Montaigne complained, in the last of his Essays, that there is more work in interpreting interpretations than in interpreting things, and that we do nothing but gloss one another. Competition law has reached the state he described. Article 102 TFEU is two sentences long. The Court of Justice and the General Court have spent more than fifty years interpreting it. On 3 September 2026 the European Commission adopted a third layer, its Guidelines on the application of Article 102 TFEU to abusive exclusionary conduct by dominant undertakings (C(2026) 6118 final), 246 paragraphs and 494 footnotes that withdraw the 2009 Guidance on enforcement priorities and take its place. Paragraph 8 describes the new text as an outline of “the current state of the case law of the European Union Courts”, with the Commission’s own positions flagged where the case law is silent. For the next decade, national courts, national competition authorities and companies will pick up this gloss before they pick up the judgments. I went back to the judgments. Everything below compares the Guidelines with the rulings they cite, not with the Commission’s decisions and not with the Commission’s own account of what the Court has held.
The method, in four sentences
The footnotes contain 1,096 pinpoint citations to 90 judgments of the Court of Justice and the General Court. Each cited paragraph was retrieved from the judgment and analyzed next to the sentence it supports. The direction was then reversed. A knowledge graph of 1,015 antitrust judgments listed, for every concept the Guidelines address, the judgments the record associates with it, and the uncited paragraphs of the 31 judgments the Guidelines rely on most were analyzed for anything that qualifies what the Guidelines say.
Four claims
The first claim concerns accuracy. Of 1,096 citations, 1,093 could be matched to the judgment text, and all but one of them say, in substance, what the footnote attributes to them. Four citations carry an error of form, three misstated case numbers and one wrong paragraph number. Whoever drafted this text read the cases. That deserves saying before anything else, because the three claims that follow are about what the text does with the cases it read.
The second claim concerns scope. Twelve statements of law go beyond what the paragraphs cited for them hold, and four of the twelve carry the enforcement architecture. Paragraph 58 states a general theory of presumptions under which “the more a given conduct is considered generally likely to distort effective competition, the less case-specific evidence is required”, and under which presumptions “shift the evidentiary burden from the Commission onto the dominant undertaking”. Paragraph 64 states that where conduct can exclude an equally efficient competitor there is no need to show separately that it departs from competition on the merits. Paragraph 128 states that once a margin squeeze leaves an equally efficient rival with a negative margin, the Commission “may conclude” that competition is distorted, “absent any evidence to the contrary”. Paragraph 199 states that conduct “by its very nature harmful to competition” is “deemed as such to distort effective competition”.
The Court has said each of these things about something narrower than the Guidelines say them about. It has recognized presumptions for two practices, exclusivity since Hoffmann‑La Roche and pricing below average variable cost since AKZO, and the General Court in the Intel case on remand described the first as a presumption and not a per se infringement. No judgment states a sliding scale between the general likelihood of harm and the quantum of evidence. Paragraph 8 says the Commission will explain its own position where the case law is silent. Paragraph 58 is not presented as such a position.
The third claim concerns omission. On the questions that decide cases, the Guidelines cite the paragraph in which the Court supports a presumption and stop before the paragraph in which the Court says what the Commission must still prove. Sometimes they stop in the middle of the paragraph. The Court’s statement in Unilever of the evidential floor under every finding of “capability” is cited at paragraph 82 for its first half, that effects must be more than hypothetical, and its second half, which says who bears the doubt, is nowhere in the text. Google Shopping and Google Android are each cited for the proposition that the as‑efficient‑competitor test is not always required, and in each the adjacent paragraph, the one that states what the Commission must still establish, is left out. TeliaSonera is cited at paragraph 128 itself, and the paragraphs of TeliaSonera that say when a squeeze produces no such effect are not. Post Danmark II is cited for the absence of a de minimis threshold and not for the sentence, one paragraph later, that states the standard of probability those paragraphs belong to.
The Guidelines also cite the judgments in which the Commission lost, in whole or in part, on a theory they now codify, Intel, Qualcomm, Google AdSense, Google Android in the General Court, and Google Shopping in the General Court (in those two, on one part of the decision only, the rest of the infringement standing). Two of them, Qualcomm and the Intel appeal, are cited for their annulment paragraphs, recast as grounds on which an undertaking may rebut a presumption, and in one case with a condition added that the judgment does not contain. The other three are cited for what they upheld and not for the paragraphs in which the Commission’s proof was found wanting. The margin‑squeeze judgment in which the General Court annulled part of the infringement period for want of proof, Deutsche Telekom in 2018, is not cited at all. The omissions do not all run one way. The Court has also said things the Guidelines could have used and did not, in Tomra on retroactive rebates, in Compagnie Maritime Belge on above‑cost selective pricing, and in Commission v Intel on how much of an as‑efficient‑competitor test a firm must dislodge before doubt arises. Those passages are few and, for the most part, superseded. The passages that state what the Commission must prove are many, recent, and drawn from the judgments the Guidelines cite most.
The fourth claim concerns vocabulary. The phrase “presumption of innocence” does not appear in the 246 paragraphs. Neither does “benefit of the doubt”. The word “doubt” appears once, in paragraph 10, in the expression “for the avoidance of doubt”. The phrase “burden of proof” appears once, in paragraph 204, where it rests on the dominant undertaking. These are not words the Court uses sparingly. The presumption of innocence appears by name in at least 130 judgments of the record, and in seven of the 90 judgments the Guidelines cite, in the Court’s own reasoning in Intel, in Telefónica and in Google Android at first instance. The rule that any doubt must operate to the advantage of the undertaking appears in at least 73 judgments of the record, and in eight of the 90 the Guidelines cite, Unilever among them. Every annulment named above turned on proof the Court found insufficient.
What the silence costs
Cesare Beccaria wrote in 1764 that no one can be called guilty before the judge has pronounced sentence. The Court of Justice applies that principle to undertakings facing competition fines. The Commission must prove the abuse, and any doubt goes to the undertaking. The Guidelines say nothing about that principle, and they set out presumptions that move the burden the other way. That silence has no effect on the law, because a guideline is not law. It binds only the Commission, and paragraph 8 accepts that national courts and authorities are free to follow it or not. The silence has its effect later, in court. When the Commission adopts a decision on the strength of paragraph 58 or paragraph 128, the undertaking will appeal, and the Court will judge that decision by its own case law, including the paragraphs the Guidelines pass over. Those are the paragraphs on which the Commission has lost before. David Hume asked that a wise man proportion his belief to the evidence. The Guidelines are a faithful account of the evidence the Commission has won with. The record also holds the evidence it has lost to, and the Court, unlike the Guidelines, will read both.
*
Below the paywall is the audit behind these claims, about 9,000 words. It contains five things the quotations alone do not give:
For each of the nine practices covered in Section 4 of the Guidelines, from predatory pricing to self‑preferencing, a short verdict on whether the text matches the case law, goes beyond it, or leaves something out.
A list that pairs each presumption in the Guidelines with the paragraph in which the Court states what the Commission must prove, and with the case the Commission lost on that point.
A description of the four ways in which the Guidelines leave paragraphs out, so that anyone can run the same check.
A comparison of what the Court and the Guidelines say about economic evidence and the as‑efficient‑competitor test.
And a checklist of what a Commission decision based on paragraph 58, 128 or 199 will have to show to survive an appeal.
Every quotation is reproduced word for word, with its paragraph number. It is written to be dropped into a brief.
