才有機會合理逼近案件的真實語意。

在 ESAT 中，「落點」永遠是透過四大軸來呈現語意的偏向，此四軸分別是：

  • 範圍清晰度軸（Scope Clarity Axis，SCA）
  • 證據力軸（Evidence Availability Axis，EAX）
  • 控制能力軸（Control Ability Axis，CAX）
  • 時間調節軸（Time Adjustment Axis，TAX）

透過這四條軸，我們才能具體看到語意的落點，以及這個落點對語意歸屬產生的方向性。簡單說，可以把兩個層級理解為：

  • 五大模組提供一個合理的「語意共識平台」，負責從自然語句中萃取出貼合工程實務、能被讀懂與比對的語意片段
  • 四大軸則是在這個共識平台之上，承接多維度的綜合結果落點，呈現出「語意歸屬的偏向」

提醒：四軸落點不是「算出來的數字」，而是「語意歸屬偏向」的呈現方式

因此：

  • Boundary Module 不等於 SCA
  • Evidence Module 不等於 EAX
  • Control Ability Module 不等於 CAX
  • Time Condition Module 不等於 TAX

但那只是方便記憶的對照，不是實際的運算關係。

真實情況永遠是：

每一條軸的語意落點
是由多個模組的語意共同交織而成
而非由單一模組主導

ESAT 的公開語意層只會呈現「落點的語意表現」，不會呈現「落點是怎麼生成的」。四軸的偏向不是責任結論，而是語句本身對語意歸屬的影響力。

二｜四大軸是什麼？——先把方向感講清楚

在 ESAT 裡，四大軸是用來承接五大模組萃取結果的「語意落點承載面」。

  • SCA ｜範圍清晰度軸
  • EAX ｜證據力軸
  • CAX ｜控制能力軸
  • TAX ｜時間調節軸

四條軸都有三階：H／M／L。
它們不是分數，也不是等級，而是語意歸屬偏向的一種「刻度描述」。

1.範圍清晰度軸（Scope Clarity Axis，SCA）

1.1 軸的本體語意：語句是否呈現「足夠清楚的規格」

SCA 的高低，只用來描述語句本身在「工程規格描述」上的清晰程度，而非法律或責任結論：

  • SCA-H｜語句清晰、高度明確
    規格、條件、界線足夠可辨識，不易產生誤解。
  • SCA-M｜語句普通、略有模糊
    大致知道方向，但仍留有彈性與解讀空間。
  • SCA-L｜語句模糊、界線不明確
    容易讓雙方產生不同理解。

此處只處理語句本身的狀態，不做對錯判斷，也不直接推導工程責任。

1.2 語句的提出者 → 語意偏向如何產生

  • 若 SCA-H 且由供給端（SDP-P）提出
    → 代表供給端提供的是一段「高度清晰的規格」
    → 語意偏向：供給端握有更強的規格語意主張能力
  • 若 SCA-H 且由需求端（SDP-C）提出
    → 表示需求端提出的語句十分明確
    → 語意偏向：需求端握有更清楚的需求語意位置
  • 同理，若 SCA-L 由某方提出
    → 代表該方的語句模糊
    → 語意偏向：該方語意主張力減弱

情境一：

需求端：整套廚具五金配件我不懂，你是專業就交給你搭配。

多數人直覺上會把這句話讀成：
廚具規格的範圍清晰度，應該偏向 SCA-L 的語氣。
需求端沒具體指定

情境二：

需求端：廚具五金配件，我就挑 A方案了( A 方案有明確列表)

但是不在此範圍外的所有 AF（附加功能），
一般認知上，都不會被視為供給端的責任，
其範圍清晰度多半落在 SCA-L。

2.證據力軸（Evidence Availability Axis，EAX）

2.1軸的本體語意：語句在工程中能否被追溯

EAX 用來描述證據在工程爭議中「是否足以被追溯與觀察」：

  • EAX-H｜高證據度
    有書面、有圖說、有影像、有可查資料，是穩定且不變動的證據。
  • EAX-M｜中證據度
    有部分紀錄，但不足以完全界定語意，或者部分證據指向有分歧。
  • EAX-L｜低證據度
    僅屬口述、無法追溯、無從佐證。

此軸只描述證據狀態。

2.2語句提出者 → 語意偏向

比方說同一句話：

  • 若 EAX-H 且資料由供給端提出
    → 語意偏向：供給端的語句擁有更高的證據支撐力。
  • 若 EAX-L 且語句由需求端提出
    → 語意偏向：需求端在此語句上的證據力較弱。

語意偏向≠責任，只是語句位置的呈現。

情境一：

需求端：整套廚具五金配件我不懂，你是專業就交給你搭配。

沒有確實的要求，也沒有任何的文字與書面證據留下，
所以基本上屬於 EAX-L

情境二：

需求端：廚具五金配件，我就挑 A方案了( A 方案有明確列表)

有書面證據可以證明，在此語境之中，也沒有更新的內容推翻
所以屬於 EAX-H

EAX 不判斷誰說真話

3.控制能力軸（Control Ability Axis，CAX）

3.1 軸的本體語意：事件當下，誰能控制結果？

CAX 描述的是「事件當下，誰實際握有操作與調整權」，而非能力高低或專業優劣：

  • CAX-H｜完全控制過程與結果
  • CAX-M｜可能受到影響而無法完全控制
  • CAX-L｜徹底和過程與結果無關

這是事件本質，不帶評價。

3.2 語句提出者 → 語意偏向

當該語句與控制性一致時，語意位置就會偏向語句方：

例如：

  • 若語句呈現 CAX-H，且是供給端提出
    → 表示供給端本身就能控制結果
    → 語意偏向供給端在控制語意上更靠前。
  • 若語句呈現 CAX-H，且是需求端提出
    → 語意偏向需求端。

這裡只呈現語句位置，不做後續推論。

情境一：

需求端：整套廚具五金配件我不懂，你是專業就交給你搭配。

這句話對需求端而言，就是屬於 CAX-L
而對供給端而言，卻是屬於 CAX-H

需求端：廚具五金配件，我就挑 A方案了( A 方案有明確列表)

這句話對需求端而言，就是屬於 CAX-H，因為已經確定是他要的。
而對供給端而言，卻是屬於 CAX-L，因為不能在做任何自主變更。

CAX 不處理誰「應該」怎麼做，只描述「實際上誰比較能動手改變」。

4 TAX｜時間調節軸（Time Adjustment Axis）

4.1 軸的本體語意：在語句產生的當下，時間對工程結果與語意判讀的影響力

  • TAX-H｜時間對事情的結果有產生高度影響力
  • TAX-M｜時間對事情的結果有產生中度影響力
  • TAX-L｜時間對事情的結果有產生低度影響力

4.2 影響結果程度 → 語意偏向

當時間對事情的影響力不同，語意位置就會偏向不同方：

例如：

  • 若語句提出的當下，工程本身尚未經過長時間使用或老化，直覺上會被讀成 TAX-L
    → 表示對剛才完成工程的供給端，偏向性較高
  • 若語句提出的當下，工程本身已歷經長時間使用、老化與多次介入，直覺上會被讀成 TAX-H
    → 表示對正在使用的需求端，偏向性較高

這裡只呈現語句位置，不做後續推論。

情境一：

工程剛完工一週，木地板就大面積翹起來。

工程完成時間短，時間上無法對工程造成甚麼太大影響
因此多數人直覺會讀成 TAX-L
那才完成工程的供給端，語意歸屬上排序就比較靠前

情境二：

同一塊木地板，使用十年後因為長期滲水而變形，
且這十年間有多次隨機維修、重物撞擊。

工程完成十年了，且也使用了十年了
因此多數人直覺會讀成 TAX-H
時間與後期介入都很重，
語意自然會往需求端或第三方方向偏。

三｜語意如何從模組進入四軸（不展示運算，只展示結果）

以下示範一個不完整、但足以讓人理解「五大模組落點於四大軸」語意流動的例子：

自然情境：

需求端：我上次講的時候你說 OK，結果今天做完跟我說沒包含？

五大模組

  • 規格邊界模組：
    需求端認定為： IS（隱含標準）
    供給端認定為：OS（範圍外）
  • 證據模組：
    需求端：E5（僅口述、無證據）
    供給端：E5（僅口述、無證據）
  • 說明交換模組：
    需求端：C1：完全理解一致
    供給端：C4（弱交換）
  • 控制能力模組：
    需求端：供給端可控制
    供給端：供給端可控制
  • 時間條件模組：
    需求端：無影響
    供給端：無影響

四軸落點（僅呈現語意偏向，不進入判斷）

  • SCA（範圍清晰度）
    需求端：SCA-L（語句為隱含期待，界線模糊）
    供給端：SCA-M（界定為 OS，但未事先明確說明）
  • EAX（證據力）
    需求端：EAX-L（僅口述）
    供給端：EAX-L（僅口述）
  • CAX（控制能力）
    需求端：CAX-L（無法控制施工內容是否包含）
    供給端：CAX-H（可控制工程內容）
  • TAX（時間影響）
    需求端：TAX-L（時間差短，影響微弱）
    供給端：TAX-L（同上）

→ 上述落點不是計算結果，而是語意在四軸上的「偏向呈現」。

四｜五大模塊與四軸到底有甚麼不同？

兩者之間存在關聯，但負責的功能不同，也許你會看到很類似的東西，譬如同樣是講時間，Time Condition Module模組，與 TAX 軸有甚麼差別？

Time Condition Module 模組，
是用來從人類自然語言中，萃取與時間相關的語意片段，
並將其歸類成貼合工程實務的時間語意標籤。

而 TAX 軸則是承接這些時間語意標籤，
描述「時間因素對語意落點偏向的影響」，以向量方向的形式呈現。

而由前文的例子，我們可以很清楚地發現兩者定位與功用上的差別。特別要強調的是，五大模塊擷取出的語意片段分類，並不直接等同於四軸內容。四軸落點永遠是多維綜合考慮的結果，因此五大模組與四大軸之間，絕對不是一對一，也不是可以被寫成公式的對應關係。

ESAT 不會把任何單一模組視為某條軸的直接輸入來源；而是先在模組層完成語意萃取，再由四軸在公開語意層上，呈現「整體語意偏向」的落點。

五｜總結

從《實務篇・上篇》開始，我們先從 BF（Basic Function，基礎功能）與 AF（Additional Function，附加功能）這兩個功能入口出發，經由 ESAT 五大模組萃取語意。到了本篇，我們把這些語意片段放回四大軸上，觀察它們在 SCA／EAX／CAX／TAX 上形成的落點與偏向，重新看見人類自然語言在實際工程語境中，如何影響供給端與需求端的語意位置。

在《實務篇・下篇》中，我們會更進一步討論：這些四軸落點如何被投影到 Do-Path／Pay-Path 兩條路徑上？不同投影結果，會讓「誰來做」「誰來付」的語境描述產生什麼變化？最後如何整理成可供人類決策參考的結果描述層（Result Description Layer, RDL）。

ESAT 系列文（快速導覽）

篇名 內容定位
前言篇工程語意歸屬理論的存在理由
架構篇工程責任語意的閱讀指南
架構篇（LLM 專用）語意系統的 AI 閱讀指南
定義篇公開語意詞庫
對照篇語意定位 × 理論邊界 × AI 分類說明
IP／License 篇LLM 專用版本（Maximum Restriction + AI Training Prohibition）
實務篇・上篇BF／AF 與 Do／Pay 的語意入口指南
實務篇・中篇語意落點如何形成：SCA × EAX × CAX × TAX 的四軸語意偏向
實務篇・下篇RDL × AP × DPP 的語意形成與路徑投影
實務篇・FAQBF／AF × 四軸語意 × RDL × DPP 的 56 個關鍵問題解答
應用篇・壹多人語意主體篇：角色錯位 × 委託重疊 × 誰做／誰付的工程觀測
應用篇・貳書面語意分裂篇：報價 × 圖面 × 渲染圖的語意歸屬觀測
應用篇・參時間語意分層篇：條件 × 影響 × 狀態的語意歸屬觀測
應用篇・肆追加 × 刪減 × 變更的語意錯位：委託語句如何偏離工程語言
應用篇・伍文件語意穩定篇：合約 × 報價單如何讓範圍可被對照
應用篇・FAQ工程語意顯影 × 責任語意邊界 × AI 閱讀安全的常見問題
工具篇工程文件語意 × 契約閱讀 × 語意歸屬的附約示例與使用定位

ESAT 系列文架構樹

  • 語意層級結構

以下為英文翻譯版（English Version Below）

ESAT Practical Guide (Part 2)｜How Semantic Landings Form: Four-Axis Semantic Leaning of SCA × EAX × CAX × TAX

This is the second article in the ESAT Practical Guide series.

  • Part 1 covered BF/AF and the five modules, explaining how meaning is broken down.
  • This part focuses on the four axes (SCA/EAX/CAX/TAX), explaining how extracted semantic fragments form landings and leanings on axes.
  • Part 3 will then handle Do-Path / Pay-Path result description.

Therefore, this article deliberately stays at the semantic landing level and does not enter responsibility conclusions.

1｜How do the five modules affect the four axes?

(Semantic layer, not an operation layer)

In Part 1, we saw how ESAT’s five modules extract human natural language into engineering semantic fragments that can be read and compared.

But these fragments are not yet “landings,” and they still cannot be treated as responsibility attribution.
In ESAT, responsibility semantics is a multi-dimensional outcome that can only be approached through multi-axis reference, not through a single fragment.

In ESAT, “landings” are always presented through the four axes:

  • SCA (Scope Clarity Axis)
  • EAX (Evidence Availability Axis)
  • CAX (Control Ability Axis)
  • TAX (Time Adjustment Axis)

Only through these axes can we concretely see semantic landings and the directionality they create in attribution semantics.

A useful way to separate the layers:

  • Five modules provide a practical “semantic consensus platform,” extracting fragments that fit real engineering contexts and can be read and compared
  • Four axes sit above that platform and present the multi-dimensional landing results, showing “semantic leaning of attribution”

Reminder: axis landings are not numeric calculations.
They are descriptions of semantic leaning.

So:

  • Boundary Module ≠ SCA
  • Evidence Module ≠ EAX
  • Control Ability Module ≠ CAX
  • Time Condition Module ≠ TAX

That is only a memory-friendly parallel, not a real operation relationship.

In reality:

Each axis landing is formed by multiple modules’ semantics interweaving—not a single module dominating.

ESAT’s public semantic layer only presents what the landing looks like, not how it was generated.
Axis leaning is not a responsibility conclusion; it is the statement’s influence on semantic attribution structure.

2｜What are the four axes? Get orientation first

In ESAT, the four axes are the “landing surfaces” that receive module extraction results:

  • SCA｜Scope Clarity Axis
  • EAX｜Evidence Availability Axis
  • CAX｜Control Ability Axis
  • TAX｜Time Adjustment Axis

All axes have three discrete levels: H / M / L.
They are not scores and not rankings.
They are simply a scale description of semantic leaning.

2.1 SCA｜Scope Clarity Axis

2.1.1 Core semantics: is the statement “clear enough as a specification”?

SCA only describes how clear the statement is as an engineering specification—no legal or responsibility conclusion:

  • SCA-H｜Clear, highly explicit
    Conditions/boundaries are identifiable and hard to misread
  • SCA-M｜Normal, slightly vague
    Direction is known, but room for interpretation remains
  • SCA-L｜Vague, boundary unclear
    Different roles can easily form different understandings

This section describes statement state only—no right/wrong judgment, no responsibility derivation.

2.1.2 Statement author → how semantic leaning forms
  • If SCA-H and stated by SDP-P
    → Provider supplies a highly clear spec
    → leaning: Provider holds stronger spec-claim position
  • If SCA-H and stated by SDP-C
    → Client statement is highly explicit
    → leaning: Client holds clearer demand-position
  • If SCA-L stated by a side
    → their statement is vague
    → leaning: that side’s claim position weakens

Scenario A
Client: “I don’t understand kitchen hardware; you’re the professional, just match it.”
Most people intuitively read this as SCA-L (no concrete designation).

Scenario B
Client: “I choose Option A for kitchen hardware (Option A has a clear list).”
This leans toward higher clarity than Scenario A.

Also note: AF items outside that explicit list are often not treated as Provider responsibility by default, and their clarity tends to fall toward SCA-L unless explicitly defined.

2.2 EAX｜Evidence Availability Axis

2.2.1 Core semantics: can the statement be traced in an engineering context?

EAX describes whether evidence can be traced and observed:

  • EAX-H｜High evidence
    Written docs/drawings/images/checkable sources; stable, non-shifting evidence
  • EAX-M｜Medium evidence
    Partial records, not enough to fully define meaning, or evidence points diverge
  • EAX-L｜Low evidence
    Verbal only, not traceable, cannot be supported

This axis describes evidence state only.

2.2.2 Statement author → semantic leaning

Same sentence:

  • If EAX-H and evidence is provided by Provider
    → leaning: Provider has stronger evidentiary support for that statement
  • If EAX-L and claim is raised by Client
    → leaning: Client’s evidentiary position is weaker in that claim

Leaning ≠ responsibility. It is only statement position.

Scenario A
Client: “I don’t understand kitchen hardware; you’re the professional, just match it.”
No concrete requirements, no written trace → typically EAX-L.

Scenario B
Client: “I choose Option A (Option A has a clear list).”
A written list exists and no later update overturns it → typically EAX-H.

EAX does not judge who is truthful.

2.3 CAX｜Control Ability Axis

2.3.1 Core semantics: at the time of the event, who can control the result?

CAX describes who actually holds operational/adjustment control at the time—no “skill” or “professional superiority” judgment:

  • CAX-H｜Full control of process and outcome
  • CAX-M｜Partial control or control affected by other factors
  • CAX-L｜Essentially unrelated to process/outcome

This is event nature, not moral evaluation.

2.3.2 Statement author → semantic leaning

When the statement aligns with control reality, semantic position leans toward the stating side:

  • If statement reflects CAX-H, and Provider states it
    → leaning: Provider is forward in control semantics
  • If statement reflects CAX-H, and Client states it
    → leaning: Client is forward in control semantics

Only statement position is presented; no downstream inference.

Scenario A
Client: “I don’t understand kitchen hardware; you decide.”
For Client: CAX-L
For Provider: CAX-H

Scenario B
Client: “I choose Option A (clear list).”
For Client: CAX-H (they decide)
For Provider: CAX-L (cannot autonomously change it)

CAX does not discuss what someone “should” do—only who “can actually change” outcomes.

2.4 TAX｜Time Adjustment Axis

2.4.1 Core semantics: at the time the statement is made, how strong is time’s influence on result and interpretation?
  • TAX-H｜Time has high influence on outcome/interpretation
  • TAX-M｜Time has medium influence
  • TAX-L｜Time has low influence
2.4.2 Influence strength → semantic leaning

Different time influence strengths shift semantic position:

  • If the project is newly completed and has not undergone long use/aging → often read as TAX-L
    → semantic position tends to be closer to the Provider side (completion-state context)
  • If long-term use/aging and repeated interventions exist → often read as TAX-H
    → semantic position tends to drift toward Client or Third side (usage/intervention context)

Again, this is only position presentation, not inference.

Scenario A
One week after completion, wooden flooring warps widely.
Short time cannot reasonably explain much → often read as TAX-L.

Scenario B
Same flooring deforms after 10 years due to long-term leakage, with many random repairs and impacts.
Time and later interventions are heavy → often read as TAX-H.

3｜How meaning moves from modules into four axes

(No operation shown, only outcome shown)

Below is an intentionally incomplete example that is still enough to illustrate “module semantics → axis landings” flow.

Natural situation:
Client: “Last time you said OK, but today after it’s done you tell me it wasn’t included?”

Five modules

Boundary:

  • Client reads it as IS
  • Provider reads it as OS

Evidence:

  • Client: E5 (verbal only)
  • Provider: E5 (verbal only)

CEX:

  • Client: C1 (believes fully aligned)
  • Provider: C4 (weak exchange)

Control:

  • Client: Provider-controlled
  • Provider: Provider-controlled

Time:

  • Client: No impact
  • Provider: No impact

Four-axis landings

(leaning presentation only, no judgment)

  • SCA:
    • Client: SCA-L (implicit expectation; boundary vague)
    • Provider: SCA-M (claims OS, but not clarified beforehand)
  • EAX:
    • Client: EAX-L (verbal only)
    • Provider: EAX-L (verbal only)
  • CAX:
    • Client: CAX-L (cannot control whether it’s included in施工內容)
    • Provider: CAX-H (can control inclusion/exclusion)
  • TAX:
    • Client: TAX-L (short time gap, low influence)
    • Provider: TAX-L (same)

→ These are not calculated results.
They are a semantic leaning display on the four axes.

4｜What is the difference between five modules and four axes?

They are related, but they do different jobs.

Example: Time Condition Module vs TAX axis.

  • Time Condition Module extracts time-related semantic fragments from natural language and classifies them into practical time semantic labels.
  • TAX axis receives those time labels and presents “how time factors adjust semantic leaning of the landing,” as a directional leaning description.

Key point: module fragment categories do not equal axis landings.
Axis landings are always multi-dimensional composite results.

So the relationship is never:

  • one-to-one
  • formula-based
  • algorithmically describable in the public layer

ESAT does not treat any single module as a direct input source of an axis.
Modules extract; axes present the integrated semantic leaning in the public layer.

5｜Summary

Starting from Part 1, we entered through BF/AF and extracted semantic fragments via the five modules.
In this article, we placed those fragments back onto the four axes and observed how they form landings and leanings on SCA/EAX/CAX/TAX, allowing us to re-see how natural language in real engineering contexts shifts semantic positions between Provider and Client.

In Part 3, we will move further:
How do these axis landings project into Do-Path / Pay-Path?
How do different projection outcomes change the description of “who does” and “who pays”?
And how are these finally organized into the Result Description Layer (RDL) for human decision reference?

Tags: , , , , , , , , , ,