Docs / How a factor is decided

What do we do when only one transcription speaks?

Part of How a factor is decided. The method's own publisher is silent about a flow, and one transcription states a number for it — this page is whether that number becomes ours.

EF 3.1's flow list has no silvicultural soil, no industrial soil, no groundwater, and no Copper, Ion. ecoinvent's has all of them, and the ecoinvent Centre's implementation of EF 3.1 characterises them: metaldehyde in forestry soil, chlorobenzene in groundwater, copper's ion in surface water. So this list holds thousands of flows for which exactly one deciding implementation speaks, and it is not the method's own.

That is a different situation from the JRC speaking alone. The reference's own number for a flow is the method; adopting it is not a decision, and 296,719 of EF's consensus factors are sole for that reason. A transcription's number for a flow the method's publisher never characterised is a claim about what the method would say, and adopting it is.

Figures are from the build of 1 September 2026, run 20260901T0451332605450000.

The number this list already carries

Metaldehyde is the poison in slug pellets. EF characterises it in agricultural soil at 5.6369 CTUe/kg for freshwater ecotoxicity, in non-agricultural soil at 5.637, in unspecified soil at 5.6369. ecoinvent adds a row for forestry soil, which resolves to our silvicultural context, and states 5.637 there.

Nobody is proposing new science. ecoinvent has carried EF's own number onto a compartment EF's list cannot express, and this list can see that, because it already publishes the same substance at that number in the category next door. Where that is so — the same substance, the same category, another context, within the same 2 % that separates a rounding from a disagreement — the factor is published as restated. The amount is always the transcription's own stated row; nothing is inherited or interpolated, and the derivation names the fact that made it publishable.

That is the whole of the 3,818 restated factors: roughly 2,750 in silvicultural soil, 950 in the unconfined aquifer and 120 in industrial soil. The census of the blanks (next page) says which neighbour each one matches, and the answer is consistent enough to be a rule: ecoinvent gave forestry and industrial soil EF's non-agricultural number in 2,747 of 2,765 silvicultural rows and 123 of 123 industrial, and groundwater EF's surface water number in 948 of 948.

The population a curator accepted

Sixty-two named pesticides reach this list under their own names, because it declines the catch-all mapping ecoinvent's correspondence table gives them — Insecticides, unspecified and its siblings (#405). ecoinvent's implementation characterises them as the bucket, so each proposes the bucket's numbers on a flow the JRC never characterised. A curator accepted that as a population: the fact that makes the numbers acceptable is a fact about these substances, written down once in data/lcia-factor-adoptions.json rather than ruled sixty-two times. Twenty-three minerals — gypsum, sodium chloride, borax — are the file's second population, where ecoinvent's number is the JRC's element factors weighted by the mineral's formula and the arithmetic was checked (#752). Those 899 factors carry derivation: adopted, and the 435 signed on 2026-09-01 -- the ions, the two conjugates and the land classes below -- carry the same word.

An entry is deliberately not pinned to a number where its reason is a relationship: a revised bucket factor is still the bucket's factor reaching a named pesticide. The minerals' entries are pinned, because their reason is the arithmetic.

The number another substance carries

Copper, Ion is a substance this list has only because ecoinvent shipped it. The JRC characterises copper, not its ion; ecoinvent gives the ion copper's own numbers, 46.47592828149169 CTUe/kg in surface water; and that number stands published on copper. Until 2026-09-01 the ion's factor was published as restated on that strength alone: some flow of the category carried the very same digits (#704).

The evidence that rule read was the amount, and the amount is the same shape whether the donor is copper for its ion or lindane for its manufacturing contaminants. EF characterises α‑, β‑ and δ‑hexachlorocyclohexane for freshwater ecotoxicity and holds them 21–34× away from lindane; for non-cancer human toxicity it characterises only lindane. ecoinvent filled the gap with lindane's number, 0.00019937 CTUh/kg from unspecified air, and this list published it on all three isomers under their own names — because the digits matched (#648). Nobody had decided that. On the build of 1 September, 311 factors over 75 substances reached their flow by matching another substance's number: 120 an ion taking its element's, 108 a stereo-relative taking its parent's, 14 a named substance taking a catch-all's, 69 with no recorded relationship at all.

The rule is gone (#760). No number crosses from one substance to another without a signature, however obvious the pair, and the signatures are data/lcia-factor-adoptions.json: one entry per recipient, naming the donor, the relationship in one word — ion-of, salt-of, conjugate-of, catch-all, land-family, formula-weighted, substitute — the categories covered, and a verdict, adopt or decline, with a comment. Copper's ion is an ion-of entry, signed; so are nine other ions, the acrylate anion and flupyrsulfuron-methyl's sodium salt with the mass correction stated, and 53 land classes taking their family's number. The three hexachlorocyclohexanes, tetrachlorvinphos, lambda-cyhalothrin and (R)-mecoprop are substitute entries with the verdict decline and #648's table as the reasoning, and 2,4-D ester, dimethyl hexynediol and fenpropimorph are declined catch-all entries; their rows stay on the proposed-factor page answered rather than asking. tools/count_identity_crossings.py lists every candidate pair with the relationship the build records, so an entry is written from evidence rather than from memory — and an entry never overwrites a value: it answers a row the reference is silent about, and nothing else.

The number nobody stated for the recipient

Everything above answers a row a transcription stated. Stepwise 2006 has one implementation, and its export names no ion of any metal but chromium. Zinc emitted to water is characterised at 133.39 kg C2H3Cl-eq per kg for non-cancer human toxicity, published on Zinc; Zinc(2+), the dissolved ion, has a flow in the same compartment and nobody states anything for it, so it is never asked about. Every ecoinvent zinc emission lands on the ion — the air and soil rows too, because ecoinvent's own correspondence table sends its metal to EF's Zinc II — so under this list's Stepwise implementation a kilogram of ecoinvent zinc scored nothing (#922).

The signature for that case says something slightly different: not "publish the transcription's number" but "this substance takes its donor's published number, context for context". An entry in the Stepwise adoptions file, data/lcia-factor-adoptions-stepwise-2006.json, carries number_from: donor, and the pass that reads it copies zinc's water number onto the ion's water flow, zinc's air number onto its air flow, and so on, as adopted, each naming the zinc flow it took the number from. It runs before the convention, so the ion's river, lake and forestry-soil flows then take the number from the ion's own rows the way zinc's do. On the build of 2 September 2026 that is 299 adopted factors over 23 ions of 16 elements, and the convention carries a further 338 from them.

What it refuses is the point. A recipient the method's own publisher characterised anywhere is refused whole: Stepwise names Chromium III and gives it no ecotoxicity number while giving the metal one, and that silence is the publisher's decision, so no entry is written for the chromium states and the pass would refuse one. It never overwrites a published triple, never copies a number that was itself carried or adopted, and refuses a pair whose units differ. Stepwise's model does not tell a metal's oxidation states apart, so the 23 entries include EF 3.1's own speciated flows — Antimony(3+) and Antimony(5+), Arsenic(3+) and Arsenic(5+), Copper(2+), Iron(2+) and Iron(3+) — which no inventory merged today emits.

What is left is a question

A transcription's number that none of the above vouches for is a proposed-factor: nothing is published, and the row asks. On this build the queue is 64 questions over 158 factors, 48 of them answered on the record -- the declines above and the earlier rulings -- and 16 open: ten rocks EF's flow list never contained, each with one ecoinvent number for mineral resource use (#651), kresoxim-methyl in silvicultural soil, where ecoinvent states 53,540 against a soil number this list publishes as 134.73, and fenoxycarb's two rows, which are #645's twin substance and wait for the merge rather than for a signature.

For a number that is genuinely one publisher's own, with no donor and no neighbour, the standard is an attributable source: sulphuryl difluoride's global warming potential was published because IPCC AR6 states it, and a rock's depletion factor will be published when somebody can say where it came from — or declined on the record.

Where it is written down

• lcia_characterization_factors.derivation — sole, restated, adopted.
• /queue/proposed-factor — the open questions and the ruled ones.
• data/lcia-factor-adoptions.json — the two populations, and the one substance it names as refused (fenpropimorph, whose numbers are mostly the bucket's and partly ecoinvent's own).
• data/lcia-factor-adoptions-stepwise-2006.json — the 23 Stepwise ions taking their element's published number where nobody stated one.
• data/lcia-factor-rulings.json — proposed rows a curator answered one at a time.