When your precursor comes from more than one installation

By Ahmed Medhat · Published22 September 2026

Most complex goods are not made from one supplier's precursor. A steel mill buys pig iron from two sources because one of them was short that quarter. A cement plant takes clinker from a sister site and from the market. The moment that happens, the calculation act stops letting you pick a number and starts telling you which number to use.

Article 14 of Implementing Regulation (EU) 2025/2547 is short, and it decides more than its length suggests. It has three paragraphs, it is titled "Precursors produced during different reporting periods or in different installations", and each paragraph answers a different question.

Paragraph 1: the same installation, different periods

"Where an installation producing complex goods receives, from another installation, precursors under a given CN code produced during different reporting periods, the embedded emissions of the complex goods shall, for the part of the emissions embedded in the precursors under that CN code, be determined as the weighted average of emissions embedded in the precursors under that CN code produced during those different reporting periods" (Art. 14(1)).

One supplier, two vintages. You do not get to use the better year. You weight them.

Note what the weighting is over: the precursors under that CN code. Not the good you make from them, and not your production process as a whole. The averaging happens inside the precursor line.

Paragraph 2: several installations, and the default

"Where an installation producing complex goods receives precursors under a given CN code from multiple installations, the embedded emissions of the complex goods shall, for the part of the emissions embedded in the precursors under that CN code, by default be determined as the weighted average of emissions embedded in the precursors under that CN code received from the different installations" (Art. 14(2)).

Two words carry the weight here. Multiple installations, and by default.

Multiple installations is the trigger, and it is worth being precise about, because it is easy to get wrong in both directions. It is not multiple suppliers in the commercial sense, and it is not multiple purchase orders. It is multiple installations: multiple physical plants producing the good. Two deliveries from the same plant are one installation, and paragraph 2 does not apply to them at all. Two plants owned by the same group are two installations, and it does.

By default means this is what happens unless you do something about it. Which brings us to paragraph 3.

Paragraph 3: narrowing it, with evidence

"Where operators provide the verifier with sufficient evidence demonstrating that, out of the precursors under a given CN code received from multiple installations, the installation producing the complex goods used, for a given production process, only precursors from a single installation, or from a subset of installations, the embedded emissions of those precursors used in goods produced through that production process shall be determined, respectively, based on the embedded emissions of the precursors obtained from that single installation, or as the weighted average of emissions embedded in the precursors received from that subset of installations" (Art. 14(3)).

This is the escape from the average, and it is an evidenced one. Three conditions are stacked in that sentence and all three have to hold.

The evidence goes to the verifier, not to a file in your drawer. It has to demonstrate actual use, not purchase: that the production process consumed only that installation's material. And it is scoped to a given production process, so it can be true for one of your processes and false for another in the same year.

The reward is proportionate. Narrow to one installation and you use that installation's figure. Narrow to a subset and you weight across the subset. You never get to simply choose the lowest.

Why this is worth effort

If your cleanest supplier is also your smallest, the default in paragraph 2 buries them. A weighted average across all installations under that CN code drags your embedded emissions towards whoever supplied the most tonnes, which is usually not whoever supplied the cleanest tonnes.

Paragraph 3 is how a genuinely low-carbon supply arrangement survives the arithmetic. It costs you traceability: you have to be able to show which process ate which material. Plants that already run batch or heat-level records have most of what is needed. Plants that buy into a common stockyard usually do not, and that is a decision about the stockyard, not about paperwork.

What the report has to say about it

This is not an internal working. The choice is a reported element.

Annex IV point 1.1 item 25 requires, "where the production process of a complex good used a precursor under a given CN code obtained from multiple installations, the specific embedded (direct and, if applicable, indirect) emissions to be used for that precursor, and an indication whether they were determined by using the default method laid down in Article 14(2) or by calculating the embedded emissions of the precursor obtained from a specific installation or subset of installations in accordance with Article 14(3)".

So the report states two things: the figure, and which paragraph produced it. Item 24 does the same for the different-reporting-periods case under Article 14(1).

An indication that says 14(3) where the truth is 14(2) is a defect in a required element of a filed document, not a presentational slip. It tells a reader the figure was narrowed on evidence when it was not.

Three ways this goes wrong

The CN written two ways. The trigger in paragraph 2 is a precursor under a given CN code. If the same good is recorded once as 2523 10 00 and once as 25231000, a system comparing the raw text sees two different goods, each from one installation, and concludes that paragraph 2 does not apply. The report then states the single-installation case over a genuine multi-installation one. Compare CN codes without their spaces, everywhere the comparison decides something.

Counting rows instead of installations. The other direction. Two deliveries from one plant, recorded as two lines, are not multiple installations. A count of lines says they are, and the weighted-average method gets applied to a supply that never needed it. This one is easy to introduce and hard to notice, because the arithmetic still looks reasonable.

Dropping the blanks out of the average. A precursor line with no stated figure does not disappear from the weighting. Leave it out and the average rises over the rows that remain, and you publish a number that is not the one the calculation used. Where a supplier has stated nothing, the act sends you to the default value for that CN code and country of production rather than to a gap.

What to do with this

Two questions will tell you where you stand. For each precursor CN code you buy, how many distinct installations did it come from last year, counted by plant and not by invoice. And for the ones with more than one, can you show which production process consumed which material.

If the answer to the second is yes, Article 14(3) is available to you and worth taking. If it is no, paragraph 2 is your method, and the honest thing is to report it as such.

We track the CBAM acts daily and write up what changes. If you want to see how the pieces fit before you build a process around them, the DeCarbonPro free CBAM tools need no account.