Materials → waste → materials → waste. Everyone can draw the arrow. Almost nobody can weigh it, trace it, and prove it. This station is a clickable field manual: open a box, run the numbers one step at a time, mark the hotspot, tick the evidence.
材料 → 废弃 → 材料 → 废弃。箭头谁都会画，能称重、能追溯、能举证的却极少。这是一份可点击的现场手册：打开盒子，一步一步算数，标出热点，勾选证据。
“You cannot close a loop you have not weighed.”
「没有称重过的循环，闭不上。」
All figures on this page are a worked illustrative plant, not a client disclosure. Replace with your own weighbridge tickets.
本页数字为演示用示例工厂，非任何客户披露数据，请以自家地磅单替换。
Five metrics carry almost all of the zero-waste and circularity conversation. Open each: the formula, what is inside the boundary, what is not, and what counts as good.
零废弃与循环议题，几乎都压在这五个指标上。逐个打开：公式、边界内、边界外，以及「多少算好」。
Tier bands follow the UL 2799 landfill-waste-diversion families (silver / gold / platinum) and the TRUE 90% floor. Verify the current edition before you print a logo.
分级参考 UL 2799 的填埋分流等级族（银/金/白金）与 TRUE 的 90% 底线。印 logo 之前，请核对现行版本。
A plant at 98% diversion where 30% is burned for energy is not circular — it is a well-managed incinerator feed. The gap between the two percentages is exactly your remaining design work.
分流率 98%、其中 30% 拿去焚烧发电，这不是循环，只是被管理得很好的焚烧料源。两个百分比之间的差距，正好等于你还没做的设计工作。
Where the gap usually hides差距通常藏在哪Yield tells you how much you machined away. Loop retention tells you whether that swarf came back as the same alloy or fell a grade. A 26% yield with 91% loop retention is a good factory. A 26% yield with downcycled chips is a leak.
成材率说明你切掉了多少；闭环留存率说明这些屑料是回到同一牌号，还是降级了。26% 成材率 + 91% 闭环留存，是好工厂；26% 成材率 + 屑料降级，就是漏点。
Typical ranges by process各工艺的常见区间Stop running two projects. One data spine feeds both: part number → material → mass → supplier site → destination. Add a destination column and it answers waste. Add a recycled-fraction column and it answers circularity. Add an emission factor column and it answers carbon.
别再开两个项目。一条数据主干同时喂两边：料号 → 材质 → 质量 → 供应商厂址 → 去向。加一列「去向」，它回答废弃物；加一列「再生比例」，它回答循环性；加一列「排放因子」，它回答碳。
If your BOM, your waste manifest and your carbon model cannot be joined on a shared key, you do not have a sustainability programme — you have three spreadsheets and a deadline.
如果 BOM、废弃物清运单和碳模型无法用同一个主键连接起来，你手上就不是可持续项目，而是三张表加一个交期。
Every number below is built in front of you. Press Next step and watch the denominator appear before the answer — that is where most claims break.
下面每个数字都当场推导。点下一步，先看到分母、再看到答案——多数声明就是死在分母上。
Fix the boundary and the period. Plant A, final assembly + CNC, one calendar quarter. Everything that crosses the gate on a weighbridge ticket is in. Wastewater to the municipal works is out, and that exclusion is written into the report.
先定边界与周期。A 厂，总装 + CNC，一个日历季度。凡是带地磅单出厂的都计入；排入市政的废水不计入，并在报告中写明该排除项。
Weigh every stream. Tonnes for the quarter:
逐流称重。本季度（公吨）：
Subtract only the two disqualifiers. 1,000 − 14 − 4 = 982 t diverted.
只减掉两项「不合格去向」。1,000 − 14 − 4 = 分流 982 公吨。
Now the honest number. Strip out burning: (812 + 96) ÷ 1,000 = 90.8% material recovery. Thermal share = 74 ÷ 1,000 = 7.4%, inside the usual 10% cap.
再算诚实的那个数。剔除焚烧：(812 + 96) ÷ 1,000 = 物质回收 90.8%。焚烧占比 = 74 ÷ 1,000 = 7.4%，在常见的 10% 上限之内。
Test the claim. Diversion 98.2% → gold band, not platinum: platinum needs 100% diversion from landfill. Landfill is not zero, so "zero waste to landfill" cannot be printed. Claimable: 98.2% diversiongold tier7.4% energy recovery
检验声明。分流 98.2% → 金级，不是白金：白金要求填埋分流达 100%。填埋不为零，因此「零废弃填埋」不能印。可声明：分流 98.2%金级能量回收 7.4%
What kills this at audit: a weighbridge ticket without a final-destination facility name; a broker who "recycles" but ships to a landfill 200 km away; on-site burning of pallets; a hazardous stream missing from W_total.
审核时最致命的：地磅单上没有最终处置设施名称；中间商声称回收却把料运到 200 公里外填埋场；厂内焚烧栈板；W_total 里漏掉一条危废流。
Pull the BOM by mass, not by part count. Four enclosure parts, one AI notebook.
按质量拉 BOM，不是按件数。一台 AI 笔电，四个外壳件。
Attach a verified fraction to each — with its evidence type.
为每件挂上已验证比例——并标注证据类型。
Choose the denominator — out loud.
公开选定分母。
Split PCR from PIR. PCR = 240 + 157.5 + 30 = 427.5 g → 62.4% of the set. PIR = 28.5 g → 4.2%. Most eco-labels only credit PCR, so the headline number is 62.4%, not 66.6%.
把 PCR 与 PIR 分开。PCR = 240 + 157.5 + 30 = 427.5 g → 组件的 62.4%；PIR = 28.5 g → 4.2%。多数环保标章只认 PCR，所以对外主打数字是 62.4%，不是 66.6%。
Guardrails before you publish. Mass-balance credits reconciled input vs output within ±2%; no credit sold to two customers; ingot lot numbers on the mill certificate match the part's traveller; the 75% is a batch average, so declare the tolerance (e.g. 75% ±5).
发布前的护栏。质量平衡额度的进出账差在 ±2% 内；同一额度未卖给两个客户；铝锭炉号与零件流程卡对得上；75% 是批次均值，须声明公差（如 75% ±5）。
One chassis, one billet. Billet in 1.28 kg → finished part 0.33 kg → swarf 0.95 kg.
一个机身、一块料。投料 1.28 kg → 成品件 0.33 kg → 切屑 0.95 kg。
Follow the swarf, not the invoice. 0.86 kg is dry, single-alloy, segregated at the machine → returns to the same 6013 remelt. 0.09 kg is coolant-soaked and mixed → sold as low-grade scrap.
跟着切屑走，不要跟着发票走。0.86 kg 干燥、单一牌号、机台旁即分类 → 回到同一 6013 重熔；0.09 kg 含切削液且混料 → 当低级废料卖掉。
Price the difference in carbon. Primary aluminium sits around 16 kg CO₂e per kg globally; remelted secondary aluminium is well under 1. Every kilo you keep in the same alloy avoids roughly 15 kg CO₂e.
把差额换成碳。原生铝全球平均约 16 kg CO₂e/kg，再生重熔铝远低于 1。每保住 1 公斤同牌号铝，约避免 15 kg CO₂e。
Then attack the 25.8%. Retention is a design number, not a housekeeping number: near-net extrusion or forged pre-form lifts yield to 45–60% and cuts machining energy with it. Recycling swarf brilliantly is second prize; not making it is first.
再回头攻 25.8%。成材率是设计指标，不是现场卫生指标：近净成形挤压或锻造预成形可把成材率提到 45–60%，同时降低加工用电。把切屑回收得漂亮是第二名，不产生切屑才是第一名。
Evidence for this loop: machine-side segregation SOP with photos, dry-swarf moisture spec, centrifuge/briquetter records, alloy certificate from the remelter naming your material back, and a mass reconciliation that ties 0.86 kg out to 0.86 kg in.
这条回路的证据：机台旁分类 SOP（附照片）、干屑含湿规格、离心/压块机记录、重熔厂出具的合金证明（写明是你的回料）、以及 0.86 kg 出与 0.86 kg 入的物料平衡对账。
Type your own tonnes for one period. The tier logic updates as you go — including the moment energy recovery breaks the 10% cap.
填入你自己一个周期的公吨数。等级判定会即时更新——包括焚烧发电突破 10% 上限的那一刻。
An AI PC's footprint is mostly decided before your gate opens. Six hotspots, ranked by typical share of product carbon. Tap one — the chemistry, the metric that moves it, the lever, and the document to demand.
一台 AI PC 的足迹，大半在进你厂门之前就定了。六个热点，按产品碳排的典型占比排序。点开任一个——化学式、可撬动的指标、杠杆、以及该索取的文件。
UPSTREAM → mine · refine · fab → TIER 1/2 → components → INSIDE THE FENCE → assemble · test · pack → USE → recover → back to refine
上游 → 采矿 · 精炼 · 晶圆 → 一/二级 → 元件 → 厂内 → 总装 · 测试 · 包装 → 使用 → 回收 → 回到精炼
Bigger dies, more DRAM, an NPU on top: every generation of AI PC pushes this rank upward. Two drivers — enormous electricity for tools and cleanrooms, and fluorinated gases used to etch and clean chambers. Those gases have staggering global warming potentials.
更大的晶粒、更多 DRAM、再加一颗 NPU：每一代 AI PC 都把这一项往上推。两个驱动因子——机台与无尘室的巨量用电，以及蚀刻与腔体清洗用的含氟气体。这些气体的全球增温潜势极其惊人。
Read the numbers读一下数字CF₄ ≈ 7,380C₂F₆ ≈ 12,400NF₃ ≈ 17,400SF₆ ≈ 25,200CH₄ ≈ 30N₂O ≈ 273
GWP over 100 years, IPCC AR6 order of magnitude. One kilogram of leaked SF₆ ≈ twenty-five tonnes of CO₂. This is why a leak survey beats a lighting retrofit.
100 年 GWP，量级依 IPCC AR6。泄漏 1 公斤 SF₆ ≈ 25 公吨 CO₂。这就是为什么一次泄漏排查胜过一轮换灯。
The abatement chemistry, in two lines减排化学，两行讲完Note the loop closing on itself: destroy an F-gas and you create fluoride to scrub, and calcium fluoride sludge to divert. A carbon win becomes a waste line. Both belong on the same table.
注意这个回路自己咬住了尾巴：销毁一种含氟气体，就产生要洗涤的氟化物，以及要分流的氟化钙污泥。碳的胜利变成一条废弃物流。两者必须放在同一张表上。
Levers, with typical effect杠杆与典型效果Hall–Héroult smelting is electricity plus a consumed carbon anode. When alumina runs low in the cell, an anode effect fires and cryolite fluorides attack the anode carbon — that is where PFCs are born.
霍尔–埃鲁法电解 = 大量电力 + 被消耗的碳阳极。当电解质中氧化铝浓度过低，发生阳极效应，冰晶石中的氟与阳极碳反应——PFC 就是这样生成的。
A PCB is copper, glass-reinforced laminate and a long chemical tail: etchants, plating baths, gold, and solder. Mass is a few hundred grams; impact is disproportionate — and most of the waste is liquid and metallic, which means it is recoverable if you keep it segregated.
一片 PCB 是铜、玻纤基材，加上一条长长的化学尾巴：蚀刻液、电镀槽、金、焊料。质量只有几百克，影响却不成比例——而且大部分废弃物是液体与金属，只要分流做好就能回收。
Cell manufacturing is dominated by dry-room dehumidification, electrode drying and formation energy, plus nickel and cobalt refining upstream. Solvent recovery is the classic hidden hotspot: NMP is expensive and regulated, so it is usually recovered — verify the recovery rate rather than assume it.
电芯制造以干燥房除湿、极片干燥与化成用电为主，加上上游镍钴精炼。溶剂回收是经典隐性热点：NMP 昂贵且受管制，通常会回收——请验证回收率，不要假设。
Levers杠杆Panel glass melting runs continuously at very high temperature; indium tin oxide is a scarce-metal thin film; and panel etch/clean historically used SF₆ and other F-gases. Ask which gases, how much, and whether abatement is fitted — this is exactly the question most PC brands never put in writing.
面板玻璃熔炉长期高温连续运转；ITO 是稀有金属薄膜；面板蚀刻与清洗历史上使用 SF₆ 等含氟气体。请追问：用哪些气体、用量多少、是否装设减排——这恰恰是多数 PC 品牌从不书面追问的问题。
This hotspot ranks last on carbon and first on tonnage crossing your gate. It is also the fastest visible win, which makes it the honest place to start building the data habit — as long as you say plainly that it is the small end of the carbon story.
这个热点在碳上排最后，在过磅吨数上排第一。它同时是最快见效的可视胜利，因此很适合用来养成数据习惯——前提是坦白说明：它在碳的故事里属于小头。
A claim is only as strong as the document behind it. Mark each yes or no — the meter at the bottom tells you whether to book a verifier or go back to the weighbridge.
声明的强度，等于它背后那份文件的强度。逐项标记「是/否」——底部的读数会告诉你：该约验证机构，还是先回地磅房。
You cannot map everything in year one, and pretending otherwise is how programmes die. Use a threshold, not a mood.
第一年不可能全部盘完，硬撑正是项目死掉的方式。用阈值判断，不要凭心情。
Never write "excluded". Write "deferred to Q3, threshold 1% of mass, owner: plant EHS". An auditor forgives a gap with a date on it. They do not forgive a silence.
永远不要写「排除」。要写「延至 Q3，阈值为质量的 1%，责任人：厂 EHS」。审核员会接受一个标了日期的缺口，但不会接受沉默。
When you have no data at all: use a secondary database factor or a supplier proxy, score the data quality 1–5, and record it. Then apply the stop rule — if a driver represents more than 20% of the impact and rests only on proxy data, you may not publish a claim about it. Go get the primary number, or say nothing about that part.
当你完全没有数据：先用二次数据库因子或供应商代理值，给数据质量打 1–5 分并记录在案。然后启动停止规则——若某驱动因子占影响 20% 以上，且仅依赖代理数据，就不得对外发布相关声明。要么拿到一手数据，要么对那部分闭口不谈。
And the sequence never changes: weigh it → trace it → attribute it → reduce it → then, and only then, claim it.
而顺序永远不变：先称重 → 再追溯 → 再归属 → 再削减 → 最后，才谈声明。
Primary sources for the standards, factors and benchmark bands used above. Standards revise — always confirm the current edition and scope before printing a claim.
上文所用标准、因子与基准区间的一手来源。标准会改版——印出声明前，请务必确认现行版本与适用范围。
Every plant figure on this page is an illustrative worked example built to show the method. Ranges for hotspot shares and lever effects are typical industry orders of magnitude, not a substitute for your own measured data.
本页所有工厂数字均为演示方法用的示例推导。热点占比与杠杆效果的区间为行业典型量级，不能替代你自己实测的数据。