Watch brief · standing · 2026 → 2050
观察简报 · 长期跟踪 · 2026 → 2050
Twenty of the heaviest carbon balance sheets on earth, read from their own annual and CDP disclosures. Ten more waiting in the wings. Thirty companies in the field of view — and one honest question repeated every quarter: what actually moved?
全球最重的二十份碳资产负债表，只读它们自己公开的年报与 CDP 披露；另有十家在候补名单上。视野里共三十家企业，每个季度重复同一个诚实的问题：到底什么真的动了？
It isn't an evaluation. It's a watching — for attention.
这不是评价，这是一种注视——为了把注意力留在这里。
Grouped by where the physics sits, not ranked by tonnes. Rankings age badly and invite argument; the sector block and the dominant emission category tell you far more about what has to change. Every figure behind these names should be pulled from the company's own latest annual report, sustainability data book or CDP response — never from a summary, ours included.
按“物理过程发生在哪里”分组，而不是按吨数排名。排名会过时，也容易引发争论；行业分组与主导排放类别更能说明什么必须改变。名字背后的每一个数字，都应回到该公司最新的年报、可持续数据册或 CDP 问卷中去取——不要引用任何二手汇总，包括本页。
“Good” here means the mechanism is transferable — something another company could copy on Monday. It does not mean the company is clean. A heavy emitter with one excellent lever is still a heavy emitter; that is exactly why the lever is worth stealing.
这里的“好”指的是机制可迁移——别的公司周一就能照做。它不代表这家公司干净。一个高排放企业拥有一个出色的杠杆，仍然是高排放企业；也正因如此，这个杠杆才值得学。
Plant-by-plant closure and conversion schedules, published, then rebuilding the same balance sheet around renewables and grids. Transferable move: put the retirement calendar in the annual report, not the sustainability report.
逐厂公布关停与改造时间表，然后用同一张资产负债表重建可再生与电网业务。可迁移之处：把退役日历写进年报，而不是可持续报告。
DRI-EAF investment with hydrogen-ready design, plus a certified low-emission steel product line so the premium has somewhere to land. Transferable move: pair the asset with a sellable product, or the capex never clears.
投资直接还原铁—电炉路线并预留氢就绪设计，同时推出经认证的低排放钢产品线，让溢价有落点。可迁移之处：资产必须配一个能卖的产品，否则投资永远批不下来。
Calcined clay and supplementary materials to cut the clinker factor, plus demolition concrete recycled back into new mixes. Transferable move: attack the material formula before you attack the fuel.
用煅烧黏土与辅助胶凝材料降低熟料系数，并把拆除混凝土回收进新配方。可迁移之处：先动材料配方，再动燃料。
The inert-anode joint venture aims at smelting that emits oxygen instead of CO₂ — process emissions removed, not offset. Transferable move: fund the one R&D bet that makes a whole emission category disappear.
惰性阳极合资项目的目标是让电解过程排出氧气而非 CO₂——工艺排放被消除，而不是抵消。可迁移之处：把钱押在能让整个排放类别消失的那项研发上。
Cradle-to-gate product carbon footprints calculated across a catalogue of tens of thousands of items, so customers can design with the number. Transferable move: make PCF a data field in the ERP, not a project.
对数万种产品逐一计算摇篮到大门的产品碳足迹，让客户能拿着这个数字做设计。可迁移之处：把 PCF 做成 ERP 里的一个数据字段，而不是一个项目。
Large-scale renewable procurement for fabs alongside abatement of high-GWP process gases — the two levers that actually control a semiconductor footprint. Transferable move: solve Scope 2 and process Scope 1 in the same plan.
为晶圆厂大规模采购绿电，同时对高 GWP 工艺气体做减排处理——这才是半导体足迹的两个真正开关。可迁移之处：把范围二与工艺范围一放在同一份方案里解决。
Supplier clean-energy commitments plus recycled aluminium and rare earths designed into the product. Transferable move: make renewable supply a condition of the purchase order, then measure recycled content per part.
要求供应商承诺使用清洁电力，并把再生铝与再生稀土设计进产品。可迁移之处：把绿电写成采购订单的条件，再按零件统计再生材料含量。
An internal carbon fee that lands on each division's P&L and funds abatement and durable removal contracts. Transferable move: give the number a currency, and behaviour follows the invoice.
内部碳费直接计入各事业部损益，并用于资助减排与长期碳移除合同。可迁移之处：给碳一个货币单位，行为会跟着账单走。
Dual-fuel methanol vessels ordered ahead of supply, with a customer-paid low-emission shipping product creating the demand signal. Transferable move: buy the asset and sell the premium in the same year.
在燃料供应成熟之前先订双燃料甲醇船，并以客户付费的低排放运输产品制造需求信号。可迁移之处：同一年内既买下资产，也卖出溢价。
Farm-level regenerative sourcing programmes for dairy and coffee, accounted separately under FLAG rules. Transferable move: for food, the factory is a rounding error — go to the farm and report land separately.
面向乳业与咖啡的农场级再生农业采购项目，并按 FLAG 规则单独核算。可迁移之处：食品行业的工厂只是零头——要走到农场，并把土地排放单列。
Co-funded technology partnerships with steelmaking customers, because the ore seller's Scope 3.10 is the mill's Scope 1. Transferable move: fund abatement one step down the chain and share the accounting.
与钢铁客户共同出资开展技术合作——因为矿商的范围三.10 就是钢厂的范围一。可迁移之处：出钱去减下游一步的排放，并共享核算成果。
Patterns, not accusations. Each of these shows up somewhere on the board, and each of them is a place where a good report and a real transition quietly diverge.
这里写的是规律，不是指控。每一条都能在名单上找到对应，也都是“一份好报告”与“一次真转型”悄悄分岔的地方。
A 2050 pledge with no 2030 capex line attached is a sentence, not a plan. Follow the money, not the ambition statement.
一个没有 2030 资本开支科目支撑的 2050 承诺，只是一句话，不是计划。要跟着钱走，而不是跟着愿景走。
In energy and mobility, Scope 3.11 dwarfs everything else. A target that excludes the use of sold products is a rounding error dressed as leadership.
在能源与交通行业，范围三.11 压倒一切。一个不包含“售出产品使用”的目标，只是被包装成领导力的零头。
Intensity can fall while absolute tonnes rise. Always read both, and read them against the same baseline year.
强度下降的同时，绝对吨数可能在上升。两者都要看，并且要用同一个基准年去看。
Market-based Scope 2 built on unbundled certificates flatters the number and moves no electron. Hourly, local matching is the honest version.
建立在非捆绑绿证之上的市场法范围二只是让数字好看，并没有让一个电子移动。逐小时、同区域的匹配才是诚实版本。
Offsets booked before abatement delay the engineering. Removals belong at the end of the sequence, priced as a liability, not worn as a badge.
在减排之前先记抵消，只会推迟工程本身。碳移除应放在顺序的最后，作为负债定价，而不是当成勋章佩戴。
Restated baselines and shifting organisational boundaries erase progress you cannot see. Check the restatement note before you compare two years.
基准年重述与组织边界变动，会抹掉你看不见的进展。比较两个年份之前，先看重述说明。
Divesting a coal plant is not a reduction. The molecules changed owner, not fate — and often changed to an owner who discloses less.
卖掉一座燃煤电厂并不是减排。分子只是换了主人，命运没变——而且新主人往往披露得更少。
Most Scope 3.1 is still spend-based. Secondary factors smooth the curve and hide the exact hotspot you were hired to find.
大多数范围三.1 仍是基于支出法。二次因子把曲线抹平，也把你受雇去找的那个热点藏了起来。
Land and FLAG emissions arrive late and are assured weakly. Deforestation risk usually sits three tiers deep, in traders you have never met.
土地与 FLAG 排放总是来得最晚、鉴证最弱。毁林风险通常藏在三级以外的贸易商那里，你从未见过他们。
Eco-design claims made at launch are rarely re-measured at end-of-life. Recyclable is a property; recycled is an outcome. Report the outcome.
发布时的生态设计承诺，很少在产品寿命终点被复核。“可回收”是属性，“已回收”才是结果。请报告结果。
Sustainability sitting beside the P&L instead of inside it. Until the transition cost is in the product cost and the buyer's KPI, nothing structural moves.
可持续被放在损益表旁边，而不是里面。除非转型成本进入产品成本与采购 KPI，否则结构性的东西不会动。
Carbon is the loudest number, not the only limit. For half of this board, water and land will bind before carbon does. Four questions to carry into every disclosure you read.
碳是最响的那个数字，但不是唯一的边界。对名单上一半的公司来说，水和土地的约束会先于碳到来。带着这四个问题去读每一份披露。
Ask for consumption, not withdrawal, and ask for it weighted by basin stress — a litre in Hsinchu, Arizona or the Atacama is not a litre in Norway. Semiconductors, mining, textiles and beverages carry the real exposure. Cross-read the CDP Water Security response against site locations.
要耗水量，而不是取水量，并且要按流域压力加权——新竹、亚利桑那或阿塔卡马的一升水，不等于挪威的一升水。半导体、矿业、纺织与饮料承担着真正的风险。把 CDP 水安全问卷与厂址位置对照着读。
Split hazardous from non-hazardous, and diversion-to-energy from true material recovery. Steel slag, fly ash, spent pot lining, mine tailings and battery scrap are the four places where waste becomes someone else's feedstock — or a century-long liability.
把危废与一般废弃物分开，把“焚烧发电”与真正的物质回收分开。钢渣、粉煤灰、废阴极、尾矿与电池废料，是废弃物变成别人原料——或变成百年负债——的四个地方。
Deforestation- and conversion-free volumes by commodity, with cut-off dates stated. Watch for TNFD-style location screening, mine closure and rehabilitation provisions, and whether biodiversity commitments have a site list attached or only a policy.
按商品披露“无毁林、无生态转换”的量，并写明截止日期。关注 TNFD 式的地点筛查、矿山闭坑与复垦拨备，以及生物多样性承诺背后是否附有具体场址清单，还是只有一纸政策。
Around eighty per cent of a product's impact is fixed at the design table. Look for PCF per SKU, recycled content per part, design for disassembly and repair, and readiness for digital product passports. A rule of thumb: if the designer cannot see the number in the tool, it does not exist.
产品约八成的影响在设计阶段就被锁定。要看每个 SKU 的碳足迹、每个零件的再生材料含量、可拆解与可维修设计，以及数字产品护照的准备度。一条经验：如果设计师在工具里看不到这个数字，它就等于不存在。
The only durable indicator: the share of revenue and capex from low-carbon products and services, trending year on year. Everything else is process. Ask whether green capex is additive or cannibalising the legacy line — the second is the real transition.
唯一经得起时间检验的指标：低碳产品与服务在收入与资本开支中的占比，以及它的逐年趋势。其余都只是过程。还要问：绿色资本开支是叠加的，还是在吞噬旧业务线——后者才是真正的转型。
Find the executive whose bonus moves with the tonnes, and the board committee that reviews the transition plan. Then check the assurance statement: limited or reasonable, and over which scopes. Reasonable assurance on Scope 3 is still rare enough to be a signal.
找出奖金随吨数变动的那位高管，以及审阅转型计划的董事会委员会。然后看鉴证声明：是有限保证还是合理保证，覆盖哪些范围。范围三上的合理保证仍然稀少到足以成为一个信号。
The first decade is decided by capital allocation and regulation. The second is decided by whether the business model itself changed. Read every transition plan against these two clocks.
第一个十年由资本配置与监管决定；第二个十年取决于商业模式本身是否改变。用这两个时钟去读每一份转型计划。
Horizon one
第一段
2026 — 2035
The decade the capex decides
由资本开支决定的十年
Horizon two
第二段
2035 — 2050
The decade the model changes
商业模式改变的十年
AI does not decarbonise anything. It shortens the distance between a measurement and a decision — which is where most transition programmes actually stall. Eight places it earns its keep, and one place it costs.
AI 本身不减排。它缩短的是“从测量到决策”的距离——而大多数转型项目恰恰卡在这一段。以下八个地方它值回票价，最后一个地方它要付出代价。
Language models classify millions of purchase-order lines into activity categories, so ledger data can be mapped to physical units and real factors. Typically the fastest route from a spend estimate to a hotspot you can act on.
用语言模型把数百万条采购订单行归类到活动类别，让账务数据映射到实物单位与真实因子。这通常是从“支出估算”走到“可行动热点”的最快路径。
Extraction from PDFs, certificates and EPDs, with confidence scores and every gap-fill explicitly flagged as a proxy. The discipline is the flagging, not the extraction.
从 PDF、证书与 EPD 中抽取数据，带置信度，并把每一次缺口填补明确标注为替代值。真正的纪律在于标注，而不是抽取。
Model-predictive and reinforcement control on setpoints delivers a few per cent of fuel or power with no new asset — the cheapest tonne available to heavy industry, and available now.
对设定点做模型预测与强化学习控制，能在不新增资产的前提下省下几个百分点的燃料或电力——这是重工业当下能拿到的最便宜的一吨。
Satellite imagery plus plume-detection models attribute leaks to a site and an operator within days. Reconcile against the company's own reported methane intensity — the gap is the story.
卫星影像加羽流识别模型，可在数天内把泄漏归因到具体场站与运营方。再与公司自报的甲烷强度对账——差距本身就是故事。
Forecasting site load against grid carbon intensity, then shifting flexible loads — compute, electrolysis, cold storage, charging — into the clean hours.
预测厂区负荷与电网碳强度，把可调负荷——算力、电解、冷库、充电——挪到清洁时段。
Machine-learning screening of low-clinker binders, catalysts, electrolytes and bio-based polymers compresses years of laboratory iteration into months of candidate ranking.
用机器学习筛选低熟料胶凝材料、催化剂、电解液与生物基高分子，把数年的实验迭代压缩成数月的候选排序。
Network design, load consolidation, routing and vessel speed optimisation — logistics emissions fall through utilisation long before they fall through fuel.
网络设计、拼载、路径与船速优化——物流排放靠装载率下降的速度，远快于靠换燃料。
Screening LCA inside CAD and PLM, so a designer sees the carbon, water and recyclability consequence of a material choice while the choice is still cheap to change.
在 CAD 与 PLM 内做筛查型 LCA，让设计师在“改动还便宜”的时候，就看到材料选择带来的碳、水与可回收性后果。
Training and inference carry electricity, water and hardware embodied carbon. Put AI's own footprint on the same page as its savings, with the fab and the datacentre included. A use case that cannot show net benefit is a demo, not a programme.
训练与推理都带来电力、水与硬件隐含碳。把 AI 自身的足迹与它带来的节省放在同一页上，并把晶圆厂和数据中心一起算进去。无法证明净收益的用例，只是演示，不是项目。
The method first, because the source is useless without it. Boundary, then activity data, then factor, then assurance. Every number you publish should survive the question: which page of which document, and which factor vintage?
先讲方法，因为没有方法，来源就没用。先定边界，再取活动数据，再选因子，最后看鉴证。你发布的每个数字都应经得起这句追问：出自哪份文件的哪一页，用的是哪一版因子？
GHG Protocol Corporate Standard for Scope 1 and 2; the Corporate Value Chain Standard for the fifteen Scope 3 categories. State the consolidation approach — operational control, financial control or equity share — because it changes the answer more than any factor does.
范围一、二用 GHG Protocol 企业标准；范围三的十五个类别用价值链标准。必须写明合并方法——运营控制权、财务控制权还是股权比例——它对结果的影响比任何因子都大。
Rank your data: metered, then supplier-specific, then average-data, then spend-based. Record the factor source and its vintage next to every line. A footprint without a factor register cannot be audited or improved.
给数据分级：计量数据 > 供应商专属 > 平均值 > 支出法。每一行都要记录因子来源与版本年份。没有因子台账的足迹，既无法鉴证，也无法改进。
Land-sector and FLAG guidance for food and agriculture; ISO 14067 and cross-industry product-footprint rules for PCFs; sector-specific SBTi pathways for steel, cement, aluminium, power, shipping and buildings.
食品与农业用土地部门与 FLAG 指南；产品碳足迹用 ISO 14067 与跨行业产品足迹规则；钢铁、水泥、铝、电力、航运与建筑用 SBTi 的行业专属路径。
Limited or reasonable assurance, and over exactly which scopes and categories. Then read the restatement note. Those two pages decide whether the previous forty are comparable.
看是有限保证还是合理保证，覆盖哪些范围与类别。然后读重述说明。就是这两页，决定了前面四十页能不能比较。
The source register
数据来源台账
No external links are printed on this page by design. Retrieve each source from its own publisher, search on the company's legal entity name plus the reporting year, and record the exact document title, page and access date in your working file. A benchmark is only as strong as its citation trail.
本页刻意不印任何外部链接。请到各来源的官方发布方处自行检索：用公司法定名称加报告年度搜索，并在工作底稿中记录确切的文件标题、页码与访问日期。一份对标的强度，取决于它的引用链条。
Candidates for the second round, chosen to close the blind spots this board leaves open: fertiliser chemistry, aviation, textiles, private agricultural traders, and the grids and datacentres now growing fastest.
第二轮的候选名单，专门用来补上现有名单留下的盲点：化肥化学、航空、纺织、私营农业贸易商，以及当前增长最快的电网与数据中心。
Or go industry-first instead. Five sectors where a single well-built benchmark would teach more than ten company files: nitrogen fertiliser, aviation fuel, textiles and dyeing, datacentre power and cooling, and marine fuel infrastructure. Company benchmarks tell you who is moving; sector benchmarks tell you whether movement is even possible yet. 或者，改成以行业为单位。有五个行业，做一份扎实的行业对标，收获会超过十份公司档案：氮肥、航空燃料、纺织与染整、数据中心的电力与冷却、船用燃料基础设施。公司对标告诉你谁在动；行业对标告诉你，动，究竟是否已经可能。
A watch is only worth keeping if it has a rhythm. Three commitments, small enough to actually hold.
一份观察，只有形成节奏才值得持续。三个承诺，小到真的能坚持。
Refresh the board from newly published reports. Record what moved, what was restated, and what quietly disappeared from the table.
依据新发布的报告刷新名单。记录什么动了、什么被重述、什么悄悄从表里消失了。
One company opened properly: transition plan, capex, product footprints, water and land, and the single decision that would change the trajectory.
认真打开一家公司：转型计划、资本开支、产品足迹、水与土地，以及那个能改变轨迹的关键决策。
Cite the page. Name the factor. Keep the tone of a colleague, not a judge — the point is attention, and attention is a form of respect.
标明页码，写清因子。保持同行的语气，而不是裁判的语气——目的是注意力，而注意力本身就是一种尊重。
Nobody here is finished. That is the whole reason to keep watching.
这里没有一家已经完成。这正是继续注视的理由。
Benchmark in progress — open for contributions 对标进行中 · 欢迎补充与指正