One EU act, twenty-four texts: which one did you read, and on what date?

Every classification you closed this year, every status you assigned yourself, every permit condition you drafted rests on one sentence, in one language, read on one particular day. The question that will reach you is not which rule you applied: it is which text you read, and when. Nothing in your archive answers that, and in six verified cases it changed the result.

They are not translations: they are twenty-four texts

Regulation No 1 of 1958 on the language arrangements lists the twenty-four official languages in Article 1; Article 4 provides that “Regulations and other documents of general application shall be drafted in the official languages”. Drafted, not translated. Hence the formula in the CILFIT judgment (Case 283/81, paragraph 18): “THE DIFFERENT LANGUAGE VERSIONS ARE ALL EQUALLY AUTHENTIC” — in capitals in the published text.

What happens when they diverge is set out in Endendijk (Case C-187/07, 3 April 2008). At paragraph 23 the Court rules out that the wording of one version may “be made to override the other language versions in that regard”; at paragraph 24, the provision “must be interpreted by reference to the purpose and general scheme of the rules of which it forms part”.

The consequence is blunt: the text you read in your own language was the text, not a convenience. If it contained an error, the decision taken on it is not disowned: it is reconstructed, as at that date.

One list, three defects, three durations

Annex I to Regulation (EU) 2021/821 is the European control list of dual-use items. At point 2B510.b.2 — additive manufacturing equipment for metal components — the Italian requires a vacuum of 100 Pa or less. The Swedish, from 15 November 2025, required “lika med eller mer än 100 Pa”, equal to or greater than: the threshold reversed, and the same reversal in Dutch and in Slovenian. Corrected with effect from 21 September 2026: three hundred and ten days.

At entry 1C351.d.4, the Italian said “ricino” — the castor oil plant — instead of “ricina”, ricin, the Chemical Weapons Convention toxin, in Annex I and Annex IV, for four hundred and forty-eight days. We requested that corrigendum (CELEX 32022R0001R(03)) in all twenty-four languages: it answers in Italian and in no other. Third, point 3B501.f.1.a: where the Italian set a light source wavelength of less than 193 nm, it now reads “non utilizzato;”, not used. Eighty-seven days. We have written about those three: what follows shows they were not an accident.

Seven corrigenda, and a consolidated text without them

On Regulation (EU) 2025/40 on packaging we queried CELEX 32025R0040R(01)-R(10) in twenty-four languages: seven corrigenda answer, covering thirteen language versions — three for Spanish, two each for Italian, Czech and Swedish, one each for the other four. English never.

The most recent, of 18 September, rewrites Article 3, point 15, subparagraph (c) — the definition of “producer”, on which registration and extended producer responsibility depend — replacing “direttamente ai consumatori”, directly to consumers, with “direttamente agli utilizzatori finali”. The English has said “directly to end users” since 22 January 2025, and the Spanish corrigendum had already touched that same subparagraph on 9 September 2025. We have covered what that means for firms selling abroad. What is new is the rest: the Italian consolidated text, checked this morning, still carries a single ►C1 marker and subparagraph (c) still reads “direttamente ai consumatori”. Five days on, the text people consult is the old one.

Two drafts and two quotation marks never closed

Commission Implementing Decision (EU) 2026/2015 touches sixty-nine rows of the list of harmonised standards for machinery, sixty-seven inserted and two replaced. We compared them one by one against the English: every row number matches, two references do not. At row 32a the Italian cites “(ISO/FDIS 3744:2024)” — a draft stage — where the English cites “(ISO 3744:2025, including corrected version 2026-01)”. At row 209a, press brakes, the Italian cites “(ISO/FDIS 6909:2025)” and the English “(ISO 6909:2026)”. Anyone copying the whole row into the EU declaration of conformity carries a draft with it.

In the same Italian annex we counted sixty-nine opening quotation marks and sixty-seven closing: rows 215a and 223a — generating sets and industrial robots — open and never close. In the English annex the count balances.

Yesterday: one Official Journal number, two different errors

On 22 September the Official Journal of the European Union, L series, published under number 2026/90789 a corrigendum to Directive (EU) 2024/1785 on industrial emissions. CELEX 32024L1785R(01), requested in twenty-four languages, answers in two — Italian and Hungarian — and the two texts do not correct the same thing.

The Italian touches recital 19 and the second paragraph of the new Article 7 of Directive 2010/75/EU, on incidents and accidents: where drinking water resources are polluted, the competent authority now informs not “i gestori dell’acqua potabile e delle acque reflue” but “i gestori dei servizi idrici di acqua potabile e acque reflue” — the operators of the water services, not of the water. That is already the wording of Article 8(6): the Italian text diverged from itself on who must be told. The Hungarian corrects something else — in Annex I, point 5.3, “biológiai tisztítás” becomes “biológiai kezelés”, biological treatment — the activity that determines whether a waste treatment installation needs the permit at all.

The transposition deadline was 1 July 2026; the Italian draft decree was approved at first reading on 22 June, on the earlier text.

Not an accident, a constant

Four acts, six defects, four years: a reversed threshold, a toxin mistaken for a plant, an entry replaced, a definition narrower than it should be, two standards cited in draft, corrigenda that exist in one language only. This is not one office slipping: it is how a legal order works that produces twenty-four equally authentic texts and corrects them at different speeds.

For you, one thing changes, and it is the one that counts: your position is not defended with the rule, it is defended with the text — that row, that day, in that language.

Why a closed system

The archive you need has four requirements, and all four rule out an external service. The languages that matter are not one: the text a customer or an authority in another Member State will read is theirs, and it belongs next to yours, with the diff. The memory is dated: a consolidated text is a page that changes by itself, and what defends you is the row as it stood on the day of the decision, with the markers then in force, and it must stay queryable that way for years. And the indexed object is not the legislation: it is the link between one of your catalogue entries, a customer, a destination, an installation and a line in a list — the technical and commercial portfolio, often bound by contract clauses. Send that to a service that passes it on to somebody else’s model and you hand over the map.

Compliance. The system watches Cellar in the languages that concern you, pulls down corrigenda and consolidated texts, computes the diff against the version you decided on and opens a check only on the rows touching your products, markets or permits, with the dated trail to show an inspector.

Decision-making. The same base holds catalogue, customers, destinations, technical files and environmental permit conditions in a single operational model, on which AI agents prepare the review and close it with one of your engineers in command — for large enterprises, defence, public administration and healthcare. On-premise on standalone machines needing no deep integration into your network, or on dedicated cloud with a data centre in Italy, always co-managed.

If you cannot say which text you decided against, and on what date: write to us.

What we do not know

We do not know how many Italian decisions went through the wrong rows: no public data says so. For technical standards we report number, title and edition as they appear in the annexes: those texts are paid-for and we have not read them. The count of corrigenda is what Cellar answers today — one published but not yet indexed we would not have seen — and testing language by language measures the availability of the document, not the existence of the error. On the Italian transposition we report what the Council of Ministers stated, not the draft.