重量移向上游：ISSB、GRI 與 TNFD 如何改變揭露前證據判斷

2026 年 9 月 24 日，ISSB 公開討論人力資本研究的工作人員文件 AP4，涉及投資人需要的資訊、衡量方法、商業模式及司法管轄區脈絡；文件不代表 ISSB 董事會已制定人力資本準則。9 月 22 日，GRI 宣布以食品飲料產業試行較短且更聚焦的 Sector Standard，仍保留辨識可能重大衝擊與導向既有 Topic Standards 的功能。9 月 21 日，TNFD 的 2026 Status Report 記錄自然揭露進入重複週期以及受訪者進行 LEAP 評估的情況。三者是不同制度與不同證據母體，不構成單一全球義務。EIS-021 原作者用 The Weight Moves Upstream 說明：當報導表面更聚焦，脈絡、方法、重大性與追溯性可能成為更關鍵的揭露前判斷。Evidence Conservation Principle 是原作者的治理比喻，Context-Bound Evidence 是其分析概念，都不是 ISSB、GRI、TNFD 或歐盟的正式術語。本台灣版的問題是：當人力、產業、自然資訊流向不同報告與客戶決策，來源的哪些條件必須隨資料同行？

On 24 September 2026 the ISSB discussed a staff paper on Human Capital research and proposed next steps. The paper is for discussion, not a Board decision or a new Standard. On 22 September GRI announced a streamlined format for Sector Standards, starting with a Food and Beverages pilot that retains likely material topics and sector impacts. On 21 September TNFD published its 2026 Status Report on recurring disclosure and LEAP assessment. These are different institutions, authorities and research populations. The controlled EIS-021 source calls the shared analytical pattern The Weight Moves Upstream. A more focused reporting surface may concentrate responsibility for context, methodology, materiality and traceability before disclosure. Its Evidence Conservation Principle is a governance metaphor, and Context-Bound Evidence is an analytical concept, not terminology officially adopted by the ISSB, GRI, TNFD or the EU. This Taiwan edition asks which facts about origin, place, period and decision use must accompany evidence as workforce, sector and nature information moves into different reports and customer requests.

ISSB 文件列出人力資訊缺少決策有用性、完整性、一致性與清楚衡量方法等研究觀察，並提議繼續研究 IFRS S1 的作用、產業、地理、司法管轄區及精確的投資人需求。員工離職率若未保存分母、僱用型態、資料期間及報告邊界，跨企業比較可能失真；承包商職安資料若未明列納入條件，也不能直接與正職員工合併。這是資料治理例子，不是 AP4 新發布的要求。GRI 的精簡試點旨在更快提供有用的產業入口，並非取消企業辨識實際重大衝擊的責任。食品加工的農業供應、用水與勞動問題與半導體或紡織產業不同；同一產業也會因地點與營運活動而異。GRI 1 與 GRI 3 的報導基礎及重大主題程序仍須區別於尚在開發的食品飲料試點。TNFD 報告則提供時間面向：它記錄逾 1,000 個組織於 56 個國家或地區有不同程度對齊揭露，超過 100 個組織進入第三年度，76% 受訪者已進行或正進行 LEAP。這些數據屬 TNFD 的調查與市場觀察，不代表所有台灣企業皆已採用。

The ISSB staff paper identifies deficiencies in decision-useful workforce information, consistency, completeness and transparent methods, while proposing further research on business model, geography, jurisdiction, investor needs and IFRS S1 implementation. A turnover figure without its denominator, employment category, period and entity boundary may not be comparable. Contractor safety information cannot be pooled with employee data if populations differ. These are evidence-design examples, not new AP4 requirements. GRI's Food and Beverages pilot aims to deliver shorter sector guidance while continuing to identify likely impacts and make use of Topic Standards. Agriculture, water and labour issues in food production are not identical to semiconductor or textile supply chains, and an organisation must still judge its actual material impacts. GRI 1 and GRI 3 provide established reporting and material-topic context, while the pilot remains under development. TNFD reports more than 1,000 organisations across 56 countries or areas with some aligned disclosure and more than 100 in a third annual cycle; 76% of surveyed respondents had undertaken or were undertaking LEAP. These are TNFD's reported findings, not a claim about all Taiwan companies.

第一個風險是指標脫離定義。跨境集團可以彙總員工資料，卻在各地對外包、派遣與價值鏈工作者使用不同分類。數字若沒有分母、來源、期間、法人及更正紀錄，投資人或確信人員無法重建。第二個風險是把較短的準則誤解成較少的實際衝擊。台灣食品、紡織或電子供應商可能同時收到不同的 GRI、ESRS 及客戶問卷。原始紀錄可以共用，但重大性判斷、契約範圍和報導目標不能自動共用；未取得資料不等於議題不重大。第三個風險是自然資訊失去地點及時間。用水、棲地與生態依存隨廠區及方法而變，平均數可能掩蓋高風險據點。LEAP 是辨識及評估方法，不是保證所有風險已被排除的證書。多年度揭露更需要保留前後兩期的坐標、範圍、方法版本與判斷，以辨別真實變化與測量變化。第四個風險是混淆權威：ISSB 研究提議、GRI 試點、TNFD 自願框架、歐盟 ESRS 正式法規和台灣 IFRS 導入各有不同地位，不能藉由共同的『精簡』語彙互相替代。

The first failure mode is a metric without its definition. Multinational groups can aggregate workforce figures while regions classify contractors, temporary staff and value-chain workers differently. Without denominator, source system, period, entity and correction history, investors or assurance practitioners cannot reconstruct the number. The second is treating shorter guidance as fewer real-world impacts. A Taiwan food, textile or electronics supplier may receive distinct GRI, ESRS and buyer questionnaires. Original records can sometimes be reused, but materiality decisions, contractual boundaries and reporting objectives cannot be copied without reassessment. Missing supplier data does not prove a topic immaterial. The third is nature information without location or time. Water stress, land use and ecosystem dependency change with site and method; an average can conceal a high-risk facility. LEAP is an assessment approach, not a certificate that all risks have disappeared. Repeated disclosure calls for prior and current coordinates, boundaries, methods and decisions so readers can distinguish operational change from measurement change. A fourth failure is authority confusion: ISSB research, a GRI pilot, TNFD voluntary recommendations, EU ESRS law and Taiwan IFRS adoption have different legal and institutional statuses.

SNN.TW 建議建立四層可回查的證據物件。來源層保存產生機構、原始檔、時間戳、版本、取得方式及權限；脈絡層標示法人、廠區、地點、活動、工作者分類、生態系、價值鏈位置與期間；方法層保留分子分母、儀器、估算、替代資料、限制及不確定性；決策層記錄納入、排除、彙總、拆分的原因、負責人、核准時點與下一次覆核條件。這是編輯台提出的實務設計，不是三個機構共同發布的技術規格。人力資料可先從一項離職率或職安紀錄進行，辨識正職與承攬的母體及薪資、事故來源。GRI 產業資料可建立活動、可能衝擊、Topic Standard、原始憑證及實際重大性決議之間的映射，避免把『可能重大』變成企業的必然結論。自然資料須保存地理參照、LEAP 階段、測量期間、現場覆核日期及前期變更。跨框架共用一份來源時，仍要分別建立 IFRS、ESRS 或其他使用情境的決策紀錄，保留目的、範圍、判斷人與限制。

SNN.TW proposes a four-layer evidence object. The identity layer retains producer, source file, timestamp, version, retrieval route and access rights. The context layer records legal entity, site, geography, activity, worker category or ecosystem, value-chain position and period. The method layer preserves numerator, denominator, instrument, estimate, proxy, limitation and uncertainty. The decision layer records why a record was included, excluded, aggregated or disaggregated, the accountable reviewer, approval time and review trigger. This is an editorial design proposal, not a specification issued jointly by the three institutions. A human-capital pilot can start with turnover or safety data, checking employee and contractor populations against payroll and incident systems. A GRI sector mapping can connect activity and likely impact to a Topic Standard, original evidence and an actual materiality decision, avoiding the assumption that every likely topic is material to every enterprise. Nature records require geospatial references, LEAP stage, measurement period, site review date and changes since the last cycle. When one source is reused across frameworks, each IFRS, ESRS or other use needs a separate decision trail for purpose, reporting boundary, authority and limitations.

證據鏈分成五組。ISSB 的 2026 AP4 與較早的研究更新支持研究進展，IFRS S1 支持投資人導向的揭露架構；它們不證明已出現新的人力資本強制條文。GRI 9 月 22 日公告、食品飲料試點，以及 GRI 1、GRI 3 支持產業準則方向與重大議題判斷，試點尚未成為生效的最終產業準則。TNFD 2026 報告與發布聲明支持其調查數據及揭露週期；LEAP 指引提供自然評估概念脈絡，但其網頁日期精度不足，未列入硬性 15 筆門檻，也不能替台灣個別企業提供確信。歐盟正式 ESRS 修訂及 EFRAG 技術建議只作精簡背景，須區分正式法規與先前草案。台灣證交所的公告支持本地 IFRS 永續揭露準則的分階段導入及實務資源，不能證明歐洲買方已修改特定契約。原稿 EIS-021 與 EIA-016 是受控系列關聯脈絡，不列入 15 筆獨立外部來源。每筆外部紀錄均需明列標題、機構、日期、查核時點、角色及支撐範圍；不得將作者的 Evidence Conservation Principle 包裝成任何機構採納的正式術語。

The evidence chain has five tiers. The ISSB's 2026 AP4 and prior research updates support the status of the project, while IFRS S1 supplies investor-oriented reporting context; they do not demonstrate that a new Human Capital rule is in force. GRI's September update, pilot materials, GRI 1 and GRI 3 support the shorter sector format and material-topic assessment, not a completed effective pilot Standard. TNFD's 2026 report and announcement support attributed findings and reporting cycles. LEAP guidance provides conceptual assessment context, but its online date lacks day precision and is not among the 15 counted records. None is an assurance opinion on a Taiwan company's controls. The final EU ESRS amendment and EFRAG's earlier technical advice illustrate a separate simplification history; a draft is not the adopted regulation. Taiwan Stock Exchange documents establish local IFRS adoption and practical resources, not a specific European buyer contract. EIS-021 and EIA-016 are controlled related publications and are excluded from the independent external source threshold. Every counted record specifies the title, institution, publication date, verification time, role and bounded claim. The source author's analytical metaphor is never attributed to a standards body.

台灣的第一條傳導路徑是跨境集團與資本市場。台灣上市櫃公司自 2026 會計年度起依實收資本額分階段接軌 IFRS 永續揭露準則；擁有歐洲子公司或客戶的集團還可能提供人力、供應鏈或自然資料給另一報告主體。該資料是否進入 ESRS 報告，仍由歐盟報導主體依自身邊界和重大性判斷。第二條路徑是採購契約與客戶問卷。全球品牌可能要求台灣食品、紡織、電子與半導體供應商提供更聚焦但更具活動、廠區、期間與來源脈絡的資料。這是可檢驗的市場傳導推論，並無第一方文件證明所有買方已改寫合約。第三條路徑是金融與確信，銀行、投資人及查核者可能要求重建人力與自然資訊如何支持風險和揭露結論。較短的海外文本不能證明台灣公司的證據成本已下降；一次自然揭露也不能證明多年資料可比。應待實際契約、問卷、報告版本和確信調整出現，才把可能的產業影響改寫成已發生事實。

Taiwan's first transmission channel is capital markets and multinational groups. Taiwan-listed companies phase in IFRS Sustainability Disclosure Standards from fiscal 2026 by paid-in capital, while a group with an EU subsidiary or buyer may supply workforce, value-chain or nature evidence to another reporting entity. Whether the record belongs in an ESRS report remains a decision under that entity's own boundary and materiality rules. The second channel is procurement terms and buyer questionnaires. Global brands may request more focused activity, site, period and method context from Taiwan food, textile, electronics and semiconductor suppliers. That is a testable editorial inference; the institutional documents do not prove that all buyers have amended their terms. The third channel is finance and assurance, where lenders, investors and practitioners may need to reconstruct how workforce or nature records support risk and disclosure conclusions. Shorter overseas guidance does not prove that Taiwan evidence costs have fallen. A first nature report does not prove three years of comparable site data. Actual clauses, questionnaires, report versions and assurance adjustments are needed to establish observed market outcomes.

SNN editorial interpretation：全球制度變動的事實邊界，是 ISSB 於 2026 年 9 月公開討論人力資本研究工作人員提議、GRI 推出較精簡的產業準則試點，以及 TNFD 2026 報告呈現重複自然揭露週期，三者未建立單一台灣法定申報義務。傳導機制是跨國集團資料請求、歐洲與全球品牌的供應鏈採購問卷、投融資盡職調查及確信程序，把人力分類、產業衝擊與自然地點脈絡帶進台灣市場。台灣食品、紡織、電子、半導體與金融業可能面對較聚焦、但更依賴來源、方法與邊界的資料要求。此影響屬編輯推論，須由實際客戶條款、問卷及確信個案驗證。於下次跨境客戶問卷回覆前，台灣供應商的採購主管與永續主管應核對客戶條款、法人及廠區、活動清單、原始憑證、報導期間和欄位版本，客戶條款更動後重新驗證。於 2026 年度 IFRS 揭露定稿前，上市櫃公司財務與人資主管應核對工作者母體、分母、薪資及事故來源、估計方法、管理階層核准紀錄，並於年報核准前保存差異說明。於每次自然風險評估週期結束前，台灣製造業廠務與風險主管應抽查據點坐標、水及土地資料、LEAP 評估版本、現場覆核日期與前期變更，重大廠區或方法更動時立即重作比較。銀行風險主管或確信團隊在引用結論前應核對證據身分、責任人、核准時間及限制。於每次授信覆核前，台灣銀行風險主管應核對客戶申報資料版本、廠區座標與管理階層核准紀錄，並記錄查核時間。揭露前證據基礎設施（Pre-Disclosure Evidence Infrastructure）保存來源、脈絡、方法及判斷的可追索路徑，並區分每一框架的權威；這是 SNN.TW 編輯觀點，不是任何機構的法定指令。

SNN editorial interpretation: the factual boundary of global institutional change is the ISSB's September 2026 Human Capital staff proposal for further research, GRI's streamlined Sector Standards pilot and TNFD's 2026 Status Report on recurring nature disclosure. They have not created a single Taiwan statutory filing obligation. The transmission mechanism runs through multinational group requests, European and global buyer procurement questionnaires, credit and investor diligence, and assurance evidence requests. These channels could bring more focused demands for worker classification, sector impacts and site-specific nature information to Taiwan's food, textile, electronics, semiconductor and financial markets. The downstream effect is an editorial inference to be tested with actual clauses, questionnaires and assurance cases. Before the next cross-border questionnaire, a Taiwan supplier's procurement and sustainability owners should inspect buyer terms, entity and facility boundary, activity list, source file, reporting period and field version, then revalidate after a term changes. Before FY2026 IFRS disclosures are finalised, a Taiwan listed company's finance and human-resources owners should check worker population, denominator, payroll and incident sources, estimate method and management approval, preserving discrepancies before annual-report approval. Before each nature-risk cycle closes, a Taiwan manufacturer's facility and risk owners should sample site coordinates, water and land records, LEAP version, review date and prior-period changes, repeating the comparison after a significant site or method change. A lender's risk owner or assurance team should reconstruct source identity, accountable reviewer, approval time and limitations before citing a conclusion. Pre-Disclosure Evidence Infrastructure preserves the route from source to context, method and judgement while keeping framework authority distinct. This is SNN.TW editorial analysis, not an endorsed institutional implementation directive.

下一輪檢驗應測量能否重建判斷，而不只計算標準頁數。若 ISSB 推進人力資本工作，應查其正式決議與諮詢文件如何處理產業、地理及工作者分類，不能將現有 AP4 當成已生效規範。GRI 食品飲料試點完成後，應查較短文本能否引導企業辨識實際重大衝擊，而非生成制式清單。TNFD 後續年度可比較同一據點的範圍、方法與決策用途，區分實質改善和資料可得性變化。台灣可抽選跨國集團人力指標、食品或紡織供應商產業衝擊判斷、製造廠區自然風險結論，由未參與編製的覆核者從公開主張回查原檔、期間、方法、框架及核准歷史，記錄缺件率與補件時間。若精簡同時降低資料蒐集與覆核成本，而可追溯性維持，原先對判斷密度的市場假設也應修正。

The next test should measure reconstructability, not pages. If the ISSB advances Human Capital work, review formal Board decisions and consultation materials for industry, jurisdiction and worker definitions; AP4 currently supports a description of research only. Once the GRI Food and Beverages pilot is complete, assess whether its shorter text still leads organisations to identify actual material impacts instead of reproducing a default list. Across later TNFD cycles, compare the same site's boundary, method and decision use, distinguishing real progress from improved data availability. Taiwan could sample a multinational workforce metric, a food or textile supplier's impact decision and a manufacturer's nature-risk conclusion. A reviewer outside the original preparation team can trace a published claim back to the source, period, method, governing framework and approval history, recording missing links and time to repair. If simplification reduces both collection and review effort while traceability remains intact, the Taiwan market inference should be revised.