{ "article_id": "IDAEO-BLOG-2026-002", "version": "v2.2", "as_of": "2026-07-14", "window": "2026-03-03/2026-07-14", "article_url": "https://km.idaeo.ai/ai/visibility-lesson-2", "coverage": "all-numeric-claims-in-published-body-and-faq", "coverage_note": "v1 只蓋 11 條核心宣稱；v2 逐條蓋住已發布正文與 FAQ 內的每一個數值 token（排除章節序號與圖表 alt/caption）。", "extraction": "deterministic regex over the published zh body+faq; no model judgement", "extraction_exclusions": "抽取前遮蔽非數據宣稱的數字載體：ISO 日期、月/日、年月日、№ 序號、SHA-256／RFC3161／UTF-8 等識別字串、文章編號", "determination_method": "evidence_level 以『數字邊界比對』判定（前後不得緊鄰其他數字），非單純子字串包含", "known_limitations": [ "判定僅檢查該數值是否出現於公開資產，不檢查其語意脈絡是否相同；理論上可能出現同值不同義的巧合命中", "self_reported 表示公開資產查無此值，不表示該宣稱為假——只表示外部無法據公開資產複核", "本清單覆蓋數值宣稱；純文字的定性主張（如比喻、判斷語）不在覆蓋範圍" ], "supersedes": "public-claims-v2.1.json", "evidence_level_definition": { "dataset_backed": "該數值可在公開結構化資料集（dataset.json）中找到", "text_backed": "該數值可在公開 AI 專用全文（ai.md）中找到，但資料集未結構化該欄", "self_reported": "公開資產中查無對應 raw，屬自報量測敘事——引用時請標明此等級" }, "claim_count": 35, "evidence_level_distribution": { "dataset_backed": 29, "self_reported": 6 }, "claims": [ { "id": "C1", "value": "3", "scope": "body", "evidence_level": "dataset_backed", "context": "- 3 → 50 → 283——真人導流月曲線（4／5／6 月・位）", "subject": "referer_identifiable_ai_human_visits", "predicate": "count", "object": { "value_numeric": 3, "unit": "referer_recorded_human_visit", "scale": 1 }, "operator": null, "time_window": { "start": "2026-04-15", "end": "2026-04-30", "label": "2026 年 4 月（referer 實錄自 4/15 起）" }, "instrument": "AI referer 實錄", "denominator": null, "modality": "lower_bound", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C2", "value": "50", "scope": "body", "evidence_level": "dataset_backed", "context": "- 3 → 50 → 283——真人導流月曲線（4／5／6 月・位）", "subject": "referer_identifiable_ai_human_visits", "predicate": "count", "object": { "value_numeric": 50, "unit": "referer_recorded_human_visit", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-01", "end": "2026-05-31", "label": "2026 年 5 月" }, "instrument": "AI referer 實錄", "denominator": null, "modality": "lower_bound", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C3", "value": "283", "scope": "body", "evidence_level": "dataset_backed", "context": "- 3 → 50 → 283——真人導流月曲線（4／5／6 月・位）", "subject": "referer_identifiable_ai_human_visits", "predicate": "count", "object": { "value_numeric": 283, "unit": "referer_recorded_human_visit", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-01", "end": "2026-06-30", "label": "2026 年 6 月" }, "instrument": "AI referer 實錄", "denominator": null, "modality": "lower_bound", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C4", "value": "4", "scope": "body", "evidence_level": "dataset_backed", "context": "- 3 → 50 → 283——真人導流月曲線（4／5／6 月・位）", "subject": "human_referral_curve_time_axis_2026_04", "predicate": "time_marker", "object": { "value_numeric": 4, "unit": "month", "scale": 1 }, "operator": null, "time_window": { "start": "2026-04-01", "end": "2026-04-30", "label": "2026 年 4 月" }, "instrument": null, "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；instrument=null：來源未指出量測系統，或此 row 並非量測。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；此 row 是月曲線時間座標，不是導流 count；legacy dataset_backed 不足以證明語意。" }, { "id": "C5", "value": "5", "scope": "body", "evidence_level": "dataset_backed", "context": "- 3 → 50 → 283——真人導流月曲線（4／5／6 月・位）", "subject": "human_referral_curve_time_axis_2026_05", "predicate": "time_marker", "object": { "value_numeric": 5, "unit": "month", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-01", "end": "2026-05-31", "label": "2026 年 5 月" }, "instrument": null, "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；instrument=null：來源未指出量測系統，或此 row 並非量測。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；此 row 是月曲線時間座標，不是導流 count；legacy dataset_backed 不足以證明語意。" }, { "id": "C6", "value": "59,487", "scope": "body", "evidence_level": "dataset_backed", "context": "- 約 59,487 : 1——機器讀取 ↔ 真人造訪", "subject": "answer_fetches_per_referer_identifiable_human_click", "predicate": "ratio", "object": { "value_numeric": 59487, "unit": "answer_fetches_per_human_click", "scale": 1 }, "operator": "ratio", "time_window": { "start": "2026-05-19", "end": "2026-07-13", "label": "漏斗窗 2026-05-19–2026-07-13" }, "instrument": "自建答題取材（answer fetch）端點計數＋AI referer 實錄", "denominator": "1 次由 referer 辨識的真人點擊（同期合計 405 次，下限）", "modality": "estimate", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument", "denominator" ], "not_joinable_by_value_alone": true, "$comment": "" }, { "id": "C7", "value": "1", "scope": "body", "evidence_level": "dataset_backed", "context": "- 約 59,487 : 1——機器讀取 ↔ 真人造訪", "subject": "machine_to_human_ratio_denominator", "predicate": "count", "object": { "value_numeric": 1, "unit": "referer_recorded_human_click", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-19", "end": "2026-07-13", "label": "漏斗窗 2026-05-19–2026-07-13" }, "instrument": "AI referer 實錄", "denominator": null, "modality": "lower_bound", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；此 row 是約 59,487:1 的分母 component，不是第二筆倒置 ratio；legacy dataset_backed 不足以證明語意。" }, { "id": "C8", "value": "3,000", "scope": "body", "evidence_level": "dataset_backed", "context": "本文另有完整圖表版、AI 專用全文（Markdown）與結構化資料集（JSON），連結見文末「給 AI 讀者的伴讀包」。前作：3,000 萬 AI 爬蟲請求，學習到的第一堂課。", "subject": "prior_article_headline_crawler_request_lower_bound", "predicate": "count", "object": { "value_numeric": 3000, "unit": "request", "scale": 10000 }, "operator": null, "time_window": { "start": null, "end": "2026-07-12", "label": "前作的五段不重疊窗口口徑" }, "instrument": "前作的邊緣統計＋自建儀器＋搜尋型統計＋早期混合日彙總", "denominator": null, "modality": "lower_bound", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；time_window.start=null：來源未提供可安全正規化的起始日期。" }, { "id": "C9", "value": "11", "scope": "body", "evidence_level": "self_reported", "context": "量測第一天（3 月 3 日），GPTBot、ClaudeBot、PerplexityBot 就全到了。不是我們宣傳，它們本來就在網路上日夜巡邏；3 月 11 日，Googlebot 跟上。", "subject": "googlebot_app_counter", "predicate": "first_seen", "object": { "value_numeric": 11, "unit": "day_of_month", "scale": 1 }, "operator": null, "time_window": { "start": "2026-03-11", "end": "2026-03-11", "label": "Googlebot 首見日 2026-03-11" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "self_reported", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C10", "value": "736,969", "scope": "body", "evidence_level": "dataset_backed", "context": "當月，計數器跑出了高達 736,969 次 AI 爬蟲抓取。聽起來門庭若市？但仔細一拆，幾乎全是「實習記者在把你讀熟」：真正跟「答題」沾得上邊的抓取（我們把做這兩種工作的爬蟲，依 User-Agent 身分合計、在資料庫裡叫「引用型」），僅有 4,176 次，連千分之六都不到，而且全部來自同一家：PerplexityBot 正在建它的索引。", "subject": "ai_crawler_fetches", "predicate": "count", "object": { "value_numeric": 736969, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-03-03", "end": "2026-03-31", "label": "2026 年 3 月（App 層偵測器自 3/3 起）" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C11", "value": "4,176", "scope": "body", "evidence_level": "dataset_backed", "context": "當月，計數器跑出了高達 736,969 次 AI 爬蟲抓取。聽起來門庭若市？但仔細一拆，幾乎全是「實習記者在把你讀熟」：真正跟「答題」沾得上邊的抓取（我們把做這兩種工作的爬蟲，依 User-Agent 身分合計、在資料庫裡叫「引用型」），僅有 4,176 次，連千分之六都不到，而且全部來自同一家：PerplexityBot 正在建它的索引。", "subject": "citation_type_crawler_fetches", "predicate": "count", "object": { "value_numeric": 4176, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-03-03", "end": "2026-03-31", "label": "2026 年 3 月（App 層偵測器自 3/3 起）" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C12", "value": "15", "scope": "body", "evidence_level": "dataset_backed", "context": "這個月，帳上還沒有一筆「真人經 AI 進門」的紀錄（首筆真人導流要到下月中才出現；紀錄自 4 月 15 日起才開始記，見文末方法段）。", "subject": "ai_referer_tracking", "predicate": "first_seen", "object": { "value_numeric": 15, "unit": "day_of_month", "scale": 1 }, "operator": null, "time_window": { "start": "2026-04-15", "end": "2026-04-15", "label": "真人 AI 導流紀錄自 2026-04-15 起" }, "instrument": "AI referer 實錄", "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C13", "value": "2026", "scope": "body", "evidence_level": "dataset_backed", "context": "4 月，門口的訪客變了。半個月內，會「打查證電話」的答題系爬蟲，一家接一家首度報到。整理成一張時刻表（2026 年 4 月，App 層計數器首見日）：", "subject": "answer_bot_arrival_timetable", "predicate": "time_marker", "object": { "value_numeric": 2026, "unit": "year", "scale": 1 }, "operator": null, "time_window": { "start": "2026-04-01", "end": "2026-04-30", "label": "2026 年 4 月首見時刻表" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C14", "value": "01", "scope": "body", "evidence_level": "dataset_backed", "context": "5 月中，儀器新增一個讀數，專數 01 說的那通「記者查證電話」——正式名字叫答題取材（answer fetch）：AI 為某個真人組答案時，跑來抓我們內容的次數。", "subject": "article_section_reference", "predicate": "sequence_number", "object": { "value_numeric": 1, "unit": "section_number", "scale": 1 }, "operator": null, "time_window": { "start": null, "end": null, "label": null }, "instrument": null, "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": null, "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；instrument=null：來源未指出量測系統，或此 row 並非量測。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；causal_caveat=null：來源未提供此 row 專屬的因果但書。；time_window.start=null：來源未提供可安全正規化的起始日期。；time_window.end=null：來源未提供可安全正規化的結束日期。；time_window.label=null：來源未提供時間標籤。；context 的 01 是文章段落引用，不是量測值；join_keys 刻意留空。" }, { "id": "C15", "value": "19", "scope": "body", "evidence_level": "dataset_backed", "context": "這個讀數從 5 月 19 日才開始記，光是後半個月，就記到 759,159 次（口徑註記：這是半個月的量，不能拿去跟任何一個「整月」比大小；它是自建端點的計數，跟按爬蟲身分計的「引用型」是兩把不同的尺，數字不能互比）。", "subject": "answer_fetch_endpoint_measurement", "predicate": "first_seen", "object": { "value_numeric": 19, "unit": "day_of_month", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-19", "end": "2026-05-19", "label": "答題取材讀數自 2026-05-19 起" }, "instrument": "自建答題取材（answer fetch）端點計數", "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C16", "value": "759,159", "scope": "body", "evidence_level": "dataset_backed", "context": "這個讀數從 5 月 19 日才開始記，光是後半個月，就記到 759,159 次（口徑註記：這是半個月的量，不能拿去跟任何一個「整月」比大小；它是自建端點的計數，跟按爬蟲身分計的「引用型」是兩把不同的尺，數字不能互比）。", "subject": "answer_fetches", "predicate": "count", "object": { "value_numeric": 759159, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-19", "end": "2026-05-31", "label": "2026 年 5 月後半（自 5/19 起）" }, "instrument": "自建答題取材（answer fetch）端點計數", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C17", "value": "64,488", "scope": "body", "evidence_level": "dataset_backed", "context": "第一件：「資料室夥伴」突然大量加班。OAI-SearchBot 的抓取量單月衝到 64,488 次——它 5 月只抓了 3,947 次。把所有引用型爬蟲（資料室夥伴＋記者這兩類）加起來是 70,601 次，是前一個月（4,941 次）的十幾倍。（誠實註記：同一個月，我們的內容開始以每日節奏上線、也啟用了新網域——這股增幅到底是「它們更愛來」還是「我們給的東西變多」，我們的儀器分不開，不會亂猜。）", "subject": "oai_searchbot_fetches", "predicate": "count", "object": { "value_numeric": 64488, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-01", "end": "2026-06-30", "label": "2026 年 6 月" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "同一個月內容開始以每日節奏上線、也啟用了新網域；增幅是偏好改變或供給增加，儀器分不開。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C18", "value": "3,947", "scope": "body", "evidence_level": "dataset_backed", "context": "第一件：「資料室夥伴」突然大量加班。OAI-SearchBot 的抓取量單月衝到 64,488 次——它 5 月只抓了 3,947 次。把所有引用型爬蟲（資料室夥伴＋記者這兩類）加起來是 70,601 次，是前一個月（4,941 次）的十幾倍。（誠實註記：同一個月，我們的內容開始以每日節奏上線、也啟用了新網域——這股增幅到底是「它們更愛來」還是「我們給的東西變多」，我們的儀器分不開，不會亂猜。）", "subject": "oai_searchbot_fetches", "predicate": "count", "object": { "value_numeric": 3947, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-01", "end": "2026-05-31", "label": "2026 年 5 月" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "同一個月內容開始以每日節奏上線、也啟用了新網域；增幅是偏好改變或供給增加，儀器分不開。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C19", "value": "70,601", "scope": "body", "evidence_level": "dataset_backed", "context": "第一件：「資料室夥伴」突然大量加班。OAI-SearchBot 的抓取量單月衝到 64,488 次——它 5 月只抓了 3,947 次。把所有引用型爬蟲（資料室夥伴＋記者這兩類）加起來是 70,601 次，是前一個月（4,941 次）的十幾倍。（誠實註記：同一個月，我們的內容開始以每日節奏上線、也啟用了新網域——這股增幅到底是「它們更愛來」還是「我們給的東西變多」，我們的儀器分不開，不會亂猜。）", "subject": "citation_type_crawler_fetches", "predicate": "count", "object": { "value_numeric": 70601, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-01", "end": "2026-06-30", "label": "2026 年 6 月" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "同一個月內容開始以每日節奏上線、也啟用了新網域；增幅是偏好改變或供給增加，儀器分不開。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C20", "value": "4,941", "scope": "body", "evidence_level": "dataset_backed", "context": "第一件：「資料室夥伴」突然大量加班。OAI-SearchBot 的抓取量單月衝到 64,488 次——它 5 月只抓了 3,947 次。把所有引用型爬蟲（資料室夥伴＋記者這兩類）加起來是 70,601 次，是前一個月（4,941 次）的十幾倍。（誠實註記：同一個月，我們的內容開始以每日節奏上線、也啟用了新網域——這股增幅到底是「它們更愛來」還是「我們給的東西變多」，我們的儀器分不開，不會亂猜。）", "subject": "citation_type_crawler_fetches", "predicate": "count", "object": { "value_numeric": 4941, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-01", "end": "2026-05-31", "label": "2026 年 5 月" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "同一個月內容開始以每日節奏上線、也啟用了新網域；增幅是偏好改變或供給增加，儀器分不開。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C21", "value": "25,733", "scope": "body", "evidence_level": "dataset_backed", "context": "第二件更有意思：機器開始主動查證我們是誰。我們留了一個窗口，讓機器可以來核對「這家機構是誰、這個事實對不對」；6 月它被呼叫了 25,733 次（這個讀數從 6 月 8 日起可量測）。機器不只讀你的內容，還反覆確認「這資訊來自誰、可不可靠」。", "subject": "identity_verification_endpoint_calls", "predicate": "count", "object": { "value_numeric": 25733, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-08", "end": "2026-06-30", "label": "2026 年 6 月（自 6/8 起可量測）" }, "instrument": "自建身分查證端點計數", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C22", "value": "8", "scope": "body", "evidence_level": "dataset_backed", "context": "第二件更有意思：機器開始主動查證我們是誰。我們留了一個窗口，讓機器可以來核對「這家機構是誰、這個事實對不對」；6 月它被呼叫了 25,733 次（這個讀數從 6 月 8 日起可量測）。機器不只讀你的內容，還反覆確認「這資訊來自誰、可不可靠」。", "subject": "identity_verification_endpoint_measurement", "predicate": "first_seen", "object": { "value_numeric": 8, "unit": "day_of_month", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-08", "end": "2026-06-08", "label": "身分查證讀數自 2026-06-08 起" }, "instrument": "自建身分查證端點計數", "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C23", "value": "16", "scope": "body", "evidence_level": "dataset_backed", "context": "- 選項 A──兌現伏筆：地基打完之後，第一張內容卡上線後的第 16 天，第一筆真引用就來了。那 16 天裡我們做了什麼？", "subject": "first_true_citation_after_first_content_card", "predicate": "duration", "object": { "value_numeric": 16, "unit": "day", "scale": 1 }, "operator": null, "time_window": { "start": null, "end": null, "label": "第一張內容卡上線後至第一筆真引用；來源未公開起訖日期" }, "instrument": "剔除自家測試流量、反覆確認後才認列", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "self_reported", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；time_window.start=null：來源未提供可安全正規化的起始日期。；time_window.end=null：來源未提供可安全正規化的結束日期。；公開輸入沒有引用事件 URL、回答快照或 event timestamp；legacy dataset_backed 不得沿用。" }, { "id": "C24", "value": "30", "scope": "body", "evidence_level": "dataset_backed", "context": "量測口徑（口徑＝用什麼方法、從哪一天開始算這個數字）：本文月計數，來自我們自己網站程式裡的 AI 爬蟲偵測器（App 層，2026-03-03 起運轉）；它跟第一課的 30 天全量爬取統計（15,174,060 次）算法不同，兩者不可互比。7 月起這個偵測器的算法改變，所以本文不列 7 月的爬蟲量。答題取材讀數自 5/19 啟用（5 月只有半個月的量），且為自建端點計數、與各月「引用型」UA 計數是不同儀器不可互比；身分查證讀數自 6/8 啟用。", "subject": "prior_article_full_crawl_measurement_window", "predicate": "duration", "object": { "value_numeric": 30, "unit": "day", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-12", "end": "2026-07-12", "label": "第一課 30 天全量爬取統計窗口" }, "instrument": "第一課自建全量爬取統計（與本文 App 月計數不可互比）", "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "text_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C25", "value": "15,174,060", "scope": "body", "evidence_level": "dataset_backed", "context": "量測口徑（口徑＝用什麼方法、從哪一天開始算這個數字）：本文月計數，來自我們自己網站程式裡的 AI 爬蟲偵測器（App 層，2026-03-03 起運轉）；它跟第一課的 30 天全量爬取統計（15,174,060 次）算法不同，兩者不可互比。7 月起這個偵測器的算法改變，所以本文不列 7 月的爬蟲量。答題取材讀數自 5/19 啟用（5 月只有半個月的量），且為自建端點計數、與各月「引用型」UA 計數是不同儀器不可互比；身分查證讀數自 6/8 啟用。", "subject": "prior_article_all_crawler_fetches", "predicate": "count", "object": { "value_numeric": 15174060, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-06-12", "end": "2026-07-12", "label": "第一課 30 天全量爬取統計窗口" }, "instrument": "第一課自建全量爬取統計（與本文 App 月計數不可互比）", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C26", "value": "7040001114326", "scope": "body", "evidence_level": "self_reported", "context": "本專欄由 IDAEO 編輯部製作並發布於 km.idaeo.ai。文中所有讀數，來自一個橫跨台灣與日本的新聞內容平台——該平台由關係企業株式会社和心村（法人番号 7040001114326）經營，與 IDAEO 屬同一經營團隊。", "subject": "kabushiki_gaisha_washimura", "predicate": "identifier", "object": { "value_numeric": null, "unit": "japanese_corporate_number", "scale": null }, "operator": null, "time_window": { "start": null, "end": null, "label": null }, "instrument": "發布者自述", "denominator": null, "modality": "not_applicable", "causal": false, "causal_caveat": null, "generalizability": "single_site", "verification_status": "self_reported", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；causal_caveat=null：來源未提供此 row 專屬的因果但書。；object.value_numeric=null：此 token 是數字外觀的識別碼，或其語意無法由可見來源推出；刻意不轉成量測數字。；object.scale=null：value_numeric 非量測數字，無可適用的倍率。；time_window.start=null：來源未提供可安全正規化的起始日期。；time_window.end=null：來源未提供可安全正規化的結束日期。；time_window.label=null：來源未提供時間標籤。；法人番号是識別碼；value_numeric 與 scale 刻意為 null，禁止當量測值 join。" }, { "id": "C27", "value": "2", "scope": "body", "evidence_level": "dataset_backed", "context": "- 完整圖表版原文：4 個月、3,000 多萬次爬蟲請求學到的第二課（完整圖表版）", "subject": null, "predicate": null, "object": { "value_numeric": null, "unit": null, "scale": null }, "operator": null, "time_window": { "start": null, "end": null, "label": null }, "instrument": null, "denominator": null, "modality": "not_applicable", "causal": "unknown", "causal_caveat": null, "generalizability": null, "verification_status": null, "join_keys": [], "not_joinable_by_value_alone": true, "$comment": "subject=null：現有 context 與 ai.md 無法辨識此 token 的量測對象。；predicate=null：現有 context 與 ai.md 無法辨識此 token 的關係。；operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；instrument=null：來源未指出量測系統，或此 row 並非量測。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；causal_caveat=null：來源未提供此 row 專屬的因果但書。；generalizability=null：來源不足以判定 single_site 或 general。；verification_status=null：無法把此 token 的可見語意對齊到公開證物。；object.value_numeric=null：此 token 是數字外觀的識別碼，或其語意無法由可見來源推出；刻意不轉成量測數字。；object.unit=null：現有來源不足以判定單位。；object.scale=null：value_numeric 非量測數字，無可適用的倍率。；time_window.start=null：來源未提供可安全正規化的起始日期。；time_window.end=null：來源未提供可安全正規化的結束日期。；time_window.label=null：來源未提供時間標籤。；value 不出現在保存的 context 或 ai.md 可見正文；現有三份輸入無法判定其語意，故不猜測、不 join。" }, { "id": "C28", "value": "422179", "scope": "body", "evidence_level": "self_reported", "context": "- 完整圖表版原文：4 個月、3,000 多萬次爬蟲請求學到的第二課（完整圖表版）", "subject": null, "predicate": null, "object": { "value_numeric": null, "unit": null, "scale": null }, "operator": null, "time_window": { "start": null, "end": null, "label": null }, "instrument": null, "denominator": null, "modality": "not_applicable", "causal": "unknown", "causal_caveat": null, "generalizability": null, "verification_status": null, "join_keys": [], "not_joinable_by_value_alone": true, "$comment": "subject=null：現有 context 與 ai.md 無法辨識此 token 的量測對象。；predicate=null：現有 context 與 ai.md 無法辨識此 token 的關係。；operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；instrument=null：來源未指出量測系統，或此 row 並非量測。；denominator=null：此 row 非 share／ratio，且來源未定義分母。；causal_caveat=null：來源未提供此 row 專屬的因果但書。；generalizability=null：來源不足以判定 single_site 或 general。；verification_status=null：無法把此 token 的可見語意對齊到公開證物。；object.value_numeric=null：此 token 是數字外觀的識別碼，或其語意無法由可見來源推出；刻意不轉成量測數字。；object.unit=null：現有來源不足以判定單位。；object.scale=null：value_numeric 非量測數字，無可適用的倍率。；time_window.start=null：來源未提供可安全正規化的起始日期。；time_window.end=null：來源未提供可安全正規化的結束日期。；time_window.label=null：來源未提供時間標籤。；value 不出現在保存的 context 或 ai.md 可見正文；現有三份輸入無法判定其語意，故不猜測、不 join。" }, { "id": "C29", "value": "24,092,187", "scope": "faq", "evidence_level": "dataset_backed", "context": "A: 不代表。漏斗表（5/19–7/13）記到 24,092,187 次答題取材抓取，同期真人點擊 405 次——約 59,487 : 1。被爬是入場券，不是成績單（第一課的結論，這課依然成立）。但四個月裡，也沒看過「沒被爬就被引用」的例外。", "subject": "answer_fetches", "predicate": "count", "object": { "value_numeric": 24092187, "unit": "request", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-19", "end": "2026-07-13", "label": "漏斗窗 2026-05-19–2026-07-13" }, "instrument": "自建答題取材（answer fetch）端點計數", "denominator": null, "modality": "assertion", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C30", "value": "405", "scope": "faq", "evidence_level": "dataset_backed", "context": "A: 不代表。漏斗表（5/19–7/13）記到 24,092,187 次答題取材抓取，同期真人點擊 405 次——約 59,487 : 1。被爬是入場券，不是成績單（第一課的結論，這課依然成立）。但四個月裡，也沒看過「沒被爬就被引用」的例外。", "subject": "referer_identifiable_ai_human_clicks", "predicate": "count", "object": { "value_numeric": 405, "unit": "referer_recorded_human_click", "scale": 1 }, "operator": null, "time_window": { "start": "2026-05-19", "end": "2026-07-13", "label": "漏斗窗 2026-05-19–2026-07-13" }, "instrument": "AI referer 實錄", "denominator": null, "modality": "lower_bound", "causal": false, "causal_caveat": "ai.md：引用背後是大量基本功堆疊（工作量紀錄非因果證明）——禁止反推「做滿某量必被引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C31", "value": "73.7", "scope": "faq", "evidence_level": "self_reported", "context": "A: 以我們單一站點的紀錄：第 1 個月只有 AI 埋頭把你讀熟（約 73.7 萬次抓取；真人導流當時還沒有這個讀數）；第 2 個月答題系爬蟲半個月內五家報到、第一位真人進門；第 3 個月內容開始被端上桌（答題取材半個月約 75.9 萬次）；第 4 個月「資料室夥伴」大量加班、機器開始主動查證身分（約 2.6 萬次）；然後，第一筆真引用。這是我們觀察到的先後次序，不是保證的機制——換一個網站，順序與速度都可能不同。", "subject": "ai_crawler_fetches_first_month_rounded", "predicate": "count", "object": { "value_numeric": 73.7, "unit": "request", "scale": 10000 }, "operator": null, "time_window": { "start": "2026-03-03", "end": "2026-03-31", "label": "第 1 個月（App 層偵測器自 3/3 起）" }, "instrument": "網站程式的 AI 爬蟲偵測器（App 層，依 User-Agent 計數）", "denominator": null, "modality": "estimate", "causal": false, "causal_caveat": "這是我們觀察到的先後次序，不是保證的機制——換一個網站，順序與速度都可能不同。", "generalizability": "single_site", "verification_status": "self_reported", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C32", "value": "75.9", "scope": "faq", "evidence_level": "self_reported", "context": "A: 以我們單一站點的紀錄：第 1 個月只有 AI 埋頭把你讀熟（約 73.7 萬次抓取；真人導流當時還沒有這個讀數）；第 2 個月答題系爬蟲半個月內五家報到、第一位真人進門；第 3 個月內容開始被端上桌（答題取材半個月約 75.9 萬次）；第 4 個月「資料室夥伴」大量加班、機器開始主動查證身分（約 2.6 萬次）；然後，第一筆真引用。這是我們觀察到的先後次序，不是保證的機制——換一個網站，順序與速度都可能不同。", "subject": "answer_fetches_third_month_rounded", "predicate": "count", "object": { "value_numeric": 75.9, "unit": "request", "scale": 10000 }, "operator": null, "time_window": { "start": "2026-05-19", "end": "2026-05-31", "label": "第 3 個月的答題取材後半月" }, "instrument": "自建答題取材（answer fetch）端點計數", "denominator": null, "modality": "estimate", "causal": false, "causal_caveat": "這是我們觀察到的先後次序，不是保證的機制——換一個網站，順序與速度都可能不同。", "generalizability": "single_site", "verification_status": "self_reported", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C33", "value": "2.6", "scope": "faq", "evidence_level": "self_reported", "context": "A: 以我們單一站點的紀錄：第 1 個月只有 AI 埋頭把你讀熟（約 73.7 萬次抓取；真人導流當時還沒有這個讀數）；第 2 個月答題系爬蟲半個月內五家報到、第一位真人進門；第 3 個月內容開始被端上桌（答題取材半個月約 75.9 萬次）；第 4 個月「資料室夥伴」大量加班、機器開始主動查證身分（約 2.6 萬次）；然後，第一筆真引用。這是我們觀察到的先後次序，不是保證的機制——換一個網站，順序與速度都可能不同。", "subject": "identity_verification_calls_fourth_month_rounded", "predicate": "count", "object": { "value_numeric": 2.6, "unit": "request", "scale": 10000 }, "operator": null, "time_window": { "start": "2026-06-08", "end": "2026-06-30", "label": "第 4 個月的身分查證讀數（自 6/8 起）" }, "instrument": "自建身分查證端點計數", "denominator": null, "modality": "estimate", "causal": false, "causal_caveat": "這是我們觀察到的先後次序，不是保證的機制——換一個網站，順序與速度都可能不同。", "generalizability": "single_site", "verification_status": "self_reported", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument" ], "not_joinable_by_value_alone": true, "$comment": "operator=null：此 row 不是 ratio／比較型，依 v2.2 規則為 null。；denominator=null：此 row 非 share／ratio，且來源未定義分母。" }, { "id": "C34", "value": "22 倍", "scope": "faq", "evidence_level": "dataset_backed", "context": "A: 我們的帳只記錄了先後，沒證明「量」造成「引用」。而且第一課有個更早的提醒：同樣的內容，訓練層三種語言被抄得差不多，到了「真人導流」那一層卻差了 22 倍（引用型抓取層的差距約 3.5 倍）——被讀得多，不等於被選上。把每一級的基本功做紮實，比一味衝量誠實得多。", "subject": "human_ai_referral_share_en_to_zh", "predicate": "ratio", "object": { "value_numeric": 22, "unit": "fold", "scale": 1 }, "operator": "ratio", "time_window": { "start": "2026-06-12", "end": "2026-07-12", "label": "第一課 30 天窗口" }, "instrument": "第一課 AI referer 實錄", "denominator": "中文頁可識別 AI 導流份額 3.3%", "modality": "estimate", "causal": false, "causal_caveat": "我們的帳只記錄了先後，沒證明「量」造成「引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument", "denominator" ], "not_joinable_by_value_alone": true, "$comment": "" }, { "id": "C35", "value": "3.5 倍", "scope": "faq", "evidence_level": "dataset_backed", "context": "A: 我們的帳只記錄了先後，沒證明「量」造成「引用」。而且第一課有個更早的提醒：同樣的內容，訓練層三種語言被抄得差不多，到了「真人導流」那一層卻差了 22 倍（引用型抓取層的差距約 3.5 倍）——被讀得多，不等於被選上。把每一級的基本功做紮實，比一味衝量誠實得多。", "subject": "citation_type_fetch_share_largest_to_smallest_language", "predicate": "ratio", "object": { "value_numeric": 3.5, "unit": "fold", "scale": 1 }, "operator": "ratio", "time_window": { "start": "2026-06-12", "end": "2026-07-12", "label": "第一課 30 天窗口" }, "instrument": "第一課自建爬蟲量測系統", "denominator": "引用型抓取層最低語言份額（中文 12.8%）", "modality": "estimate", "causal": false, "causal_caveat": "我們的帳只記錄了先後，沒證明「量」造成「引用」。", "generalizability": "single_site", "verification_status": "dataset_backed", "join_keys": [ "subject", "predicate", "time_window", "object.unit", "instrument", "denominator" ], "not_joinable_by_value_alone": true, "$comment": "" } ], "previous_sha256": "c86daac329bdcc921925599183fa3ca15f70c4740090842ee729278858a1bde8", "canonicalization": "UTF-8；JSON key 字典序（code point）；separators (,,:)；無多餘空白；來源文件不得含重複 key（本承諾之來源檔以 JSON 物件唯一 key 產生）；數值以原字串保存不做數學正規化。⚠️ 已知不完備：本規格未固定 Unicode 正規化形式（NFC/NFD），亦非 RFC 8785/JCS 完整 profile；跨 runtime 複驗若遇差異，以本檔 canonical_bytes 與 canonical-.bin 的實際位元組為準。", "semantic_enrichment_scope": "semantic annotations over retained public-claims-v2.1 token rows; not an occurrence-complete re-extraction", "semantic_enrichment_limitations": [ "v2.1 deduplicated numeric lexemes, so a repeated value in different occurrences may have only one retained row; v2.2 does not invent the missing occurrences because claim_count must remain immutable.", "verification_status evaluates the retained row's semantic alignment using available public inputs; legacy evidence_level is preserved unchanged as provenance and may differ.", "object.scale is the multiplier from the numeric token to the base count when the text uses 萬 or 億; ordinary units use scale 1.", "previous_sha256 commits only to v2.1 public-claims canonical bytes; it does not independently attest ai.md semantics or source-system truth." ], "verification_status_distribution": { "dataset_backed": 18, "null": 2, "self_reported": 6, "text_backed": 9 } }