绿色发展指数系列Index Series
五大绿色发展指数,双信号联动Five Green Development Indices, Dual-Signal Convergence
资本信号层 × 实物资产信号层,五指数联动,年度发布 + 季度更新。Capital signal layer x physical-asset signal layer, annually released with quarterly updates.
INDEX 01 · 资本信号层Capital Signal
ESG 评级跟踪指数ESG Rating Tracking Index
蒸馏共识模型——中证、华证、万得三大评级全量采集(800 家 A 股)叠加 MSCI/商道融绿等多机构种子样本,按社会采用权重加权蒸馏为 [0,100] 共识分。产出 DCS 共识评分 + DIV 分歧度 + MOM 动量三重信号,基期 2026-09-30 = 1000,季度更新。全部分数由 zci 引擎真实计算,可审计、可复算。Distilled consensus model — CSI, Huazheng and Wind ratings collected in full (800 A-shares) plus multi-agency seed coverage, weighted by social adoption into a [0,100] consensus score. DCS + DIV + MOM signals, base 2026-09-30 = 1000, quarterly.
三机构全量采集3 agencies, full coverageDCS+DIV+MOM基期 1000Base 1000
INDEX 02 · 实物信号层Physical Signal
零碳园区发展指数Zero-Carbon Park Index
"1+5+4"评价体系(1 核心指标 + 5 引导指标 + 4 加分/特色项),对标发改环资〔2025〕910号文,区域层。覆盖国家级 52 家 + 省级约 200 家 + 全球 20 家园区,跨区域横向可比。"1+5+4" framework, benchmarking 52 national + 200 provincial + 20 global parks.
1+5+4 体系区域层Regional名单池 177 个(国家级 52+省级 125)Pool: 177 parks (52 national + 125 provincial)参评池公示已上线Pool prerelease live
INDEX 03 · 实物信号层Physical Signal
零碳工厂建设发展指数Zero-Carbon Factory Index
"3+8+3"评价体系 + 334号文三项核心指标(一票否决):单位能耗碳排放基本≤2/目标≤0.2 吨CO₂/吨标煤;非化石能源消费占比基本≥30%/目标≥95%;非化石能源电力物理认定量占比基本≥10%/目标≥35%(算力设施除外),企业层。"3+8+3" framework + 334 circular three core indicators, enterprise level.
3+8+3 体系334 三指标全国台账 311 家名单(七省以上)National ledger: 311 factories名单公示已上线List prerelease live
INDEX 04 · 实物信号层Physical Signal
零碳供应链指数Zero-Carbon Supply Chain Index
"4+8"指标体系,产品链层。对标欧盟 DPP / CBAM 与 334 号文零碳供应链条款,评价链主企业及其供应商的碳合规竞争力。链主候选池 570 家全部取自中证800,35 个采集字段公开可得性 94.3%(A 级 20 项直接披露 / B 级 13 项可推算),已由「企业填报」改为「我方公开采集 + 企业确认」。首谈名单 v1.0:★45 家全量实测,由「每赛道 DCS 前 3」名额制改为「DCS ≥ 85 × 披露得分 ≥ 55」达标制,19 家入围。首谈梯队得分卡包 v1.0:20 家逐家给出缺证要件 → 补强动作 → 服务赛道映射,查出「目标与底数已成披露标配、量化执行才是共性短板」(供应商碳强度下降 12/20 家缺证、零碳工厂 9/20、ESG 纳入考核 8/20)。评级可行性已判定:24 项末级指标过三道闸门,公开资料可覆盖 92.5% 权重(37.5% 无假设直取 + 55% 须先固定分母口径),仅 7.5% 须企业侧数据——可发布「公开披露基准评级」,须双轨呈现。分母口径手册 v1.0:12 项条件可评指标逐项写明「公开近似口径 + 企业确认口径」双版本(6 种折算方法、11 个确认字段即可把 55% 权重升为真值);54 家实测证明折算只改绝对档位、不改相对位次(秩相关 0.979,16 家/21 家跳档)。集团—工厂主体归集表 v1.2 已建成(五指数共用基础层):链主池 570 家 × 官方名单 9,174 条,命中由首跑的 32 家升至 273 家(采信层 216 家)。缺口性质查明:「工厂法人主体 ≠ 上市公司集团主体」——官方名单登记的是「格力电器(南京)有限公司」,上市公司是「格力电器」,不是数据缺失而是缺一张主体映射表。三层匹配(L1 归一同名 / L2 主体词关联 / L3 品牌前缀线索)+ 三张排除表(伞形集团/泛化主体词/通用地名品牌)。首谈落地包 v1.0:把「能覆盖 92.5% 权重、但 12 项卡分母」的结论落成两条可立即执行的动作——对外一封「只补分母、不查绩效」的定向征询函(请求分两栏:分母补全 11 个汇总数 + 证据补位「某项已公开但我方未检索到」),对内一张 299 家链主未归集的派工清单(P1 首谈名单内 3 家 / P2 DCS≥85 139 家 / P3 其余 157 家)。首谈 20 家平均只需确认 3—9 个字段,其中 2 家只需 3 个——每家企业收到的只是自己缺的那几个数,不做全员同表。核心设计:不问「你们做得好不好」,只问「一共多少家」,把审查性问询变为技术性核对。"4+8" framework, product-chain level. All 570 chain-leader candidates come from CSI 800; 94.3% of the 35 collection fields are publicly obtainable. Shortlist v1.0: all 45 measured, quota rule replaced by a threshold rule — 19 names qualify. Tier scorecards map each gap to an action and a service track: targets and baselines are now table stakes, quantified execution is the common gap. Rating feasibility: public sources cover 92.5% of the 24-metric weight, so a public-disclosure-based rating can be released on a dual-track design. The Caliber Manual v1.0 defines both a public-proxy and a company-confirmed caliber for each of the 12 conditional metrics (6 proxy methods, 11 confirmation fields), with a 54-name sensitivity test showing ranks are stable (rho 0.979) while absolute bands shift. The Group-Factory Entity Map v1.2 is now built as a shared base layer: hits rose from 32 to 273 of 570 chain leaders (216 trusted tier). The gap was a naming divide, not missing data (listed groups vs. factory legal entities). Three match tiers plus three exclusion lists keep precision tight. The Shortlist Action Pack v1.0 turns the assessment into two executable moves: an outbound inquiry letter that asks only for denominators, never for performance (two sections: 11 aggregate figures + evidence pointers for items we could not locate publicly), and an internal dispatch list covering the 299 unmapped chain leaders (3 priority-1 from the shortlist, 139 with DCS >= 85). The 20 shortlisted firms each need only 3 to 9 fields, tailored per firm rather than a blanket form.
4+8 体系产品链层Product Chain中证800 链主池 570 家570 CSI 800 leaders公开可得性 94.3%Public availability 94.3%首谈名单 v1.0 · 54 家实测Shortlist v1.0 · 54 measured得分卡包 v1.0 · 梯队 20 家Scorecards v1.0 · 20 names评级可实现权重 92.5%Rating coverage 92.5%分母口径手册 v1.0 · 6 法 11 字段Caliber Manual · 6 methodsFactoryZCI 反查已跑通Back-check operational
INDEX 05 · 资本信号层Capital Signal
REITs 绿色溢价定价REITs Green Premium Pricing
REITsZCI-GP 三法递进定价:配对法(同类同权市值加权基准)→ 横截面回归(披露度因子弹性 β)→ 合理分派率定价(错价信号)→ 隐含碳价 ICP。全市场 84 只全字段公开采集(26 字段,A 级 68 只)· 绿色披露指数 GDI 实测 · 按「月度行情 / 季度运营 / 年度披露」三频周期发布。REITsZCI-GP three-step pricing on all 84 C-REITs (26-field public collection, 68 grade-A) with a measured Green Disclosure Index (GDI); released monthly, quarterly and annually.
三法递进84 只 C-REITs26 字段全覆盖26 fields, 100%GDI 实测周期发布已启用Periodic release on
发布节奏Release Roadmap
2026.10
预发布上线:RatingZCI v0.9(基期 1000)+ REITsZCI-GP v1.0 全真实层(84 只 × 26 字段全量采集 + GDI 实测)+ ParkZCI/FactoryZCI 参评池公示(177 园区 / 311 工厂)+ SupplyZCI 链主候选池榜单(570 家 DCS 真算,中证800)· 公开可得性核查报告(35 字段 94.3% 公开可得)· 首谈名单 v1.0(★45 家全量实测,双条件重排 19 家入围)· 评级可行性判定(公开资料可覆盖 92.5% 权重)· 分母口径手册 v1.0(12 项双口径 · 6 种折算方法 · 11 字段确认清单 · 54 家敏感性实测)· FactoryZCI 反查通道首跑(命中 32 家)· 集团—工厂主体归集表 v1.2(五指数共用基础层,命中 273 家 / 采信层 216 家)· 首谈落地包 v1.0(企业确认版定向征询函 · 20 家逐家缺口单 · 299 家归集派工清单)Pre-release live: RatingZCI v0.9 + REITsZCI-GP v1.0 + Park/Factory pool listings + SupplyZCI pool (570 CSI 800 names), availability check (94.3% of fields), Shortlist v1.0 (19 qualify), rating-feasibility assessment (92.5% of weight coverable from public sources), Caliber Manual v1.0 (12 dual-caliber metrics, 6 proxy methods, 11 confirmation fields) the first FactoryZCI back-check run (32 hits) and the Group-Factory Entity Map v1.2 (273 hits, 216 in the trusted tier) and the Shortlist Action Pack v1.0 (targeted inquiry letter + per-firm gap sheets + 299-name dispatch list)
2026.11
园区指数首发《2026 全球零碳园区转型发展指数报告》Park Index debut report
2026.12
工厂指数报告 + REITs 基期锚定日(2026-12-31,与 RatingZCI L3 组合基期一致)Factory Index + REITs base date 2026-12-31
2027.Q1
五指数联合发布会(评级跟踪 + 园区 + 工厂 + 供应链 + REITs)并同步北欧零碳委员会官网Five-index joint launch
2027+
季度更新 + 年度白皮书 + 方法论评审Quarterly updates + annual white paper
零碳双优标杆榜Dual-Excellence Leaders
评级曲线 × 园区工厂碳绩效交叉验证——上市公司 ESG 共识改善且核心生产基地碳绩效双优,年度发布。Rating curve x carbon performance cross-validation, annual.
双信号预警榜Dual-Signal Alert
评级分歧 + 碳绩效背离——DIV 飙升且工厂核心指标未达标,季度发布,前瞻预警。DIV spike + factory miss, quarterly forward alert.
算力设施零碳榜 TOP20Zero-Carbon Computing TOP20
呼应 8/17 新政零碳算力设施纳入建设范围,覆盖数据中心/智算中心,TOP 20 排名。Data centers / AI computing, TOP 20.
RatingZCI 预发布(v0.9)已上线RatingZCI Pre-release (v0.9) Live
基期 2026-09-30 = 1000 点 · 共识样本 798 家(中证 800 全覆盖,三机构评级蒸馏)· 全部由 zci 引擎复算。10 月本站预发布,正式版 2027-01 首发并同步欧美同学会北欧零碳委员会官网。
Base 2026-09-30 = 1000 · 798 consensus names from full CSI-800 coverage. Formal v1.0 launches Jan 2027 on this site and the Nordic Zero-Carbon Committee website.
查看预发布榜单 →View Pre-release →
REITsZCI-GP 定价报告(v1.0 全真实层)已上线REITsZCI-GP v1.0 Full-Real Live
数据日 2026-09-30 收盘(月度刷新 2026-10-01):C-REITs 全市场 84 只(合计市值 2,093 亿元,市值加权 TTM 分派率 6.59%)· 26 字段公开采集全覆盖(A 级 68 / B 级 16 / C 级 0)· 配对法绿色溢价真算(可配对 80 只,正溢价 44 只)· 绿色披露指数 GDI 实测与披露度因子回归(n=80,β=2.2bp/分,R²=0.30)· 溢价榜首易方达广西北投高速REIT +639bp(新上市品种 TTM 口径偏高,不具可比性,见报告标注)· 隐含碳价待碳排放披露覆盖达标后发布 · 月度/季度/年度三频周期发布。
Data as of Sep 30, 2026 close (monthly refresh Oct 1): all 84 C-REITs (RMB 209.3bn, cap-weighted TTM payout 6.59%); 26-field public collection (68 grade-A / 16 B / 0 C); matched-pair premium on 80 names (44 positive); measured Green Disclosure Index and disclosure-factor regression (n=80, beta 2.2bp/pt, R2 0.30); top premium E Fund Guangxi Beitou Expressway REIT +639bp (newly listed, TTM not comparable); ICP pending disclosure coverage. Monthly/quarterly/annual cadence.
查看定价报告 →View Report →
ParkZCI / FactoryZCI 参评池公示(v0.9)已上线ParkZCI / FactoryZCI Pool Listings (v0.9) Live
ParkZCI 参评池 177 个园区(国家级 52 全名单 + 省级 125)· FactoryZCI 全国台账 311 家名单(锚点 391,七省以上)· 只公示名单与口径,不含评分——指标数据经申报/采集核对后才进入计算。两池分属发改委线与工信部线,互不混同。
ParkZCI: 177 parks (52 national full list + 125 provincial); FactoryZCI: 311 factories nationwide (anchor 391). Names and criteria only — no scores until data collection completes. NDRC line and MIIT line kept strictly separate.
园区参评池 →Park Pool →
工厂名单 →Factory List →
SupplyZCI 链主首谈名单 v1.0 + 梯队得分卡包 + 评级可行性判定 + 分母口径手册 v1.0 + 集团—工厂归集表 v1.2 + 首谈落地包 v1.0 + 公开披露基准版 v1.0 + 报告内页精读 v0.4 + 证据到指标判定规则 v1.3SupplyZCI Shortlist v1.0 + Scorecards + Rating Feasibility + Caliber Manual v1.0 + Entity Map v1.2 + Action Pack v1.0 + Public Baseline v1.0 + Report Insight v0.4 + Indicator Judge v1.3
54 家实测(★45 家名单全量跑完 + 9 家名单外对照),按 SupplyZCI 4+8 框架 17 项要件逐项核验公开证据。本版查明原名单的结构性缺陷:「每赛道 DCS 前 3」是名额制——电子赛道 149 家候选也只给 3 个名额,导致 DCS 93.2 的立讯精密、91.0 的工业富联、88.7 的比亚迪被名额挤出;而纺织赛道只有 3 家候选,DCS 83.0 也能入选。改为「DCS ≥ 85 × 公开披露得分 ≥ 55」达标制后 19 家入围;宁德时代(披露 95 分、DCS 82.6)列为方法论优先合作对象。本轮新出得分卡包:梯队 20 家逐家给出缺证要件 → 补强动作 → 服务赛道映射,并查出「目标与底数已成披露标配、量化执行才是共性短板」——供应商碳强度下降(12/20 家缺证)、零碳工厂(9/20)、ESG 纳入考核(8/20)最集中。入选资格≠SupplyZCI 评级。
54 names measured against 17 SupplyZCI 4+8 requirements. Structural flaw found: the old "top-3 DCS per sector" rule was a quota, not a threshold — electronics (149 candidates) got only 3 slots, pushing out Luxshare (93.2), FII (91.0) and BYD (88.7). Switched to a threshold rule: 19 qualify; CATL (95 disclosed, DCS 82.6) is a methodology-priority partner. New tier scorecards map each gap to an action and a service track; targets and baselines are now table stakes while quantified execution is the common gap.
评级路线已判定:24 项末级指标过三道闸门——公开资料可覆盖 92.5% 权重(37.5% 无假设直取 + 55% 须先固定分母口径),仅 7.5% 须企业侧数据;54 家实证试算显示分档拉得开(极差 83 分、五档全有),但 SC-5 占 35.2% 含大量「未披露」而非「未做」的企业,故须「披露完备度 + 绩效分」双轨呈现——可发布「公开披露基准评级」。
Rating route determined: public sources cover 92.5% of the 24-metric weight (37.5% assumption-free, 55% pending denominator definitions); only 7.5% needs corporate-side data. A 54-name trial shows sufficient spread, but 35.2% falling into SC-5 reflects non-disclosure rather than non-performance — hence a dual-track "disclosure completeness + performance" design.
分母口径手册 v1.0:可行性判定查明的真瓶颈是「分母口径」——企业给「374 家供应商披露碳排放」却不给总数。本手册把 12 项条件可评指标(55% 权重)逐项写成双口径:公开近似口径(6 种折算方法 M1—M6,每种配必须公开的假设与页面标注)+企业确认口径(去重后只有 11 个字段,全部来自企业现成台账,一次确认长期复用)。54 家实测折算敏感性:相对位次高度稳定(秩相关 0.979)、绝对档位会移动(保守假定下 21/54 家跳档、15 家仍稳在 55 分线以上,中微公司/新宙邦/比亚迪/海澜之家/稳健医疗 5 家跨线须标「口径敏感」)。同步跑通 FactoryZCI 反查通道(A1-2 唯一闭环项):570 家链主 ×(当时仅 2,477 条的)官方名单,首跑确定命中 32 家、疑似候选 696 条——暴露「企业法人主体名 ≠ 工厂名单主体名」,须建五指数共用的「集团—工厂主体归集表」。
集团—工厂归集表 v1.2(本次升级):查明首跑反查命中率低(5.6%)有两个独立根因——① 名称鸿沟(集团名 ≠ 工厂法人名);② 名单域只覆盖 2025 年度一个批次,仅占工信部累计培育量的 24.4%。补全工信部 2017—2024 全部 9 批绿色工厂/绿色园区名单(工厂域 2,437 → 9,174 条)后,在完全不改匹配算法的前提下,链主命中即从 32 家升至 273 家;再加 2 字主体词的双重校验(行业残部一致性 + 同名异企排除)与降级留痕,采信层收敛至 216 家 / 542 个名单主体(采信层 L1/L2)。首谈 20 家的 A1-2 认证证据由 13 家提升到 17 家(85%),「贵司旗下已有 N 家工厂在官方名单中」这句话现在对 17 家都能说准。Entity Map v1.2 (upgraded): the low 5.6% hit rate had two independent root causes — the naming gap, and a list-domain gap (only one 2025 batch was loaded, 24.4% of MIIT cumulative totals). Loading all 9 historical batches (2017–2024; factory domain 2,437 → 9,174) lifted shortlist hits from 32 to 273 with no algorithm change. Two extra validations for 2-character brand cores (industry-remainder consistency and same-name-different-firm exclusion), plus a full demotion log, converged the trusted layer to 216 groups / 542 named entities. A1-2 evidence for the first-interview 20 rose from 13 to 17 (85%).首谈落地包 v1.0(本次新出):把结论落成两条可立即执行的动作。对外——企业确认版定向征询函,请求分两栏:A 栏分母补全(11 个汇总数,全部来自企业现成台账,不要名单、不要单价、无需新测)+B 栏证据补位(对「已公开但我方未检索到」的事项,只需指出文件与位置)。首谈 20 家逐家生成缺口单,每家只需确认 3—9 个字段(2 家只需 3 个),不做全员同表。核心设计:不问「你们做得好不好」,只问「一共多少家」——把审查性问询变为技术性核对。对内——299 家未归集链主的派工清单,P1 首谈名单内 3 家(45 分钟可完成)/P2 DCS≥85 139 家/P3 其余 157 家,每家附检索词与三步检索法。配套承诺 6 条:只收汇总数、不单独披露绝对值、双向标注、未检索≠未披露、可撤回、年度复用。
Caliber Manual v1.0: the real bottleneck is the denominator, not data access. Each of the 12 conditional metrics (55% of weight) now has a public-proxy caliber (six methods, each with disclosed assumptions) and a company-confirmed caliber (11 de-duplicated fields). A 54-name test shows stable ranks (rho 0.979) but shifting absolute bands. The FactoryZCI back-check channel for the only fully closed-loop metric (A1-2) is now operational: 32 confirmed hits and 696 candidates pending entity mapping.
Shortlist Action Pack v1.0 (new): two immediately executable moves. Outbound — a targeted inquiry letter in two sections: denominator completion (11 aggregate figures, all from existing records; no supplier lists, no unit prices, no new measurement) and evidence pointers (for items already public but not located by us). Each of the 20 shortlisted firms gets its own gap sheet needing only 3 to 9 fields. Core design: we ask how many, never how well. Internal — a dispatch list for the 299 unmapped chain leaders (3 priority-1 from the shortlist, 139 with DCS >= 85, 157 remaining), each with search terms and a three-step procedure.
查看集团—工厂归集表 →Entity Map v1.2 →
查看首谈落地包 →Action Pack v1.0 →
查看分母口径手册 →Caliber Manual →
查看评级可行性判定 →Rating Feasibility →
查看首谈名单 v1.0 →Shortlist v1.0 →
查看得分卡包 v1.0 →Scorecards v1.0 →
查看可得性核查 →Availability Check →
查看公开硬事实层 v1.0 →Public Hard-Fact Layer v1.0 →
查看候选池榜单 →View Pool →
零接触公开成果九件(全部零新增采集、零企业配合)Nine Zero-Contact Releases (all zero new collection, zero company involvement)
① 零碳制造资产盘点 v1.0:链主池 570 家中 216 家握有可核验的国家级绿色制造资产,合计 542 个认证主体(按名称全局去重,仅采信层 L1/L2);其中 16 家旗下已有零碳工厂(334 号文线,最稀缺信号);渗透率 37.9%,最高赛道汽车与出行 70.6%。
(1) Zero-Carbon Manufacturing Asset Inventory v1.0: 216 of 570 shortlist groups hold verifiable national green-manufacturing assets — 542 named entities globally deduplicated (trusted layer L1/L2 only); 16 already own zero-carbon factories.
② 评级共识与实体减碳行动一致性检验 v1.0:DCS 共识分与旗下认证主体数 弱正相关但逐档单调(Spearman ρ = 0.213;最高档有认证资产比例 52.7% vs 最低档 25.2%)。关键反直觉发现——市值与认证数几乎无关(r = 0.060),控制市值后各档内相关性依然为正,说明该关联不是「大公司工厂多」造成的。
(2) Rating-vs-Action Linkage Test v1.0: DCS consensus vs certified entities shows a weak but monotone positive relation (Spearman rho 0.213). Counter-intuitively, market cap is nearly unrelated (r 0.060) — the link is not a size artefact.
③ 园区与工厂指数公开可得性核查 v1.0:ParkZCI 公开可覆盖 37% 权重(C 级须台账 63%,核心层+引导层几乎全落在 C 级);FactoryZCI 仅 15%(三项一票核心 60% 权重全部不可得)。结论:两个指数都不能靠公开信息出本体评级;核查副产品是把园区函件从 28 项砍到 6 个数据包、工厂函件从 11 项砍到 5 个数据包。
(3) Park/Factory Public Availability Check v1.0: public coverage is 37% of ParkZCI weight and only 15% for FactoryZCI (all three veto-core metrics, 60% of weight, are internal). Neither index can be rated from public sources alone; the check shrinks the park letter from 28 items to 6 data packs.
④ 零接触模式 · 指数发布路线图 v1.0:一条判据定五个指数的命运——评价对象本身是否负有强制披露义务。强制披露主体(上市公司 / 公募 REITs)数据天然公开 → 3 个可发布(RatingZCI 与 REITsZCI-GP 正式可达、SupplyZCI 公开硬事实层 v1.0 已落地,可达预发布,覆盖 92.5% 权重);非披露主体(园区管委会 / 工厂法人)核心数据在内部台账 → ParkZCI 仅公开信号版(37.0%)、FactoryZCI 本体评级不可得(15.0%),改发名录版。
(4) Zero-Contact Publish Roadmap v1.0: one criterion decides all five indices — whether the evaluated entity is itself a mandatory-disclosure subject. Three can be published; ParkZCI and FactoryZCI can only be signal/roster products.
⑤ ParkZCI 公开信号版 v0.1:国家级 52 家零碳园区结构性图谱。关键发现——46 家(88%)的评价范围是「园中园」,评价边界是开发区内的子区域,导致指标的分母(面积 / 企业总数 / 能耗总量)在公开渠道无从查证;这不是数据未公开,而是分母不可外部确定,正是 63% 权重必须园区台账的结构性原因。另 21 家(40%)将于 2028 年前到期验收。
(5) ParkZCI Signal Edition v0.1: 46 of 52 national zero-carbon parks (88%) are evaluated as an “enclave within a park” — the denominator is not externally determinable, which is the structural reason 63% of weight requires park ledgers.
⑥ FactoryZCI 零碳工厂名录与培育漏斗 v1.0:省级零碳工厂 311 家 × 国家级绿色工厂(9 批 6,529 家)交叉——113 家(36.3%)已取得国家级绿色工厂认定,约三分之一走完了「绿色工厂 → 零碳工厂」这一步。含分省图谱、地市分布、培育漏斗与双认定清单。
(6) FactoryZCI Roster & Pipeline v1.0: 113 of 311 provincial zero-carbon factories (36.3%) already hold national green-factory status.
⑦ SupplyZCI 公开硬事实层 v1.0:把 v0.1 的「结构」做成「数据」,L2 层由两维扩到五维全覆盖。新增 ① 国家级绿色供应链管理企业认定——工信部绿色制造名单四大类之一(官方定义即「带动供应链上下游工厂实施绿色制造的主导企业」,与链主同题),此前完全缺失;本轮补齐 2019—2024 六批 699 家原始名单(源 PDF 已归档可复核),与池匹配得 47 家(下限:2017—2018 三批为扫描件未纳入);② ESG 报告披露状态——巨潮资讯公开公告机采,467 / 570 家 = 81.9% 已披露,含原文链接可直达报告。③④⑤ 沿用官方绿色制造资产(216 家)、评级共识、市值行业。三维交叉得证据密度谱:三重齐备 38 家 / 两维齐备 160 家 / 单维可查 296 家 / 公开层无证据 76 家。两项新维度零发函、零填报、零企业配合——因为证据本就写在链主自己必须发布的报告里。
(7) SupplyZCI Public Hard-Fact Layer v1.0: two new machine-collected dimensions. (a) National Green Supply-Chain Management Enterprise designation — one of the four MIIT green-manufacturing list categories, officially defined as the lead firm driving upstream and downstream factories toward green manufacturing, i.e. the same object SupplyZCI evaluates. All six batches 2019–2024 (699 names) were recovered from official PDFs (now archived); 47 pool matches, a lower bound since the three 2017–2018 batches are image-only scans. (b) ESG report disclosure — 467 of 570 (81.9%), each with a direct link to the report. Cross-tabulating all three evidence types yields an evidence-density map: 38 triple, 160 double, 296 single, 76 none. Both new dimensions required zero letters, zero forms and zero company involvement, because the evidence already sits in reports these firms are obliged to publish.
⑧ SupplyZCI 报告内页精读 v0.4:从「入口采集」推进到「正文取证」——467 份 ESG 报告逐页抽取,35 个字段 / 7 组行级匹配(v0.3 新增「分母口径」组),每条证据带页码与原文片段。三处升级:① 不可读样本全部补回——47 家(结构性乱码 5 / 英文版 5 / 图片型 34)经四层补数收口(私用区数字无损修回 3 + 同年度可读文本 6 + 同年度 OCR 17 + 换年度 21),0 家仍不可用,判定分母 420 → 467(+11.2%)。其中 5 家「英文版」实为采集器「取最新公告」误选(公司先发中文版、隔几天再发英文版),3 家「乱码」只是数字字形被搬进私用区、根本不需要 OCR。② 抽检抓到材料级缺陷——「温室气体排放总量」有值样本 42.2%(256 家中 108 家)实为废水/废气/二氧化硫/氮氧化物/有害废弃物排放总量,量纲同为「吨」,原有校验一条都拦不住;修后剔除纯跨介质 36 家、另捞回 7 家真值(长城汽车旧基线 5 条有值全是废水,而报告里写着「温室气体总排放量 153,244.61 吨」)。③ 换引擎即解决问题——CID 字体污染从「标记剔除」变成「彻底消除」,同一份曾 93.9% 不可读的报告恢复为 57,967 字完整可读,467 份处理从约 8 小时压到 302 秒;「表格式数据才是主流形态」——数值行落在触发词下方 1–3 行,补上「表格向下取数」通道后范围一/二「有值率」由 23%/18% 修正到 54%/56%。全量结果:范围三仅 20.3%(却是供应链排放的唯一入口);净零目标年 33.2%、碳达峰年 16.7%、SBTi 提及 17.3%;核心项 供应商碳数据收集覆盖率仅 2.1%、绿色采购占比 0%——指标树最核心的 D1-1 / A2-1 在公开报告里几乎没有供给。质量抽检:随机 12 家回原文复核,12/12 一致。v0.3 新增「分母口径」组:直击「有分子无分母」这一最大堵点——句式普查证实公开报告里唯一真实稳定的分母是「供应商总数」(101 家有值),采购总额与产品总数实测不可得(前者多为农产品采购、后者多为研发案例数),故不强行配分母。配套 5 类分子后,5 项比例型指标首次可独立复算(分子÷分母),与「报告自述百分比」分列呈现、不合并、不加分。为防止「看起来配上了」冒充「真的能算」,建立六道闸门:无分母/分母污染(子集句)/无分子/分母<50 家/分子越界/同源重复与量纲错配。闸门的价值在拦下假结果——实测分母污染 50 例、分子越界 3 例、同源重复 3 例、量纲错配 1 例;未做闸门时,分母会大量混入「通过 IATF 16949…供应商数量为 79 家」这类子集句,一算就 >100%。最终可复算 53 组 / 32 家,其中 3 项为近似口径(指标3 名义「行为准则含碳条款」,实测多为《廉洁承诺书》《质量保证协议》,以 ≈ 标注)。32 家即当前公开信息的可验证上限——这是本轮最重要的量化结论。v0.4 新增第十三节「证据强度闸门」:解决了另一半问题——不是「有没有抽到」,而是「抽到的东西够不够硬」。抽取层新增字段级闸门统计(命中行/强要件/词组级/否定句),且对全部命中行统计(不受「每家只留 4 条证据」的展示配额截断——否则强要件恰好落在第 5 行的样本会被误判成「只提到词组」)。实测:8 个线索字段里,「回收/循环利用体系」3,286 条命中中 2,846 条是词组级(86.6%)(多出现在「水资源循环利用」这类无关语境),而「供应商 ESG 评估与准入」376 条全部是强要件——「有值家数」与「有强要件家数」的落差,就是该字段的可信区间。三类缺陷两种处置:极性错误(均胜电子「暂未将气候绩效纳入高管薪酬考核」曾被判成已挂钩)属逻辑错误、直接改主分;词组/标题级命中(「尽职调查实施路径」是目录标题、「违反海外供应链尽职调查法」是风险描述)不属错误、只是证据强度不足,另算强证据口径并列呈现、不动主分。
(8) SupplyZCI Report Full-Text Extraction v0.2: from collecting entry links to extracting evidence from the body text — 467 ESG reports parsed page by page against 29 fields in 6 groups (two new groups: governance, disclosure platforms), every finding carrying a page number and quoted source text. Three upgrades: (a) all 47 unreadable samples recovered — 5 structural garbling / 5 English-only / 34 image-only, closed out through four tiers (3 private-use digit remaps needing no OCR, 6 same-year readable versions, 17 same-year OCR, 21 alternate-year), zero still unusable, lifting the denominator from 420 to 467 (+11.2%). The five "English-only" cases turned out to be a collector bug that always picked the latest announcement, and three "garbled" files merely had their digit glyphs remapped into the private-use area. (b) sampling caught a material defect — 42.2% of "total GHG emissions" values (108 of 256 firms) were actually wastewater, waste-gas, SO2, NOx or hazardous-waste totals; the unit is "tonnes" in every case, so unit checks caught nothing. 36 firms were dropped as pure cross-medium hits and 7 true values recovered (Great Wall Motor's old baseline had five values, all wastewater, while the report itself states 153,244.61 tonnes of GHG). (c) switching the PDF engine eliminated, rather than merely flagged, CID font corruption — a report that was 93.9% unreadable placeholder glyphs became 57,967 fully readable characters, and the whole 467-report run fell from about 8 hours to 302 seconds; a table-aware lookup channel lifted Scope 1/2 value rates from 23%/18% to 54%/56%. Scope 3 reaches only 20.3% even though it is the sole entry point to supply-chain emissions; supplier carbon-data coverage is just 2.1% and green procurement 0%, showing how thin public supply is for the core D1-1 and A2-1 metrics. Quality check: 12 randomly sampled values re-verified against source, 12/12 consistent.
⑨ SupplyZCI 证据到指标判定规则 v1.3:从「正文取证」再推进到「指标可判」——把字段级证据判定到 24 项末级指标,三档证据强度(E3 量化 / E2 要件 / E1 提及,合格线 E2)。21 项本层可判 + 1 项外部通道 + 2 项须企业侧,公开层可覆盖权重 92.5%,与上一轮按字段可得性独立测算的结果完全一致(交叉验证)。关键发现:若按原建议门槛(覆盖 ≥14/24 项),全池 467 家中只剩 21 家(4.5%)——公开层无法支撑一个够规模的绩效分比较池,故本版只发披露完备度 Coverage、不发 SC 评级;这不是回避,是诚实——门槛只有 4.5% 通过时,任何评级都是噪声。另一个结构性事实是权重倒挂:占权重最高的「减碳行动」(30%)平均达标率仅 10.2%,而覆盖率最好的「协同与赋能」(15%)达 45.4%(多为「提及即算」的软指标);真正量化的「供应商碳数据收集覆盖率」只有 2.1%。v1.2 新增第九节「比例复算」:把 12 项比例型指标里长期卡住的「有分子无分母」推进一步——全库句式普查证实唯一真实稳定的分母是「供应商总数」(101 家有值),采购总额/产品总数实测不可得,故不强行配分母;配套 5 类分子后 5 项指标首次可独立复算(53 组 / 32 家),并建立六道闸门分列「自述百分比」与「可验算结果」。闸门实测拦下分母污染 50 例(如「通过 IATF 16949…供应商数量为 79 家」这类子集句)、分子越界 3、同源重复 3、量纲错配 1——32 家即当前公开信息的可验证上限。本版直接建立在第 ⑧ 件的 467 家全量样本 / 35 字段与已修复字段之上(补数四层、跨介质假阳性剔除均见第 ⑧ 件),证据可逐条回溯至页码与原文。v1.3 新增第八节「证据强度分层」:把「依赖线索字段的达标率是上限估计」这句免责声明变成可测量的量——全部达标格中 1,613 格(53.4%)属于单点软支撑(该指标达标与否完全等价于某一个线索字段的命中与否,中间没有任何第二个字段做交叉验证),8 项 soft 关联指标全部 100% 单点依赖。并给出双口径:主分口径(仅修极性错误)21 家(4.5%)、强证据口径(再要求实质要件)15 家(3.2%)——差值 6 家即「靠词组级命中凑到门槛」的样本,换言之公开层能稳定识别的、有实质证据的链主约在 15 家量级,这个数比主分更接近「经得起复核」的口径。为什么不直接把主分改成强证据口径?因为那等于「为了让数字好看而收紧口径」——本项目的纪律是逻辑错误直接改分、口径宽严只做分列,把宽口径留着,读者才知道不确定性有多大。
(9) SupplyZCI Evidence-to-Indicator Judging Rules v1.1: from extracting evidence to deciding whether that evidence can support an indicator. Field-level evidence is mapped onto 24 end indicators through three evidence grades (E3 quantified / E2 documented / E1 mentioned, threshold E2). Twenty-one indicators are judgeable in this layer, one runs on an external channel and two require company-side data; publicly coverable weight is 92.5%, exactly reproducing the independent estimate from the previous round. The pivotal finding: under the originally proposed threshold of 14 of 24 indicators, only 21 of the 467 firms (4.5%) would qualify — public sources cannot support a performance-scoring pool of usable size, so this release publishes Coverage only and no SC rating. When just 4.5% clear the bar, any rating is noise. A second structural fact is the weight inversion: the highest-weighted dimension, Decarbonisation Action (30%), averages a 10.2% pass rate, while the best-covered dimension, Collaboration (15%), reaches 45.4% — largely because the latter rests on soft "mentioned counts" fields; the one genuinely quantified metric, supplier carbon-data coverage, sits at 2.1%. This release is built directly on the full 467-firm sample and the corrected fields of item (8), so every figure traces back to a page number and a quoted source.
查看公开硬事实层 →Hard-Fact Layer →
查看发布路线图 →Publish Roadmap →
查看园区信号版 →Park Signal →
查看工厂名录 →Factory Roster →
查看资产盘点 →Asset Inventory →
查看一致性检验 →Linkage Test →
查看可得性核查 →Availability Check →
查看报告内页精读 →Report Insight →
查看判定规则 →Judge Rules →
查看绿供名单结构分析 →List Structure →
政策锚点:Policy Anchor: 工信部 2026 年 8 月 17 日印发《关于组织开展国家级零碳工厂建设工作的通知》(工信厅节函〔2026〕334号),零碳工厂建设进入全面实施阶段。三项核心指标一票否决:单位能耗碳排放(基本≤2、目标≤0.2 吨CO₂/吨标煤)、非化石能源消费占比(基本≥30%、目标≥95%)、非化石能源电力物理认定量占比(基本≥10%、目标≥35%,算力设施除外)。零碳算力设施纳入建设范围。"十五五"目标培育建设约 500 个国家级零碳工厂。本指数系列据此校准评分基准。MIIT Aug 17, 2026 Circular 334 enters full implementation. Three core veto indicators calibrate the index scoring baseline. Target: 500 national zero-carbon factories.