Docs / Concepts

Characterisation factors

Kresoxim-methyl is a fungicide. Sprayed on agricultural soil, how much does it contribute to freshwater ecotoxicity? EF 3.1 answers that, and this list publishes three answers to it, because EF 3.1 is a method and a method has to be implemented against a flow list before it is a number you can use:

397× apart, from one method. Neither team made an arithmetic mistake. EF's own files carry kresoxim-methyl twice in every context — the same name and the same CAS number, 143390-89-0, under two EC numbers — and the two rows carry different factors. This pipeline merged them into one flow and kept one number of the two; the ecoinvent Centre, reading the same two rows, kept the other.

So this is not two teams disagreeing about toxicology. It is one question about EF's own duplication, answered twice, differently — and a curator answered it here, in data/lcia-factor-rulings.json, with the reasoning written down. Until they did, this list published nothing for it, which is the default: where two implementations disagree and no ruling exists, the honest artifact is a question.

That is what this layer is for. It does not compute impacts, it does not average, and it does not pick a winner where two competent teams disagree without saying who picked and why. It puts every number on one flow list, says who said it, and makes the disagreements findable.

Four implementations, one method

implemented_by is the field the whole layer turns on. There is no correct implementation of a method:

• European Commission — JRC — 318,774 factors over 88,907 flows. EF 3.1 is the base list, so these arrive already on consensus flows: an identity join, and one hop where the build has since merged the flow they were read off into another (below). 51,944 of them are a stated zero, which is a claim (#329): "assessed, and zero" is not "never assessed".
• ecoinvent Centre — 26,479 factors over 7,709 flows, read from the workbook fetch-lcia ecoinvent-3.12 downloads. These take two hops: the workbook names an ecoinvent flow, the merge says which consensus flow that became, and only then is there a factor about a flow of ours.
• GreenDelta — 25,266 factors over 1,509 flows, read from the openLCA method package fetch-lcia bafu-2026-v1 downloads: EF 3.1 Method (adapted) from openLCA LCIA methods 2.8.0, as BAFU's openLCA distribution ships it. Their flows are found two ways. Most take the same two hops as ecoinvent's, through whichever release has the flow: the package identifies flows by the ecoinvent UUID it was built against, declaring compatibility with ecoinvent 3.6 to 3.11, so 1,144 are reached through ecoinvent 3.12, 50 through 3.8, and 29 are EF 3.1 flows already.

The other 381 are reached by what the substance is, because a method package is not a flow list and 558 of their rows name a substance under an identifier nothing here has. A curated decision first, then a registry number — theirs are written zero-padded, 000056-23-5 for carbon tetrachloride — and then an exact name, all three inside the compartment the factor was stated in. Never a substance alone: a freshwater number reaching a flow in air would be a wrong factor rather than a missing one, so the compartments are 21 curated rows saying what each of theirs means. Where two flows of ours carry one registry number — Water and Water vapour both carry 7732-18-5 — the name settles which, and where nothing settles it the row is reported and matched to neither.

The curated decisions are 91 rows, in data/lcia-substance-decisions.json, and 25 of them are about ions. Their package writes Copper ion where this list publishes Copper, Ion, and Cadmium II where it publishes Cadmium(2+), with no registry number to fall back on. Nothing about the strings settles that, but their own file does: each of those names appears in it once per compartment, nine to thirteen times, and most of those copies carry an ecoinvent UUID that already resolves — to Copper, Ion seven times out of twelve, and never to a second substance of ours. A row carries that reading into the compartments where their identifier does not resolve, and records the count it rests on. Seven of the 25 are declined instead: they are metals emitted to a lake, where this list publishes only the element-named flows BAFU brought, and the finding that reports each one carries the reasoning.

The other 66 are a land class or a kind of water. Three of their compartments — Resource / land, Resource / in water and Resource / unspecified — name a place where this list names an action, so nothing about Occupation, agriculture reaches a flow: the word Occupation is our context and never our label, and agriculture is the land class. What says which class is BAFU's own list. The package is distributed inside BAFU's database and names its flows as BAFU names them, and this project has already read every one of those names, once, into data/land-flow-classes.json and data/water-flow-materials.json. Their file agrees where it can — 76 of their Resource / land flows carry an ecoinvent UUID that resolves, 23 to land occupation and 53 to land transformation and none anywhere else — and so do the numbers: 640 of the 1,528 factors these rows carry are bit-identical to what the JRC states for the same flow, category and country, and none of the 1,528 disagrees (#785).

195 of their rows write the country into the flow name, and are set aside rather than matched. Water, well, CH is stated beside Water, well, and the site-generic row already carries one water-use factor for each of 209 countries, Switzerland's among them; a factor's place is its geography, so the coded row is the same number written a second time. Before this rule 47 Water, XX rows reached Water in Environmental → Water → River by their registry number and put 47 countries' numbers on one flow under no country, which the build reported as a collision. Each row set aside is reported with the reason, the way a declined decision is, and counted apart from the nine curated declines. Where a set has no site-generic member the GLO row stands for it: Water, unspecified natural origin, GLO reaches Water in Resource → Water → Unknown by a curated decision, which is where BAFU's 78 rows of those names already go. Three rows have neither — Water, well, RER under Resource / in ground, where their Water, well is not — and nothing of them is published (#785).

913 of their factors reach nothing on the build of 2026-08-29: the 195 rows above carry 376 of them, and the rest are the ties and unit crossings #784 reports.

It is here to be compared with, and it decides nothing. The consensus implementation is derived from the JRC's and ecoinvent's alone (#766): a number of GreenDelta's is never evidence for a number of ours, never a candidate in a curator's queue, and publishing it changes no factor this list states. What it does is put a third reading of EF 3.1 in the difference report, where a disagreement is a question somebody can look at. - consensus-flow-list — 324,629 factors on the 2026-08-24 build, and every one of them is one of the other two's. 21,770 are agreed — both implementations stated the same number, so there is nothing to decide — 297,544 are sole, where only one of them spoke at all, and 32 are ruled: a curator looked at a disagreement and said which number to publish.

Two more take a number only ecoinvent states, and both say so on the factor. 4,402 are restated: the number is one this list already publishes in the same impact category, so publishing it decides nothing new. Mostly that is a compartment EF's flow list does not have — no silvicultural soil, no industrial soil, no groundwater — where ecoinvent characterises the missing compartment with the value it states for the nearest one EF does have, a value this list already publishes for the same substance, allowing for the last few digits a unit conversion can move. The adopted factors are the ones a person signed for, in data/lcia-factor-adoptions.json: a substance taking another substance's number, with the relationship named on the entry. EF characterises copper and has no Copper, Ion flow; ecoinvent ships the ion and gives it copper's numbers; the entry says ion-of, and the ion's factors publish as adopted. Ten ions, two conjugate pairs and 53 land classes taking their family's number were signed on 2026-09-01, beside the file's first two populations: 62 pesticides whose numbers are a catch-all's, accepted as a population because ecoinvent's own correspondence routes them into Insecticides, unspecified and its implementation characterises them there, while this list publishes them under their own names; and 23 minerals — sodium chloride, gypsum, potassium chloride, borax — that EF's flow list carries and the JRC never priced, whose ecoinvent number is the JRC's own element factors weighted by the mineral's formula, checked (#752).

Where the implementations disagree, or where ecoinvent states a number of its own that nothing else in the list carries anywhere, this publishes nothing and asks. On the 2026-08-24 build that is 34 open questions over 128 factors where the two implementations contradict each other, and 37 over 39 where only ecoinvent speaks — the second down from the 5,326 factors that queue would hold with no rules at all.

Each implementation has its own 25 impact categories, so EF 3.1 accounts for 100 of the published set. All of them are long term: excluding a source list's long-term compartments is that list's convention, not this one's, and a list that published both a long and a short set would be publishing a choice it has no basis to make.

Two methods, and nothing that assumes either

EF 3.1 is a method this list characterises against, and it is not the method this list is about. Everything above is written per method: a method states its own categories, its own reference unit for each, its own implementations, which of them decide anything, and where each of their factors is read from — and it states all of that in a file of its own. data/lcia-impact-categories.json is EF 3.1's. Adding a second method is adding a second file, and there is now a second file.

Stepwise 2006 is it: data/stepwise-2006-impact-categories.json, 19 categories under two implementations rather than 25 under four. Two, because one publisher renders it — 2.-0 LCA consultants, whose SimaPro export is the whole of it — so there is nothing to cross-check and every consensus factor of it is sole. That is the shape a method with one publisher has, and it needed no case of its own: the method's own publisher decides, ours does not, and the derivation that reconciles four implementations of EF reconciles one of Stepwise.

So the published set is 138 categories: EF 3.1's 25 times four, and Stepwise 2006's 19 times two.

What a Stepwise number means is written in its method file, with whose words those are. An EF 3.1 category's indicator is transcribed from a publisher: acidification's is the JRC's Accumulated Exceedance (AE). Stepwise's export gives a category a name and a unit and nothing else, so there is nothing to transcribe. But Stepwise takes most of its categories from EDIP 2003 and IMPACT 2002+, whose documentation names each indicator, and documents the rest itself. So Stepwise's acidification publishes Area of unprotected ecosystem (UES) — EDIP 2003's words for what m2 UES counts — and the category's meta.indicator_source names EDIP 2003 and quotes it. That key appears only where the words are not a publisher's own; where nobody's words exist at all, indicator is null.

Two methods share the flow list and nothing else. That is not a modesty about the code, it is a fact about the science: Stepwise 2006 also has a category called Acidification, counted in m2 UES from a different model, and it is not EF's. So no category name, no category slug, no reference unit and no curated ruling crosses from one method to the other. A ruling written about kresoxim-methyl under EF's freshwater ecotoxicity says which method it is about, and is read for that method alone; a question in a factor queue is named by its method as well as its substance and its category; every row of the difference report and the coverage summary says which method it is a row of; and every number a run counts about a method — how many factors an implementation stated, how many the curators have ruled on — is counted under that method's name.

Whether an implementation decides is on its row, not in a list in the code. GreenDelta's says decides: false and is published, transcribed and compared like the rest; the JRC's and ecoinvent's say decides: true and are what the consensus implementation is derived from. The method's own publisher always decides, and ours never does, because ours is the thing being decided.

The one thing that is shared is this list's own name: consensus-flow-list is the implementation this list publishes for every method it characterises, because it is the same people saying it. Which method a row of ours belongs to is said by the category it is under, not by the name on it.

A factor about a flow this build merged away

Two files come out of one build, and they have to agree about which flows exist. harmonised-flows-simple.json.gz publishes only live flows and files every deprecated identifier under redirects, saying what replaced it and why. lcia-factors.json.gz used to go on publishing factors against those identifiers — 1,674 of them over 166 flows on the 2026-08-25 build — so a consumer that loaded both saw numbers about flows the flow list does not have (#778).

The factors now follow the build's own redirects, and follow exactly the ones it tells a consumer to follow:

• identity-merge is the same flow reached twice, so a factor about one copy is a factor about the flow. 151 identifiers, and their numbers move onto the survivor.
• context-collapse is not: two source contexts our vocabulary cannot tell apart, whose factors may legitimately differ (#49). 15 identifiers, 80 factors, every one of them the same number the survivor already publishes — they are reported as redirect-refused and not carried across.

Where a redirected factor meets one the surviving flow states itself, the flow's own number stands. That pair was settled when the build collapsed the two flows: the more precise number was kept and the other written onto it as a superseded value (#375), which is where kresoxim-methyl's 53,540 against 134.73 is on the record. Asking again here would publish neither and withdraw 132 factors the build has already decided.

Where the surviving flow states nothing for that category and place, the number is carried onto it and the factor names the retired identifier as its source_flow_uuid. 72 factors, over three substances, and each one is a redirect-followed finding — because this is the one case where following a redirect adds a factor rather than restating one, and deduplication deliberately leaves it open: it settles a number two rows both publish and unions nothing.

What makes two numbers comparable

Almost all of the work is here, and none of it is arithmetic.

The category. The JRC calls it Ecotoxicity, freshwater; ecoinvent calls it ecotoxicity: freshwater. One category, two names, and nothing in either file says so — the method's own file, data/lcia-impact-categories.json for EF 3.1, is what does: one row per category with every implementation's words and ours. Comparing on either publisher's own name compares nothing with nothing: every factor comes out sole, no factor is ever agreed, and the counts look entirely plausible. Comparing across two methods on the strength of a shared name would be the same mistake one level up, which is why nothing does.

The flow. A factor is about a flow, so a factor can only be as good as the merge that placed it. 9,848 of ecoinvent 3.12's source flows reach a consensus flow, and of the 7,960 its workbook characterises exactly two do not — a question for the merge, not for this layer, and a finding rather than a silent loss.

The place. EF states a geography on 42,871 of its factors — 223 codes, mostly ISO 3166-1 alpha-2 — and Land use gives one flow 213 of them, from -522.81 in ES-CA to -227.0 in YE. Two places are two factors, which is why the identity of a factor is the triple (impact category, flow, geography) and not a pair. ecoinvent's workbook states no place at all, so on the raw count every located EF factor looks "dropped" by its implementation; that is two file formats, not a gap in anybody's science.

The unit. A factor stated per kilogram, on a flow published per megajoule, is not a number you can use. Those are converted by the flow's own property — a uranium factor of 560,000 per kg over 560,000 MJ/kg is 1.0 per MJ, which is the JRC's own number for the megajoule flow, so the conversion is checkable rather than asserted.

The conversion comes from the row itself where the row has one, and otherwise from the build: every multiplier the merge recorded is indexed by the flow and the pair of units it joins, so a factor reached by a name or by a curated decision can use the number a merged list already established for that same flow. GreenDelta's water use factors are stated per kilogram and this list publishes water per cubic metre; BAFU's own kilogram water rows carry 0.001 m³ per kg onto the very same flows, and 213 factors are published on that (#804). A flow whose rows state two different multipliers for one pair of units — natural gas, at 34.5 and 36.0 MJ per cubic metre — is not converted, because the evidence is that there is more than one answer. Where no conversion exists at all the factor is not published and the crossing is a finding.

The sign. Taking a cubic metre of water out of a French river costs +6.98 under EF 3.1's water use; putting a cubic metre back is −6.98. A process that withdraws and returns the same water scores about nothing, which is the point of the pair — and which half a factor is, is not written on the factor. It is the direction of the flow: water taken from the environment is a resource, water put back is an emission.

GreenDelta's package states both halves positive. Their 1,438 withdrawals are right; all 416 of their factors on water emitted to water are the returning half with the minus sign missing, so as shipped, giving the water back would add to a water-use score. The magnitudes are the JRC's exactly — on the 1,046 that can be compared, every one agrees to the digit — so this list reads the sign off the direction their own package states, rather than publishing a number that says the opposite of what it means (#807).

Water use is the only EF category counted as a pair like this, and the rule is about water returned to water. Water evaporated to air is a real loss from the catchment and counts positively — the ecoinvent Centre states +42.95 for water vapour in every air compartment. Land transformation looks like a pair and is not: from forest and to forest are two flows rather than two directions of one, and GreenDelta's 5,733 negative land-use factors are right as they stand.

Where they disagree, nothing is published until somebody rules

Two implementations, one triple, two numbers. There is no faithful third number, so this list publishes none and asks. That is the contested-factor queue: on the 2026-08-24 build, 38 questions over 152 factors, of which 34 are still open, each carrying what a ruling needs — where each flow came from, what this pipeline did to it, and every number anybody states about it, including the ones a merge already declined.

The mirror case is proposed-factor, the factors ecoinvent states for flows the JRC does not characterise. Most never reach the queue: where the number is one this list already publishes for the same substance in a neighbouring compartment it is published as restated, because nobody is proposing anything, and where it is another substance's number — the way an ion carries its element's — a person's signed entry in data/lcia-factor-adoptions.json publishes it as adopted or declines it on the record. What reaches the queue is a number of ecoinvent's own: 40 questions over 48 factors on the 2026-08-24 build, of which 37 are still open and 33 of those are rocks and minerals ecoinvent prices for Resource use, minerals and metals and EF never did — granite, gypsum, talc. Adopting one means publishing a factor the method's own publisher never stated, about science only ecoinvent has spoken to. That is a decision, and it is asked rather than taken.

Twenty-three of those minerals were then approved (#752), and the reasoning is worth having in one place because it is checkable. The JRC prices the elements in the crust — chlorine at 2.71e-5, sulfur at 1.93e-4, boron at 4.27e-3 kg Sb-eq per kg — and ecoinvent's implementation report (v3.12, Table 20) prices a mineral as those factors weighted by mass share. Sodium chloride is 60.7% chlorine: 0.607 × 2.71e-5 + 0.393 × 5.5e-8 = 1.646e-5, which is ecoinvent's number to four digits. Gypsum is 18.6% sulfur and calcium carries nothing: 3.59e-5, again ecoinvent's number. Where the formula reproduces the number the mineral is those elements in the crust and the JRC has already priced it, so the number is adopted and published as adopted; where it does not — feldspar, olivine, and eight rocks with no formula to check — the question stays open. What that population records is different from the pesticides': the number it was made about, so that a revised number from ecoinvent is a new question rather than an automatic yes. The mechanism is the same one a ruling's ruled_about uses.

A ruling is consulted before the twin is, and that order is what lets a curator overrule a restatement. The twin once had a second half that read a matching number on any flow of the category as the same number — retired on 2026-09-01 in favour of the signed entries above (#760), because the same digits carried lindane's numbers onto its manufacturing contaminants — and once it read the same quantity: EF 3.1 counts water consumption as withdrawal less return, +42.95 on every water taken and −42.95 on every water given back, and the ecoinvent Centre counts it as what evaporates, 42.95 on Water to air and nothing on the rest. Every withdrawal of the category carries 42.95, so the air factor was restated off them — and scoring 500 ecoinvent datasets both ways found water use doubled on nearly all of them, the same cubic metres counted at each end of the process. The decline on water vapour (#749) is a ruling about an accounting convention rather than a number, and it is filed on the proposed queue because that is the question the row would have asked had the twin not answered it first.

None of the three queues blocks anything. A factor nobody has ruled on is simply not in the consensus implementation, and both publishers' numbers are published regardless.

Carbon dioxide, and the one rule underneath it

Three carbon dioxides are published, and a reader meeting them for the first time usually reads them as three kinds of carbon:

That reading breaks immediately. Land-use-change carbon is biological in origin — it was in trees and in soil organic matter — so how can biogenic carbon be neutral and this be charged the same as coal?

Because "biogenic" here is not a claim about where the carbon atom came from. The JRC's own sentence has the answer in its plural: "the biogenic carbon uptakes and emissions were considered neutral". It is a statement about a matched pair. One rule, applied three times:

A carbon flow is neutral if and only if its counterpart falls inside the study period.

A crop grows and is burnt: both halves are in the study, so they cancel and both are 0. A forest is cleared: that carbon was fixed decades ago and the cleared land will not take it back, so there is nothing to cancel against and the emission is charged in full. Fossil carbon is the same case with a longer gap. EF says as much on the flow itself — land-use-change CO₂ is "treated as fossil CO₂" — and so does ecoinvent, in almost the same words: it "came from the atmosphere to the stock much earlier than the scope of any LCA, like fossil carbon".

So the names describe three residence times, not three chemistries, and the one that trips people is (land use change), which names an event where the other two name an origin.

A fourth substance follows from the same rule and is not EF's: Carbon Dioxide, To Soil Or Biomass Stock, carbon entering a long-lived land pool, credited −1 as derivation: ruled on the ecoinvent Centre's number. EF's carbon model has no row for it — a removal is only expressible there as a resource from air — so it is adopted deliberately rather than carried in, and the ruling says whose number it is and what it rests on. The evidence is in plans/carbon-accounting.md.

One air factor covers every air sub-compartment, where the gas is well mixed

Sulphuryl difluoride is the gas a house is tented with for termites. At the end of a fumigation the whole charge is vented to the open air, over a suburb, and it lasts about 36 years up there. Until recently this list said that kilogram was worth 4,630 kg CO₂-eq released from a rural stack and nothing at all released over the suburb — a claim neither of its sources makes.

Nobody decided that. ecoinvent has a flow for two air sub-compartments and EF 3.1 has six, so four of the six had no factor from anyone, and this list published what an implementation stated for the flow in front of it. ecoinvent's own report states the opposite rule in §7.3.1:

The IPCC only supplies values for air emissions, without specifying the sub-compartment. The same CF is assigned to an exchange emitted to air for all the sub-compartments.

This list does not follow ecoinvent's rule, or SimaPro's, or anybody else's: it has one convention of its own for every context nobody characterised, the same for every method, in data/context-carry-rules.json. Where a substance has a flow in a context that no deciding implementation states anything for, and the convention names a neighbour for that context, the number this list already publishes for the substance in that neighbour is carried in, as derivation: carried, with the flow it came from as source_flow_uuid. The rural stack and the urban ground level take unspecified air's number; the other outdoor strata take the rural stack's; a river takes surface water's; silvicultural and industrial soil take non-agricultural soil's. Indoor air, long-term air and water and the ocean are walls nothing crosses, so the fumigant's indoor factor is not carried — that is a different fate model, and the earlier rule that carried it in three well-mixed categories (#731) was retired when the convention replaced it. Each rule carries the census numbers it was decided on; How a factor is decided walks through them.

What it will not do, each of them counted rather than assumed:

• Fill a compartment somebody speaks about. Where an implementation states a number the merge withheld — a contested or contradicted question — the compartment belongs to a queue, and filling it would answer that question by accident. This is why the pass runs last, after every ruling.
• Choose between numbers. Where a substance's own air compartments carry more than one value, there is no "the" number to carry and the substance is skipped.
• Publish onto a withdrawn flow. A deprecated flow is a redirect, not somewhere to put a number.

It is also not the inheritance rule #431 removed, and the difference is the same one restated turns on: that rule invented a narrow context's number from a broader one nobody had characterised. This carries a number between compartments the indicator itself says are one statement. Where that is not true of a category, the category is not in the file.

Where the model underneath says something else

Biphenyl is two benzene rings joined together. It is used to make dyes and plastics, and as a fungicide on citrus fruit — an ordinary industrial chemical. EF 3.1 says that a kilogram of it emitted into city air causes 0.2 cases of disease, so a tonne causes two hundred. Ranked against every substance EF characterises for non-cancer human toxicity in that compartment — 3,380 of them — biphenyl comes third, above mercury and just below a rodenticide designed to be lethal in milligrams.

EF 3.1's three toxicity categories are USEtox 2.1. The JRC's own report says so and describes the adjustments made on top of it. USEtox 2.1 says that kilogram of biphenyl causes 0.00000014 cases.

Both columns are in comparative toxic units for humans, which count disease cases per kilogram emitted.

Neither published implementation can see this. The JRC's says 0.19957 because EF 3.1 says 0.19957; the ecoinvent Centre's says the same, because it transcribed the same file. Two copies of one file agreeing is not evidence about the file — and agreed would call it agreement, which is the one case where comparing two implementations gets the answer wrong.

So there is a third queue. contradicted-factor holds 8 questions over 110 factors, from the four substances where USEtox 2.1 is more than a hundredfold away in every compartment — biphenyl, o-phenylphenol, which is biphenyl with a hydroxyl group added, benfluralin and the 2,4/2,6-toluenediisocyanate mixture. The narrowest is benfluralin at 133× and the widest biphenyl at 1,381,461×. What the queue asks is not which of two readings is right but whether the method's number should be published at all, and until somebody answers, the consensus implementation publishes nothing for that substance and category — in every context, not only the six compared, because the number that is out is the substance's. EF's factor for biphenyl in indoor air is 17.194 CTUh, and withholding the six compared numbers while publishing that one would leave the worst of them in the list.

Three things are deliberately not in it. The metals — all 27 disagree with USEtox completely, and by design: the JRC report says EF replaced USEtox's metal factors with ones derived from EU data sources and tabulates each change, copper falling by 100% and silver rising by 813%. Any substance that disagrees in one compartment and not another, which is the two models routing an emission differently rather than disagreeing about the substance.

And freshwater ecotoxicity, whatever the gap — which is 34 of the 38 pairs that were measured. For human toxicity the two publications count the same thing: cases of disease per kilogram, one stated directly and one as the years of healthy life those cases cost, so a gap is the method departing from the model it cites. For a river they do not. The workbook publishes LC-Impact's ecosystem-quality result, a potentially disappeared fraction of species, beside marine and terrestrial numbers EF has no counterpart for at all; EF's CTUe is a potentially affected fraction. Those are two ideas of what damage to a river means, and the third constant is fitted between them rather than converting within one. It shows in the agreement: over the substances not in dispute, 55.3% of the cancer factors and 32.5% of the non-cancer ones land within 1% of the model, against 20.0% of the ecotoxicity ones, whose middle half runs from 0.45× to 1.06× where cancer's runs from 0.9996× to 1.043×. A hundredfold gap measured across that says the two were never the same number, which is true of every substance and is a finding about none of them. The rows stay in the evidence file; nothing is asked about them.

The evidence is a file and the rule is code. data/lcia-underlying-model-factors.json states what USEtox 2.1 gives, per substance, category and compartment, converted from LC-Impact's damage unit into EF's midpoint one by three constants that are measured rather than assumed — where the two files agree the ratio between them is exactly 11.5, 2.7 and 0.0175906. It states no verdict. Which categories are compared, and whether a row in one of them is a contradiction, is decided on each run against the numbers that build states, so a re-transcription that brings the two together publishes the substance again with nobody editing the file.

And this changes nothing about what the JRC published. EF 3.1 states 0.19957, and this list goes on publishing 0.19957 under the JRC's name, in the same table and the same file as before. What changed is that this list no longer says it too.

A ruling is a file, and it carries its reasoning

data/lcia-factor-rulings.json answers a queue item by its own key — the queue, the substance and the category, not a flow, so it survives a rebuild and goes on answering when EF publishes the same substance in a fourteenth context.

Two verdicts, and neither names a source list. publish names the implementation whose number this list takes; decline takes nobody's. A verdict per implementation — jrc, ecoinvent — reads well with exactly two of them and cannot express a third, and every source list that ships factors is one. A factor published from a ruling carries derivation: ruled, distinguishable from agreed, which is the pipeline finding nothing to decide.

A second curated file, data/lcia-factor-adoptions.json, answers about a substance rather than about a question, and publishes as adopted. It holds two populations, and they differ in what they record. The pesticides whose only characterisation is a catch-all's — because ecoinvent's correspondence put them in one and this list took them out again — deliberately do not record the numbers they were written about: if ecoinvent revises the bucket's factor, the revised number is still the bucket's number reaching a named pesticide. The minerals whose number is the JRC's element factors weighted by the formula do record it, in adopted_about, because the arithmetic was checked against today's element factors and a revised number is a new question. Every entry carries one or the other, and the loader refuses one with neither.

Four things the mechanism refuses:

• A ruling with no comment. Every one is a statement about somebody else's published science, and one with no reasoning is unreviewable. The loader raises.
• A ruling that publishes an implementation the question is not about, and a decline that names one anyway.
• A ruling with no ruled_about. It records what every implementation stated when it was made, and one without it is a ruling the staleness check below could never fail — which is the same as not having the check.
• A ruling written about numbers that have moved. A row stating something else asks again rather than being settled by a decision taken about something else. So does a row where an implementation has appeared or gone silent: a third implementation changes what "which of these is right" means.

Eight substances are ruled today, over fourteen rulings, and not one of them is a verdict on somebody's toxicology. Kresoxim-methyl is EF's duplication above: four rulings on the contested-factor queue, publishing the JRC's number in all 24 rows. Carbon Dioxide, To Soil Or Biomass Stock is two rulings on the proposed queue, publishing the ecoinvent Centre's −1 for a removal EF's carbon model cannot express, and sulphuryl difluoride is two more, adopting AR6's own GWP100 for a gas EF never characterised. Sulfur hexafluoride is two contested rulings taking the method's own table. Coal mine off-gas is one more proposed adoption, and the clearest case of the four: EF's own rule is that the CF is 1 MJ/MJ for every fossil resource, and EF files this fossil fuel among the material resources in standard cubic metres, where no indicator reaches it — so the 36 adopted is that rule times the flow's heating value (#774). The remaining three are the declines, and each is a compartment or a convention rather than a number: carnallite, where ecoinvent alone characterises a salt taken from the sea for crustal depletion; water vapour, where publishing ecoinvent's evaporation factor beside the JRC's withdrawal-minus-return factors counted every consumed cubic metre twice; and horticultural peat, where ecoinvent's 9.76 is fuel peat's calorific value on a flow EF files as a renewable material from the biosphere (#772). Where a ruling names an implementation it is never because of whose implementation it is — there is no such rule, and a draft of this design that had one was rewritten to remove it.

A ruling can also be withdrawn, and one was. Two vanadium rulings stood here until 2026-08-18 on the argument that a decimal point moving while five significant figures stay put — 1.3532e-06 against 1.3532e-05, and ten more pairs like it — is a unit slip rather than a model. The digits were real and the conclusion was wrong. EF 3.1 ships two vanadium flows in every context, vanadium (7440-62-2) and vanadium (v), and states the smaller number for the first and the larger for the second in the same compartment: the factor of ten is EF's own speciation factor for vanadium(V), which is exactly why the mantissa survives it. ecoinvent's implementation was reading the right EF row all along, and this project was comparing it against the wrong one, because the correspondence sent ecoinvent's Vanadium V rows to the element. With the routing corrected the two implementations agree to every digit and there is nothing left to rule (#560). A comparison between two implementations is only ever as good as the mapping that decides which two rows are the same question.

An answered question stays on its page, as info rather than blocking, with the verdict and the reasoning on it. The queue is a work list; a settled row is the record.

Read the source before ruling on the transcription

Three greenhouse gases arrived at the contested queue together, all three apparently the same defect — the JRC's EF 3.1 and ecoinvent's implementation of the same method stating different numbers for one substance in one category. Two implementations of one method, disagreeing: a transcription problem, surely, and whoever transcribed later is presumably right.

They are three different things, and only reading IPCC AR6 itself tells them apart. This is the worked example for a habit rather than a rule: a disagreement between two transcriptions is evidence about the transcriptions only after you have looked at what they were transcribing.

AR6 prints the same assessment twice

CFC-11 and HFC-134a are one number each, printed at two precisions in two places in the same report.

• Chapter 7, Table 7.15, Emission metrics for selected species, states CFC-11's hundred-year global warming potential as 6226 ±2297 and HFC-134a's as 1526 ±577.
• Table 7.SM.7, in the chapter's supplementary material, prints 6230 and 1530 — the same values at three significant figures. Its own caption calls it the full table and Chapter 7's points at it.

The JRC transcribed the supplement; ecoinvent transcribed the chapter. Nobody made a mistake and nobody is more current. Where both implementations state a number the merge already resolves it — 0.06% apart is well inside FACTOR_TOLERANCE, so the row is agreed and the more precise value is kept — but in the two air compartments ecoinvent's flow list does not have, aircraft cruise height and indoor air, the JRC was alone and the rounded number went out as sole. The list published 6,230 for CFC-11 released at cruise height and 6,226 for CFC-11 released anywhere else: one substance, one category, two numbers, and nothing on the page to say they were the same number.

That is what data/lcia-rounded-printings.json fixes, and its whole content is those four rows (two substances × two climate categories). A refined factor says derivation: refined, which is neither agreed nor sole, because neither happened: nobody agreed with anybody, and the value published is not the one the lone implementation stated.

It is deliberately not a tolerance rule, and the measurement says why. On the 2026-08-25 build, 97 (substance, category) groups publish two values that agree within 2%. Ninety-three of them are freshwater ecotoxicity, where the two values differ because the compartments differ and both numbers are computed and correct — Alanycarb at 134,083 in silvicultural soil and 135,091 in unspecified soil is a model saying two things about two places. Nothing in the numbers distinguishes those from a rounding. Only the documents do, so a row is written by hand and cites the two printings.

One trap for anybody checking the arithmetic: 7.SM.7's CFC-11 row reads 5560, not 6230, because that row uses the unadjusted radiative efficiency of 0.259 W m⁻² ppb⁻¹. The 12% difference is the tropospheric adjustment, which §7.3.2 assesses as non-zero for CFC-11 and which Table 7.15 includes — 6226/5560 = 1.12. Both implementations use the adjusted value. The rounding is the only difference between them.

Sulfur hexafluoride is two lifetimes, not two printings

The same shape and a different cause, which is why the same fix would have been wrong.

AR6's Table 7.SM.7 gives SF₆ a lifetime of 3,200 years, a radiative efficiency of 0.567 W m⁻² ppb⁻¹ and GWP-20/100/500 of 18,300 / 25,200 / 34,100. That is the JRC's number, and it is the published report: the same in the final government distribution, in the report on ipcc.ch today, and in the authors' machine-readable supplement, which has not changed since September 2021. The February 2022 errata amends one row of Table 7.15 — methane — and nothing else in the chapter.

ecoinvent's v3.12 report (§7.3.4, Table 9) moves SF₆ to 18,200 / 24,300 / 29,000 and calls the change negligible. It is not a rounding, and it is not a corrected printing. Hold AR6's radiative efficiency and replace the 3,200-year lifetime with one near 1,000 years:

Three horizons, three significant figures, one substituted input. A shorter lifetime is what a decade of measurements suggests: WMO's 2022 Ozone Assessment annex lists SF₆ at the 3,200-year lifetime with an alternative range of 850–1,280 years (Ray et al. 2017; Kovács et al. 2017), and its own metrics — 18,400 / 24,700 / 29,800 — sit about 1.6% above ecoinvent's, which is the ratio of WMO's radiative efficiency to AR6's.

So this is not two teams reading one table differently. It is ecoinvent updating an input the method's own publisher has not updated, and the ruling publishes the JRC's 25,200 — because this list publishes EF 3.1, and EF 3.1's climate change category is that table. Adopting a lifetime AR6 did not adopt would make our EF 3.1 something other than EF 3.1, silently, in the column a practitioner reads as EF's. The ruling says in as many words that this is not a claim about which number is better science; ecoinvent's is newer and probably closer, it stays published under ecoinvent's name, and ruled_about ends the ruling if AR7 adopts it.

What each case needed

Same queue, same symptom, three answers. What made them separable was reading Table 7.15 beside Table 7.SM.7 and doing the lifetime arithmetic — perhaps two hours of work, against a rule that would have picked "the newer number" for all three and been wrong twice.

Where to check any of this

The documents, and how to get at them, because the retrieval is the annoying part:

• AR6 WGI Chapter 7 and its Supplementary Material — https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_Chapter_07.pdf and …_Chapter_07_Supplementary_Material.pdf. Table 7.15 is in the chapter, Table 7.SM.7 in the supplement. www.ipcc.ch returns HTTP 403 to most automated fetchers; curl with an ordinary browser user-agent works.
• The machine-readable supplement, which is what ecoinvent's report cites for the metrics — https://raw.githubusercontent.com/chrisroadmap/ar6/main/data_output/7sm/metrics_supplement_cleaned.csv. One row per gas, columns for lifetime, radiative efficiency and every GWP and GTP horizon.
• The AR6 errata — https://www.ipcc.ch/site/assets/uploads/2022/02/AR6_WGI_Errata_20220202.pdf. Worth checking before calling any printed value superseded.
• WMO's 2022 Ozone Assessment annex, for lifetimes and metrics the IPCC has not reassessed — https://csl.noaa.gov/assessments/ozone/2022/downloads/Annex_2022OzoneAssessment.pdf.
• ecoinvent's own report, LCIA Implementation 3.12.pdf, shipped with the release and cached by fetch-lcia beside the workbook. §7.3.1 says where their IPCC numbers come from and states the rule that one air factor covers every air sub-compartment; §7.3.4 and Table 9 list every climate factor that changed since 3.11; §15.3.1 and Table 19 do the same for their EF implementation.

The difference report is a deliverable

Not a backlog. lcia-differences.json is a published, citable artifact whose audience is the implementation teams and anybody comparing implementations, and its headline is the agreement: over the four-list build of 2026-08-29 the published implementations of EF 3.1 agree exactly — same float, full precision — on 36,065 of the 36,744 triples more than one of them states. 98.2%.

Every one of those triples is EF 3.1's. Stepwise 2006 has one published implementation, so nobody is beside it to agree or disagree, and all 9,624 of its factors are counted below rather than banded.

That 98.2% is what makes the other 1.8% readable as signal rather than noise, and it is not a number anybody could quote before: until the implementations are on one flow list, the triples cannot be formed.

A band is not a verdict. 679 differences between competent teams are mostly modelling choices. The 36 over 100× are worth a conversation, not a correction.

Where only one implementation speaks there is no comparison to record, so those are counted rather than listed — 293,104 of EF 3.1's, which is a fact about differently sized flow lists rather than a difference between two readings, and Stepwise's 9,624 on top of them.

A second report sits beside it: coverage across sibling contexts, per implementation rather than between them. Where does one of them characterise a substance in one context and skip the context beside it? It is a summary by compartment and by category and names no flow, deliberately: asked as a list it is 6,885 rows, and the largest populations are substances with no global-warming potential, where absence is the right answer and nobody should be asked to fill it in.

The same page counts this list's own blanks: a flow that exists here, whose substance is characterised in a neighbouring context of the same class — the soils, the fresh waters, the air strata — and for which no deciding implementation states anything. Ammonium to a river was one, with ammonium to surface water characterised next door, until the convention in data/context-carry-rules.json carried the surface-water number in. What the page counts is what is left after the convention has run: on the 2026-09-01 build 3,336 flows for EF 3.1, 3,093 of them the unconfined aquifer, and 2,591 for Stepwise 2006, 2,126 of them aircraft cruise height — each a context the convention leaves blank by decision. Whether a blank takes a neighbour's number, and which, is that convention, written by people from these counts rather than a rule that reads them — How a factor is decided walks through it, and tools/count_context_blanks.py prints the census per pair of contexts, with what each publisher did where it characterised both.

And it counts the blanks beside a substance: an ion with a flow in a compartment, nothing published on it, and its element characterised in the same compartment and category — zinc's ion emitted to water, where every ecoinvent zinc emission lands, beside Stepwise's zinc to water at 133.39 (#922). On the 2026-09-02 build that was 634 pairs over 25 ions for Stepwise 2006 and 168 over 15 for EF 3.1. Whether an ion takes its element's number is a signed entry in the method's adoptions file, substance by substance: 23 Stepwise ions were signed that day in data/lcia-factor-adoptions-stepwise-2006.json, each taking its element's published number in the same compartment as adopted, and the 24 pairs left are chromium's, which Stepwise characterises itself.

Where to find all of this

Two things to know before querying:

• elementary_flows.lcia_factor_count is the JRC's non-zero factors, and keeps that meaning now that there are four implementations, so a query written before characterise existed still answers what it always answered. The per-implementation count is lcia_flow_factor_counts, a view over the factors themselves so the two cannot drift apart.
• lcia_methods on the flow record is where the JRC's factors come from, not a place to read a factor. It is the raw ILCD entry, with no category record behind it — no unit, no area of protection, no statement of who implemented it. It stays because it is a published field with readers.

What this layer does not do

• Decide which published implementation is right. Not this list's question. The consensus implementation answers a different one: what should this list say?
• Aggregate factors into a score. This publishes factors. It does not do LCIA.
• Any method but EF 3.1. ecoinvent's workbook ships 45; the crosswalk covers one, in four implementations.
• Any short-timeframe set, ours or ecoinvent's.