Knowledge base · 01-embodied-carbon

01 - Embodied carbon, from first principles

Status. Draft v0.1 · First draft: 17-03-2026 · Pre-discussion.

Why this matters. Every conversation about GREMI, SIDK, CBAM, verifiers, and passports is downstream of one fact about the physical world: in 2024 the global steel industry emitted 2.18 tonnes of CO₂-equivalent per tonne of steel produced on average (Scope 1+2+3, per the World Steel Association’s 2025 indicators report ¹), and the world has decided that number has to be counted, attested, and trusted across organisational boundaries. If you do not have a working mental model of what is being counted and why it is hard to count, the rest of the platform reads as bureaucratic plumbing. It is not. The plumbing exists because the underlying measurement problem is genuinely difficult.

1. What “embodied carbon” actually means

When you buy a tonne of hot-rolled steel coil, you are buying two things: the steel itself, and an invisible quantity of greenhouse gas that was emitted into the atmosphere while the steel was being made. The second thing is the embodied carbon of that coil. It is the cumulative greenhouse-gas footprint of every physical and chemical step that produced it - from the iron ore in the ground, through the coke oven, through the blast furnace, through the converter, through casting, through rolling, through every kilowatt-hour of electricity the mill drew off the grid along the way.

Embodied carbon is contrasted with operational carbon - the emissions a product generates while it is being used. A car’s tailpipe emissions over its lifetime are operational; the emissions from making the steel that became the chassis are embodied. For industrial inputs like steel, cement, aluminium, and chemicals, embodied carbon is the entire story - these materials are not “operated”, they are consumed.

The unit is kilograms (or tonnes) of CO₂-equivalent per unit of product. “Equivalent” because the atmosphere does not care only about CO₂ - methane, nitrous oxide, and a few industrial gases all warm the planet too, at different intensities. We convert them all to a common CO₂ basis using Global Warming Potential factors (more on this below). So a steel coil might be described as “1,950 kgCO₂e per tonne, cradle-to-gate” - meaning every greenhouse gas emitted from raw-material extraction up to the moment the coil leaves the gate of the mill, expressed as the equivalent mass of CO₂.

2. Why steel is one of the hardest cases

Roughly 7% of energy-system CO₂ emissions come from making steel ² - about 2.6 Gt of CO₂ per year, more than all road freight combined. That alone makes it strategically important. But the technical difficulty of measuring steel’s embodied carbon is what makes it interesting as a platform problem.

Three properties of the steel industry make the measurement genuinely hard:

The carbon is in the recipe, not just the energy. A coal-fired power plant emits CO₂ because it burns coal for heat. You can measure that with a stack analyser on the chimney and be roughly done. A blast furnace burns coal and uses coal’s carbon as a chemical reductant to strip oxygen from iron ore. The carbon ends up in the off-gas, in the slag, in the molten iron, in the dust, and in fugitive emissions across the site. You cannot point one sensor at one chimney and call it measured. You have to account for carbon across multiple streams using mass-balance logic - what came in as carbon, what left as carbon, what stayed behind.

Multiple production routes give wildly different numbers. The same finished product - a coil of hot-rolled steel - can be made via the integrated route (BF-BOF: blast furnace plus basic-oxygen furnace, fed by coke), via the scrap route (EAF: electric arc furnace melting recycled scrap), or via direct-reduced iron (DRI: reducing iron ore with natural gas or, increasingly, hydrogen). BF-BOF runs around 2.0–2.3 tCO₂e/t of crude steel. Scrap-EAF runs around 0.3–0.7 tCO₂e/t depending on the grid mix powering the furnace ³ ⁴. DRI sits in between when fed by natural gas, and can in principle approach zero when fed by green hydrogen on a clean grid. Three steels, same spec sheet, up to a ten-fold spread in carbon. The number is a property of how it was made, not of what it is.

The grid factor swings the answer. A modern EAF mill’s emissions are dominated by the electricity it draws. Whether you count that electricity at the location-based average for the local grid, or at the market-based factor reflecting a green-power contract the mill has signed, can change the headline number by more than an order of magnitude for the same physical kWh. Both numbers are legitimate under the GHG Protocol’s Scope 2 Guidance, which since 2015 has required both to be reported ⁵. We will return to this in 02 - Scope 1, 2, 3, because it is one of the most consequential and most-argued-about details in carbon accounting.

The combination - multi-stream carbon accounting, route-dependence, and grid-factor sensitivity - is what forces a serious steel carbon-accounting system to look like an industrial data platform rather than a spreadsheet.

3. The basic equation, and why it is already a simplification

The textbook equation for an emission is:

E = AD × EF

E is emissions (kgCO₂e). AD is activity data - how much of something you did, like tonnes of coal burnt or kWh of electricity consumed. EF is an emission factor - how much CO₂e is associated with a unit of that activity, like kgCO₂e per tonne of coal.

This equation is correct, useful, and a simplification of what really happens, in three ways:

The factor is a lookup, not a measurement. When you say “natural gas emits 56.1 kg CO₂ per gigajoule” ⁶, you are pulling a number from a published table - IPCC, DEFRA, the national inventory, or a supplier-specific factor. The accuracy of your answer is bounded by the accuracy of that factor. The Carbon Engine cares deeply about factor provenance: which factor, which version, for which geography, valid for which dates, and at which tier of confidence. A whole governance layer exists to track this.

For some sources, the equation is genuinely wrong and you need mass balance. A blast furnace’s carbon flow does not factor into AD × EF cleanly. You write it as E = Σ (mass_in × C_in) − Σ (mass_out × C_out) − ΔC_stock − C_captured, multiplied by 44/12 to convert from carbon to CO₂. Each m × C is itself a sub-problem with its own measurement and uncertainty. This is Method 3 in the Carbon Engine. We explain it in 06 - The Carbon Engine; for now, just hold that the textbook equation is one of several tools the engine reaches for, not the universal answer.

For some sources, you do not have AD directly and have to back into it. Supplier emissions, transport emissions, and small intermittent sources are sometimes computed from spend (currency) rather than physical quantity, or from category-average defaults, or from supplier-declared product carbon footprints. Each of these is a different tier of evidence - and the platform tracks which tier any number came from, so a verifier reading a claim can see how much of it was directly measured versus inferred from a default. The five-tier evidence model is one of the load-bearing concepts in the architecture.

4. CO₂-equivalent and the GWP choice

Not every greenhouse gas is CO₂. Under the IPCC AR6 GWP-100 values ⁷, methane traps 27.0 times more heat than CO₂ over a 100-year horizon if it is of biogenic origin, 29.8 times if fossil-derived. Nitrous oxide (N₂O) traps 273 times. Some industrial gases (SF₆, certain HFCs) run into the thousands or tens of thousands. To produce a single comparable number, every gas is multiplied by its Global Warming Potential factor and the result is summed:

GHG_total (kgCO₂e) = Σ gas_i (kg) × GWP_i

Three things to know about GWP that constantly trip people up:

The GWP set has versions. IPCC AR5 GWP-100. IPCC AR6 GWP-100. AR4 is still cited in some legacy regimes. The values are slightly different between versions because the underlying climate science is refined. Two carbon numbers that differ only because one used AR5 and one used AR6 will look like a real discrepancy and are not. The Carbon Engine carries the GWP reference on every CalculationResult so a verifier can see which set was used.

Time horizon matters. GWP-100 (a 100-year horizon) is the standard. GWP-20 (a 20-year horizon) ranks short-lived gases like methane much higher and is used in some sustainability-policy contexts. Mixing them silently is a category error.

For most steel-industry purposes, CO₂ alone dominates and the GWP machinery feels like overkill. Then someone reports a fugitive methane leak at a coke oven, or an SF₆ leak at an electrical substation, and the GWP machinery is the reason the number does not implode.

5. Why this is a trust problem, not just a measurement problem

Suppose a steel mill computes its embodied carbon carefully, honestly, and competently. What standing does the number have outside the mill?

If the mill simply publishes the number, the answer is: none. The buyer cannot tell whether the mill measured carefully or pulled a flattering number from a spreadsheet. The regulator cannot tell whether the methodology was applied correctly. The mill’s competitor next door, who is being undercut on a “low-carbon steel” contract, cannot tell whether they are losing on physics or on accounting.

This is why the carbon-accounting world is built around third-party verification. An accredited verifier - a firm accredited under ISO 14065 ⁸ by a national accreditation body - reads the mill’s data, walks the audit chain, raises challenges, and ultimately issues a signed VerificationStatement with one of a small set of outcomes. The verified number has external standing; the unverified number does not.

A second layer of trust is regulatory mandate. Under CBAM, the EU does not allow an importer to self-attest the embodied carbon of imported steel - declarations are tied to verifier-attested data, with specific accreditation requirements ⁹. (Full detail in 03a - CBAM, in depth.) The regulatory regime is what turns “verification” from a marketing exercise into a compliance gate.

A third layer is cryptographic verifiability. Even a verified number, written into a PDF, is only as trustworthy as the chain of custody between the verifier’s audit and the buyer’s hand. A platform that makes the verification chain re-checkable - a buyer can scan a QR code on a shipment, fetch the claim, fetch the verifier’s attestation, fetch the platform’s signature over the response, and verify all three signatures independently on their phone - is a platform that has removed itself from the trust path. The buyer does not need to trust the platform; they verify the cryptographic chain directly. This is what the three-signature trust chain is for, and it is the most architecturally distinctive thing about SIDK. We come to it in 05 - SIDK.

The progression - measure carefully → verify independently → make the verification re-checkable - is the through-line of the entire platform. Everything else is in service of that progression.

6. The minimum mental model

If you remember nothing else from this doc:

Embodied carbon is the cumulative greenhouse-gas footprint of a product up to a defined boundary. For industrial inputs, this is the whole story; there is no operational phase. The boundary vocabulary - cradle-to-gate, cradle-to-grave, EN 15804 modules - is covered in 02a - Boundaries.
Steel is hard because carbon is in the recipe (not just the fuel), because routes produce 10× different numbers for the same spec, and because the grid-factor choice swings the answer.
The textbook equation E = AD × EF is one tool of several. Mass balance, supplier PCFs, defaults, and direct measurement all coexist; the engine picks among them per source.
GWP-converts non-CO₂ gases to a single comparable number. The version of the GWP set is part of the answer.
A number’s standing comes from verification and cryptographic re-checkability, not from the number itself. This is the entire reason an architecture this elaborate has to exist.

The next doc, 02 - Scope 1, 2, 3, is where these ideas get the structure that the GHG Protocol imposes and that every regulation downstream of it inherits.

01 - Embodied carbon, from first principles

1. What “embodied carbon” actually means

2. Why steel is one of the hardest cases

3. The basic equation, and why it is already a simplification

4. CO₂-equivalent and the GWP choice

5. Why this is a trust problem, not just a measurement problem

6. The minimum mental model

References & further reading

Primary sources (cited inline)

Further reading (not cited above, but worth your time)

01 - Embodied carbon, from first principles

1. What “embodied carbon” actually means

2. Why steel is one of the hardest cases

3. The basic equation, and why it is already a simplification

4. CO₂-equivalent and the GWP choice

5. Why this is a trust problem, not just a measurement problem

6. The minimum mental model

References & further reading

Primary sources (cited inline)

Further reading (not cited above, but worth your time)

Footnotes