Stratosphere · 03
Six ways a proof fails
On a hard deadline on 27 September, four in five promises withdrawn, a data centre that is not coming back — and why trust infrastructure breaks in known patterns, not by chance.
On 27 September 2026 a particular kind of sentence becomes expensive in Germany: “climate-neutral by 2030 …”, “plastic-free by 2028 …”. Whoever makes such a promise to consumers is, from that day and by statute, acting misleadingly unless a detailed and realistic implementation plan stands behind it. The plan must contain measurable, time-bound targets and be reviewed regularly by an independent external expert — and the expert’s findings must be made available to consumers.1 The date comes from the directive, the wording from the Federal Law Gazette in February.54
One can read this as a story about regulation, and that is how it is being received. The more interesting part sits a little more hidden in the same publication, where the legislature works out what the rule will cost. The estimate — 52,500 affected products, 22 hours of effort per product, €2,000 in external costs per case — rests on a single assumption that is neither justified nor discussed: it is expected that 20 per cent of companies will substantiate their environmental claims. The rest delete them.3
Four in five so-called green claims will disappear, and that is no critic’s forecast. It is the arithmetic of the statute that was meant to make promises hold. This series has described in two pieces how trust infrastructure fails — once invisibly, once as theatre. This piece argues that the cases added since then do not form a list but a typology: six ways in which a proof can fail, each with its own tell, and each from this year, documented with a date.
Six ways
The first two are known. With cereulide in infant formula and the tampered jar in Burgenland, the system knew nothing; it learned of its own failure from the people it was supposed to protect. That was invisible failure. With the foodwatch salmon count, the system knew everything — the certificates existed, the codes were printed — and the interface offered to the person holding the product returned noise. That was theatre.
The other four have been documented in the months since August, and they differ in what is missing. In the third, what is missing is not knowledge but queryability: nobody can say which articles carry a claim that becomes inadmissible on 27 September — a master-data problem that presents as a compliance problem. In the fourth, nothing is missing at all: the dependency is quantified, the risk is named, and nothing is done. In the fifth, the system cannot read the proof because it was issued in another jurisdiction. And in the sixth, it no longer matters who owns the infrastructure, because it no longer exists. For a business the distinction is not academic: what helps depends on what is missing.
What was priced in
The 27th of September is the best-documented of the six, because the legislature described its own effect in advance The requirement is meant seriously: clear, objective, publicly accessible and verifiable commitments, a plan with resources, a review body, a publication. The cost calculation behind it is meant seriously too, and it says something different. For the external review of an implementation plan it allows one working day, eight hours, priced at the daily rates of environmental verifiers.
Who that reviewer is stands in the explanatory memorandum more precisely than the reporting conveys. First in line are the environmental verifiers licensed under the Umweltauditgesetz; auditors and statutory accountants come after, “in individual cases”, limited to their area of competence.2 The Institute of Public Auditors nonetheless published, on 10 September, an interpretation paper with sixteen questions on reviewing such plans, and framed the audit object as the statute frames it: what is reviewed is the implementation plan, not the environmental claim. Whether the target is reached is not part of the review. Whether the path to it is plausibly described is.6
Promise Plan Review Answer
● ───────────── ● ───────────── ● ○
claim delivery one day on request
That is a weaker assurance than a consumer will read into a reviewed promise. One day of review of a plan whose fulfilment is not reviewed produces a document that holds legally and says little operationally. It is the same mechanism as the seal that returns noise — only this time built by the statute itself, with a cost estimate attached. And companies had evidently done the arithmetic already: four days after the auditors’ paper, their board spokeswoman warned that the rule must not lead to meaningful sustainability communication declining. On the same day, two agricultural marketing associations in the north-east reported that businesses were already taking their regional-origin labels off products, as a precaution, for fear of cease-and-desist letters.7
"Four in five so-called green claims will disappear, and that is no critic's forecast. It is the arithmetic of the statute that was meant to make promises hold."
Warning, occurrence and pricing-in are thus all on the table, in that order reversed: the publication of September 2025 calculated the effect, the institute warned of it in September 2026, and the associations reported it in the same week. The reflex is older than this amendment. In August an animal-welfare foundation counted 314 aquaculture products across eight retail chains: the private labels were 90 to 100 per cent certified, and not a single chain had a binding purchasing policy. What happens almost everywhere in practice is not promised, because the promise would be actionable and the practice is not.8 The 27th of September turns that caution into a rule.
What knows itself
The fourth kind has the clearest shape, because it answers the question of knowledge completely and still changes nothing. In July 2026 the ifo Institute asked the firms in its business surveys how dependent they are on digital products from US providers — cloud, software, AI. 88 per cent use them; 31 per cent see themselves as strongly dependent. Asked how risky that is, 32 per cent of those affected answer “very”, and of those who see any risk at all, a good four in ten plan no countermeasure whatsoever.9
Food and beverage manufacturers sit at the bottom of this survey: a quarter strongly dependent, a fifth concerned, the lowest risk rating of any sector. The authors read that not as composure but as underestimation — infrastructure embedded deep in the working day is no longer perceived as dependency. And they give this kind of failure an explanation better than inertia: it is a coordination problem. Switching to European alternatives is the most frequently named measure and at the same time the one whose feasibility lies least in the hands of the individual firm. Demand is waiting for a supply that is waiting for demand.
The occasion for the special question was a date. On 12 June the US Department of Commerce had prohibited one provider from making its most capable models available to foreign nationals; the company switched them off worldwide, and on 30 June the order was lifted again. Economically it left no trace, and that, the authors write, is the point: it showed that the instrument exists and is used. A proof that is fully present and moves nobody is a kind of failure of its own, and it is the most common.
What six states cannot read
On 11 May the federal government notified Brussels of a draft third act on animal-husbandry labelling. It was to extend the state husbandry logo from the supermarket shelf to out-of-home catering, to processed products and — this is the part that counts — to imported meat. On 12 August the Commission issued a detailed opinion and held central provisions to be contrary to Union law; the standstill period was extended to 12 November. Latvia, Poland, Austria, Czechia, France and Spain filed detailed opinions of their own.10
The objection is not that the German standard is too high. It is that the system can only read proofs shaped like its own. Foreign producers would have to demonstrate compliance with the German scheme or be classified automatically into the lowest tier; other certification schemes would not be recognised. That breaches the free movement of goods — and, second ground of complaint, the EU organic regulation.11 The tell of this kind of failure is an inversion: a farm working to a standard the logo cannot express lands not in the middle but at the very bottom. The trust system is national. The flow of goods is not. Six member states have written the same sentence.
Where ownership no longer changes anything
The sixth kind is the youngest and the only one in which the question of who owns something plays no part. On 1 March, drones struck two Amazon Web Services data centres in the United Arab Emirates directly; a third, in Bahrain, was damaged by a nearby strike — structural damage, power loss, firefighting water. On 15 September, six and a half months later, the provider stated in its own status portal that access to resources and data held exclusively in one of the three Emirati data-centre segments — availability zones, in the provider’s language — cannot be restored; the same holds for Bahrain, with an update early in 2027 at the earliest. The damage, in its wording, spanned several of these zones and exceeded what its regional and multi-zone services are designed to withstand.12
This year’s sovereignty debate turns on ownership, jurisdiction and licence withdrawal — on whether someone may switch off access. Here nobody switched anything off. The site is gone. Whoever had their only copy there has lost it, regardless of which country the provider sits in and which law governed the contract. Exactly one thing helped in this case, and it is not a contract clause but an operating habit: a second copy in another region. That removes the convenient answer the fourth kind suggests. A European provider would have been just as gone in Bahrain.
What a typology does
A list of incidents invites answering each one on its own — with a recall, a seal, a lawyer, a migration. A typology asks a different question: which of the six ways a business is standing in before the incident arrives. The six differ in what is missing, and therefore in what would help. Where knowledge is missing, measurement helps. Where queryability is missing, master data helps. Where nothing is missing but the consequence, only an occasion helps — and it comes, as in June, from outside and within hours. Where the system cannot read the proof, no better proof helps. And where the site is missing, only what already lay in a second place helps.
Promise Plan Review Answer
● ───────────── ● ───────────── ● ───────────── ●
claim delivery outcome on request
"That is not the failure of the rule. It is its first measurement."
The series began with a question that has since sharpened rather than softened: when a specific claim is challenged, can an organisation answer it with evidence — or only with further reassurance? The 27th of September gives that question a procedure and a responsible party, and the cost calculation behind it predicts that most businesses would rather withdraw the claim than answer it. That is not the failure of the rule. It is its first measurement. What remains standing afterwards are the promises with substance behind them — and from that date on, which of the six ways one’s own company stands in is no longer a theoretical question.