Docs / What was changed in each source

Changes to ecoinvent 3.12

Part of What was changed in each source. What is still open about ecoinvent is on the issue tracker under the label ecoinvent.

ecoinvent 3.12 ships 9,850 elementary exchanges, and it is the first list merged into the transformed EF 3.1 on the build of 2026-08-29 (commit 54b0f9d). Three defects that build showed were corrected the next day, and where a passage below says so the corrected state is the build of 2026-08-30 at 9fbb0d5. Its rows carry a uuid that is the same in every ecoinvent release, so every row quoted below can be found in the ElementaryExchanges.xml of 3.12 and, where the row existed then, of any earlier release too.

The curated corrections are in ecoinvent-3.12-manual-fixes.json: 24 entries in all, of which 9 alter a registry number, 5 a name, 7 a synonym list and 3 a unit. Where a row is sent somewhere ecoinvent's own correspondence table to EF 3.1 would not send it, the target is in ecoinvent-match-overrides.json, which applies to every ecoinvent release because the uuids are shared. The splits and merges below are what the build did with the corrected rows.

One name, two substances

On the 2026-08-29 build of 54b0f9d five names in ecoinvent 3.12 ended up on more than one consensus substance. Two of those splits are deliberate and follow the compartment. The other three were the same substance published twice, and all three were corrected the next day (#825, #826, #827): on the build of 2026-08-30 at 9fbb0d5 only the two compartment splits remain. The corrected three are kept here, in the past tense, so that a reader of either build finds the rows.

Deliberate: the compartment decides

Water (CAS 7732-18-5, 11 rows, m3). Water released to air is water vapour; water released to a river or the sea is liquid water. The consensus list keeps those apart, so the five air rows land on Water vapour and the six water rows on Water.

Water, unspecified natural origin (7732-18-5, 3 rows, m3). The name says nothing about where the water is taken from; the sub-compartment does, and the consensus list names water resources by their origin. So the three rows become three resources.

One substance published twice, since corrected

Carbon (7440-44-0, 7 rows, kg). ecoinvent's plain Carbon emissions are soot — elemental carbon, or black carbon — and this project sends them to the consensus Elemental Carbon substance by a curated target (#671). The target had been written for the four rows ecoinvent's own correspondence table covers. The other three rows, all to soil, had no table entry and were matched on their registry number instead, and 7440-44-0 is also the CAS number on EF 3.1's Carbon — so on the 2026-08-29 build they landed there, and the same ecoinvent name, with the same number, was published as two substances. The curated target now covers all seven uuids, in ecoinvent-match-overrides.json (#827). EF 3.1 publishes elemental carbon in soil only as soil, unspecified, so the three soil rows are published on that flow: a sub-compartment given up for the right substance, and the expectation file for the issue says so.

Whether EF's own Carbon (7440-44-0) and its Elemental carbon (no number) are one substance is not decided by this, and BAFU's five Carbon rows still land on EF's Carbon.

Epoxiconazole (135319-73-2, 8 rows, kg). EF 3.1 carries this fungicide under BASF's development code, Bas 480f, with the registry number 106325-08-0 — a CAS number CAS has since replaced by 133855-98-8, the racemate that is the commercial product. ecoinvent's 135319-73-2 is the CAS number for the compound with its stereochemistry unspecified, and it appears on no EF 3.1 row. Six of ecoinvent's rows also carry the synonym BAS 480F, and those six reach EF's substance on that name. The two rows without the synonym had nothing left to match on and, on the 2026-08-29 build, minted a second substance, Epoxiconazole — the same fungicide twice, and which copy a row reached depended on whether ecoinvent had happened to write the synonym. The synonym is now restored on those two uuids in ecoinvent-3.12-manual-fixes.json (and in the 3.11 file, where the rows first appear), so on the build of 2026-08-30 at 9fbb0d5 all eight take the route the six already took (#826). That EF labels the substance by a development code rather than the name every inventory uses is a separate question, still open on the issue.

Hydrocarbons, unspecified (no registry number, 9 rows, kg). Seven rows are in ecoinvent's correspondence table and go to EF 3.1's Hydrocarbons (unspecified). The two that are not — air / unspecified (3fdc4e95-dc17-5f9f-9ddc-251ecd7c66c0) and soil / agricultural (614d5544-9edf-5b76-90e8-e9c7b98a8084) — carry no CAS number and a name that differs from EF's only in punctuation, and on the 2026-08-29 build they minted a second substance, Hydrocarbons, Unspecified, a duplicate of the first that BAFU's five rows of the same name then joined. Two curated targets in ecoinvent-match-overrides.json now send those two uuids to EF's flows in their own contexts, and five more send BAFU's rows there, so on the build of 2026-08-30 at 9fbb0d5 the duplicate is no longer minted (#825). A synonym written on EF's rows was tried first and moved nothing: a base-list row's synonyms are deleted before the merge builds its name index. The seven rows that reached EF are f9abb851-8731-4c5b-b057-863996a1f94a (air, non-urban), 042e5892-cd59-4e95-949e-cacce0e6a590 (soil, unspecified), 9d508263-9cfd-444a-b1de-73f2bcf39c02 (groundwater), 5e08c84c-69a2-4f0b-b3b3-6b7f0925712b (groundwater, long-term), 049a1473-3a62-4121-982b-5d15d0f2c683 (ocean), 49c42751-1b0b-4ab1-8e70-032e991ce6fd (surface water) and e620a933-c348-4e7e-8893-ab0a0f681c7e (water, unspecified).

An ionic charge was added or corrected

ecoinvent 3.12 writes the charge of a dissolved metal as a Roman numeral — Chromium VI, Sodium I — and the consensus list writes it as Chromium(6+) and Sodium(1+). That is a change of notation and nothing else: 20 names on 265 rows, and every one of them carries the ion's own registry number, so the charge was ecoinvent's and the list only respelled it. Two more names state a charge under the element's number, Strontium II and Caesium I; they are corrected to the ions' numbers below, the way Molybdenum VI was (#934).

Two more names state a charge and were, until #934, published as the bare metal: Rhodium III (16065-89-7, one row) and Palladium II (16065-88-6, six rows). EF 3.1 holds neither ion, so the number found nothing and the vendor's own bare synonym reached the element. The merge now refuses the element for a name that states a charge, and the seven rows mint their ions, Rhodium(3+) and Palladium(2+), spelled by a curated name the way the minted generic ions are. The same uuids were Palladium and Rhodium in 3.8, so 3.8's rows follow 3.12's identity, as Zinc follows Zinc II.

What is listed here is the opposite case: six metals ecoinvent names as an ion without a charge, and publishes without a registry number.

Between 3.8 and 3.9.1 ecoinvent renamed these flows from the bare element (Copper, Iron) to Copper ion, Iron ion, and at the same time deleted the registry number that had been on them — the element's, which was wrong for a dissolved species. The vendor's newest name is taken as its current opinion of what the flow is: a metal ion whose oxidation state the inventory did not record. Picking a charge for it would be this project's guess rather than ecoinvent's statement, so the list publishes the generic ion — written X, Ion — and keeps the charge unstated. Titanium ion is the worked example in When a name and a CAS number disagree: EF 3.1 ships the same name with a CAS number that says Ti⁴⁺, and the name is kept generic there too, because it is the string both vendors write and ecoinvent's rows reach the substance by exactly that label (#698). One charge was confirmed by correcting a number. ecoinvent names its dissolved molybdenum Molybdenum VI on thirteen rows, writes the formula Mo+6 and the synonyms Molybdenum (+6) and Molybdenum ion on them, and from 3.10.1 on registers every one as 7439-98-7 — molybdenum the metal; 3.9.1 shipped the same thirteen uuids under the same name with the ion's own number, 16065-87-5. A stated number outranks a name, so on the 2026-08-29 build all thirteen were published as Molybdenum, the element. The name is believed, as for EF's vanadium (v) and BAFU's Cadmium II, and the vendor's own earlier release says the same: the ion's number is written on the name in ecoinvent-3.12-manual-fixes.json (and the 3.10.1 and 3.11 files), and on the build of 2026-08-30 at 9fbb0d5 the thirteen are published as Molybdenum(6+), where EF 3.1 publishes molybdenum (vi) in the same thirteen contexts (#828). Worth reporting to ecoinvent.

Rows merged into one flow

On the 2026-08-29 build, 25 consensus flows carry more than one ecoinvent 3.12 row. They fall into seven groups.

Endosulfan and its two isomers — 21 rows onto 7 flows. ecoinvent ships Endosulfan, Alpha-endosulfan and Beta-endosulfan in the same seven compartments, and writes the parent's registry number, 115-29-7, on all twenty-one rows. A shared registry number is how this list decides that two rows are one substance, so all three names land on Endosulfan in each compartment. The isomer distinction ecoinvent's names draw is lost: the alpha isomer has its own CAS number (959-98-8) and so does the beta (33213-65-9), and ecoinvent did not use them.

Cypermethrin and its two isomer mixtures — 21 rows onto 7 flows. The same situation: Cypermethrin, Beta-cypermethrin and Zeta-cypermethrin all carry 52315-07-8 in ecoinvent, so all three land on Cypermethrin.

Two names for allyl alcohol — 4 rows onto 2 flows. Allyl alcohol and 2-Propen-1-ol are the common and the systematic name of one compound, and ecoinvent gives both 107-18-6. In air / unspecified, c8d084b4-a980-49a3-a385-8476ed020c6b and f8828e44-6c1e-50ee-afa4-1b2f2a89590c land on one flow; in water / unspecified, 3dbfcc62-4898-48a7-969b-74270615e36e and 0d8d3729-7b68-5669-bc51-fa692e31c561 on another. Published as Allyl Alcohol.

Two spellings of total organic carbon — 2 rows onto 1 flow. Organic carbon (9a891f6c-937c-4226-9702-4a552973ae3f) and TOC, Total Organic Carbon (73b225ab-ddc4-4a38-9ed0-ceedee987424), both in groundwater, both without a CAS number, land on Total Organic Carbon in the unconfined aquifer.

Two groundwater rows of 1,4-butanediol — 2 rows onto 1 flow. ecoinvent's correspondence table sends the whole 1,4-Butanediol family (110-63-4) to a different substance, the epichlorohydrin adduct 2425-79-8, and this project overrides every row of it back to EF 3.1's butylene glycol, which is the same number (#670). EF 3.1 has no groundwater flow for it, so the water / ground- row (c5de5e4d-85cf-4102-9ff1-5248d8928ba1) is sent to the same unspecified-water flow as the water / unspecified row (d835b7aa-288b-4b3a-966b-3f64f36ed220), and the two share it.

Carbon into and out of a soil or biomass stock — 2 rows onto 1 flow. Carbon, organic, increase in soil or biomass stock (9c77a9ae-fcb2-48ed-ae10-4dda59dd6c61) and Carbon, organic, decrease in soil or biomass stock (8c2fe757-6866-4ed2-9f89-81012ad774a0), both resources in ground in kilograms, are mapped by ecoinvent's own correspondence table onto EF 3.1's single Carbon, organic, in soil or biomass stock. The direction is in the sign of the amount in an inventory, so the table treats the two as one flow, and the list follows it.

Groundwater under two names — 2 rows onto 1 flow. Water, unspecified natural origin taken in ground (478e8437-1c21-4032-8438-872a6b5ddcdf) and Water, well, in ground (67c40aae-d403-464d-9649-c12695e43ad8), which ecoinvent files under natural resource / in water despite its name, are both groundwater, and both land on Groundwater.

Land use under two names — 10 rows onto 5 flows. ecoinvent carries an older and a newer spelling of the same land class side by side, and the consensus land vocabulary names the class once:

Registry numbers corrected

Twelve entries in ecoinvent-3.12-manual-fixes.json change a registry number. Each is applied before the row is matched, so the row reaches the substance the corrected number names.

Other corrections that change what a row says

Five entries rename a row, three change a unit, and two restore a synonym — BAS 480F on the two Epoxiconazole rows that ship without it (#826, above). Most renames are housekeeping — Zirconia, as baddeleyite becomes Baddeleyite, the mineral's own name, and Peat becomes Peat, horticulture so that the kilogram row of growing-medium peat stops landing on EF 3.1's megajoule row of fuel peat (#450). Three change the substance the row names, and belong here:

The three unit corrections are the (obsolete) arable occupation rows — conservation, conventional and reduced tillage (fdb1b2d0-f537-401e-b845-1d93da512174, e489cce4-a80f-417d-9ae6-9fc14cc7dd49, 81e07a67-28e0-4392-a553-d86e54a9b8a9) — which ecoinvent ships in m² while its other 57 occupation flows, and every one of EF 3.1's, are in m²·a. They are read as one year's occupation, with a factor of 1.0, which is the conversion ecoinvent's own correspondence tables state for them.