Eubulides of Miletus, a Greek logician of the fourth century, asked how many grains make a heap. One grain is not a heap. Adding one grain to what is not a heap does not make it one. So there is never a heap. The puzzle has kept philosophers busy for twenty‑four centuries because the answer is obvious and the reasoning is not. On 9 September 2026, the General Court gave merger control its own answer. An increase in market share that, in the Court’s words, may be “limited to a few tenths of a per cent” can be a significant impediment to effective competition, because the harm lies in what the increase keeps in place. What matters is not the grain. It is the pile it lands on.
The judgment in Booking Holdings v Commission (T‑1139/23) dismisses in full the action against the Commission’s decision of 25 September 2023 blocking Booking’s acquisition of eTraveli, a Swedish flight agency. The two companies barely compete. Booking sells hotel rooms and holds a share of the hotel online travel agency market in the European Economic Area that the decision puts between 60 and 70 per cent. eTraveli sells flights, a market where Booking is not dominant. The Commission’s case was that flights are the “entry point” of a trip, so that a flight customer can be sold a hotel room before, during or after the flight purchase. Owning eTraveli would therefore let Booking pull more hotel customers into a platform already too large for rivals to contest.
I analyzed the judgment against a knowledge graph built from EU courts competition judgments. I matched every authority the Court cites to the paragraph it cites and traced which neighbouring authorities the Court left out. What follows is what came out, in four points. But first, the method.
The method, in three sentences
The Court’s 527 paragraphs name 26 earlier rulings, two of them in Booking’s submission. Each citation was matched to the cited paragraph in the official text and checked for fit. The judgment was then placed in the graph, which shows which authorities sit closest to the propositions it decides, so that omissions could be identified and each confirmed by a full‑text search of the judgment.
The judgment in four points
The first point concerns what makes an impediment to competition “significant”. Booking argued that an increment of the size at issue could not fetter competition and that the Commission had presumed harm from any increase in an already dominant position. The Court agreed with the premise. Dominance does not automatically produce a significant impediment (paragraphs 463 and 464). It then held that the Merger Regulation does not prevent the Commission from relying on a theory of harm in which the concentration “would sustain” an already low level of competition rather than reduce it, the significance lying in that low level “becoming consolidated and perpetuated” (paragraphs 467 and 468).
I checked those two sentences against the corpus, 1,015 judgments of the Court of Justice and the General Court, 155 of them on mergers. Neither sentence has a predecessor, in those words or in that logic. No merger judgment locates the significance of an impediment anywhere other than in a reduction of competitive pressure. I repeat, no EU court had written that before. The two paragraphs carry no citation, and their origin is visible one paragraph earlier, where the Court records that the Commission said at the hearing that the “primary intention” of its theory of harm was to stop the expansion of rivals from becoming increasingly difficult (paragraph 466).
The Court fenced the formula with two circumstances. A market with strong network effects and a wide gap between the leader and its rivals, in which even a small increase in share can strengthen the network effects. And growth achieved through flights, one of the few customer acquisition channels Booking does not yet dominate (paragraphs 470 and 471). Both circumstances describe the market and the channel. Neither describes the increment.
How small is small? The decision’s figures are redacted from the public judgment, every one of them shown as a band between zero and five per cent. The Court’s adjectives are not redacted. The increment is “marginal” (paragraph 144), “modest” (paragraphs 396 and 440) and “relatively small” (paragraph 399), and after the Court’s corrections it cannot be ruled out that it “would be limited to a few tenths of a per cent” (paragraph 465). That is the sentence the title paraphrases, and it is the sentence an appeal will test.
The second point concerns the theory of harm itself, which the Court admitted, and a defence it removed. Booking called the theory “reverse leveraging”. Leveraging means using a position on one market to strengthen a position on another, and for more than twenty years it had run in one direction, from the market where the firm has power to the one where it does not. Tetra Laval, in 2002, was a dominant carton maker pushing into plastic bottling equipment. General Electric, in 2005, was the leading engine maker pushing into avionics. Booking runs the other way, from flights, where Booking is not dominant, to hotels, where it is. The label enters the record at recital 197 of the decision, in the Commission’s summary of Booking’s reply to the statement of objections, and Booking described the theory to the Court as novel and unsubstantiated (paragraph 65).
The Court held that the Non‑Horizontal Merger Guidelines do not preclude the theory (paragraph 87) and that the impact on competition is “essentially the same, irrespective of the direction of the leveraging effect” (paragraph 88). It added that it “would be contradictory” to let the Commission rely on leveraging only where it strengthens a non‑dominant position (paragraph 89). I checked the phrase and the logic. “Reverse leveraging” appears in none of the 1,015 judgments and in one of the 8,850+ Commission merger decisions in my decisions corpus, this one. In the 155 merger judgments, leveraging never runs from a market without power into a market with it. Booking is the first judgment in the corpus in which it does, and the Court accepted the direction in three paragraphs without an authority for the step. The Court then measured the theory against the Guidelines’ own three‑step test for conglomerate foreclosure, ability, incentive and significant detrimental effect. It found the first two steps common ground, because Booking’s aim was to cross‑sell hotel rooms to flight customers (paragraphs 84 to 86), and it sent the third to the formula described in the first point (paragraph 92).
The defence the Court removed is competition on the merits, the phrase Article 102 case law uses to separate normal competition from abuse. Booking’s argument was that offering a hotel room to a flight customer, with no pressure on the customer, is competition on the merits. The Commission had engaged the argument on its own terms. Its decision says that Booking’s growth would come “not from competition on the merits but to a significant extent from taking advantage of its brand strength and end‑customer inertia” (recital 208). The Court went further than the Commission. It accepted that cross‑selling is not tying (paragraph 105) and held that the concept “does not play a similar role in the context of merger control”, because “the implementation of a concentration cannot be regarded as a form of competition on the merits” (paragraphs 106 and 107). No merger judgment in the corpus says that. The consequence is practical. A benefit that an abuse case would weigh as competition on the merits now counts only as an efficiency, where the burden lies on the notifying party and the evidence must have been on the file during the administrative procedure (paragraphs 101, 102 and 479).
The third point concerns the arithmetic. The Commission had calculated the increase in Booking’s hotel market share using two methods derived from Booking’s own valuation model. The Court excluded the figures that counted hotel bookings combined with taxis and car rentals, since those sales do not depend on flights (paragraphs 322 to 326). It discarded the second method entirely, because the Commission could not show that it had applied a cannibalisation rate and could not explain, at the hearing, the gap between its two results (paragraphs 357 to 359). It held that the remaining figure was an overestimate, because the Commission had frozen the size of the 2025 market at its 2022 level although its own decision assumed inflation of 5.4 per cent for 2023 (paragraphs 367 and 368).
Two more findings went against the Commission. It had failed to properly consider an argument Booking raised about channel managers, the software that lets hotels list on many agencies at once, in breach of its duty to examine all relevant aspects carefully and impartially (paragraphs 415 to 417). And it had rejected Euromonitor’s market data as unreliable and relied on it de facto (paragraph 185).
None of this annulled anything. The Court of Justice held in CK Telecoms in 2023 that the General Court must conduct an overall assessment of the factors that survive review, and the General Court did exactly that (paragraphs 156 and 473). A decision can carry its arithmetic wrong and its conclusion intact. The first point explains why. Once the harm is consolidation rather than a measured loss of competition, the number stops deciding the case, and what was left of the number here was small by the Court’s own description.
The fourth point concerns what the Court did not name and what it did not answer. Commission v Tetra Laval, the 2005 judgment of the Court of Justice, is the highest‑degree authority in the judgment’s neighbourhood of the graph, at 44 edges, and it is nowhere named by the Court. It is the ruling which says that in a conglomerate merger the chains of cause and effect are “dimly discernible, uncertain and difficult to establish”, so that the quality of the Commission’s evidence is “particularly important”. The General Court cites its own Tetra Laval judgment of 2002 twice, for the definition of a conglomerate merger and for the proposition that prior dominance is not enough (paragraphs 72 and 463), and not for the paragraphs that require “convincing evidence” and a “particularly plausible” analysis of the future. General Electric is cited only for the form of a decision (paragraph 154). Impala is not cited at all.
The route taken instead is paragraph 77 of CK Telecoms, cited at paragraph 457, which says that the Tetra Laval line reflects only the essential function of evidence and does not affect the standard of proof. The paragraph before it restates that in a conglomerate case the quality of the evidence is particularly important. The paragraph after it says that the complexity of a theory of harm counts in assessing its plausibility. The Court cited the middle paragraph and neither neighbour, then stated on its own authority that merger control “would be undermined if it were subject to a particularly high burden of proof to demonstrate the likely growth of a dominant undertaking’s competitors” (paragraph 439).
Booking had also invoked Continental Can and Towercast, which say that strengthening a dominant position through an acquisition is caught only where the degree of dominance reached substantially fetters competition, so that only firms dependent on the dominant one remain (paragraph 455). Continental Can appears once in the judgment, in that paragraph. The Court answered with its two circumstances and did not say whether that test applies in merger control or how a few tenths of a per cent meet it.
Would citing these authorities have changed anything? For most of them, no. The result would stand and the reasoning would be longer. Commission v Tetra Laval and General Electric would have obliged the Court to say how the quality requirement was met once the increment could not be sized, which paragraph 77 of CK Telecoms lets it avoid. Impala would have answered the word “presumption” by name.
Continental Can is different. Its test asks whether only firms dependent on the dominant one remain, and the Court’s own findings show Expedia with a share of 10 to 20 per cent on the business side of the market and four named rivals able to expand (paragraphs 19 and 439). A court that applied that test would have had to explain how consolidation meets it. The audit below sets out, authority by authority, what a citation would have changed.
The EC decision, as I analyzed it before the judgment
When I analyzed the Commission’s decision, four things stood out. The Commission reasoned from position to harm. It inferred that Booking’s ecosystem would become harder to contest from what Booking would hold rather than from a tested capability that rivals could not reproduce. It discounted the counterfactual the parties proposed rather than running it, although the two companies had operated under a commercial agreement since 2019 and extended it on the day of the prohibition. It never asked whether Booking’s fuller funnel reflected a better product that consumers prefer or lock‑in through control of a customer acquisition channel. And its treatment of conglomerate proof sat uneasily with Commission v Tetra Laval.
The Court takes a position on each. It accepts reasoning from position, provided the market has the two features described in the first point (paragraphs 397 to 399 and 470). It holds the counterfactual dispute incapable of changing the outcome once the increment is marginal in every scenario (paragraphs 143 and 144), and records Booking’s submission that eTraveli could not renegotiate the existing agreement before 2028 (paragraph 134). It answers the reproducibility question from Booking’s own internal documents, which describe an ecosystem “difficult for competitors to replicate” (paragraphs 293 to 295), without naming the data and ranking system beneath it. On the merit question it goes further than the Commission, in the two sentences. And on Commission v Tetra Laval it says nothing.
What it does to the draft merger guidelines
The Commission’s draft merger guidelines name entrenchment as a head of harm. Paragraph 252 defines it as the merged firm gaining “control over certain assets” in a way that “structurally creates or reinforces existing barriers to entry and expansion resulting in reduced market contestability”, within a core market or across an “ecosystem”. Paragraph 256 lists the assets that count, data, intellectual property, infrastructure and “an important customer acquisition channel”, and the footnote to that last item cites one decision, Booking/eTraveli. Paragraph 32 states the standard of proof, “more likely than not”, and says it applies “irrespective of the nature and the complexity of the theories of harm”.
The judgment does three things to that text:
It gives paragraphs 252 to 256 the judicial anchor they lacked. The draft’s only example for the customer acquisition channel was a decision under appeal. It is now a decision upheld, and the Court’s two circumstances, network effects with a wide gap between leader and rivals, and a channel the leader does not yet dominate, read like paragraphs 254 and 256 of the draft.
It also shows that the draft’s fence is wider than the Court’s. The draft speaks of control over assets in general, capabilities included. The Court upheld the theory on the narrower facts of a channel and a market, and said nothing about assets. A decision that reaches for the draft’s wider wording will be citing a judgment that did not go that far.
And on evidence, the draft is more complete than the judgment. Its footnote on the standard of proof cites paragraphs 76 and 77 of CK Telecoms together, the quality requirement and its limit. The Court cited 77 alone.
One vocabulary note. The draft says at paragraph 56 that firms may acquire market power legitimately “through internal growth and competition on the merits”. The judgment says that implementing a concentration is not competition on the merits. The two sentences are consistent, and read together they draw a line the Commission had not drawn in the decision itself, between growing and buying.
What the outcome settles
Pascal wrote that if Cleopatra’s nose had been shorter, the whole face of the earth would have changed. His point was that a small cause can have vast effects. The judgment states the mirror image. A small cause can be enough to keep everything as it is, and keeping everything as it is can be the harm. Whether the Court of Justice agrees is the next question. Until it answers, the Commission holds a formula for significance that does not require the level of competition to fall, and a theory of harm that runs from the market a company does not dominate to the one it does. It also holds a prohibition that survived the loss of most of its quantification. Companies planning acquisitions in markets with network effects should take the judgment at its word.
*
What follows below the paywall is the audit that sustains the four points above. It is written to be dropped into a brief. It contains:
The authorities the Court did not cite, adverse and supportive, with the passages the Court would have had to confront and what citing each would have changed.
The draft merger guidelines set against the judgment, paragraph by paragraph.
The judgment in one map, plea by plea, with the paragraph where each point is decided, and two figures, the five‑step theory of harm with what survived review, and the precedent path with the authorities the Court took and the ones it left.
The passages that define significance as consolidation and perpetuation, verbatim, with the two circumstances and the hearing statement they grew from.
The reverse leveraging and competition on the merits passages, with the Commission’s own words in the decision and the authorities the Court relied on.
Every error the Court found in the Commission’s calculations and file.
A citation table for all 26 rulings named, each matched to its paragraph in the official text with a note on fit.
My pre‑judgment reading of the decision set against what the Court did with each point, the efficiencies section, the posture on appeal, and the method.
The audit
Case T‑1139/23, General Court, Tenth Chamber, extended composition of five judges, judgment of 9 September 2026.
Disposition. The Court rejected all three pleas:
The first plea, that the Commission departed from its Non‑Horizontal Merger Guidelines and misclassified competition on the merits, fails because the Guidelines reserve other theories of harm and because competition on the merits has no comparable role in merger control (paragraphs 70 to 109).
The second plea, on the counterfactual, fails in part on the facts, since the decision did not adopt a “zero‑flights” scenario, and in part as ineffective, since the difference between the parties’ scenarios is negligible once the increment is marginal in any scenario (paragraphs 110 to 145).
The third plea fails after the Court upheld dominance, upheld the importance of a flight offering, struck out most of the Commission’s increment calculations, found a breach of the duty of care, upheld the strengthening of network effects and the harm to hotels and consumers, and held in an overall assessment that the impediment was proved (paragraphs 146 to 473). The efficiencies part fails on admissibility, pass‑on, verifiability and the out‑of‑market rule (paragraphs 474 to 524).
Exhibit 1: The authorities not cited, and what citing them would have changed
Every omission below was confirmed by a full‑text search of the judgment. For each, the last line says what a citation would have changed.
Adverse omissions.
Commission v Tetra Laval, C‑12/03 P, paragraphs 42 to 44. The Court of Justice’s statement of what a conglomerate prohibition must show.
Commission v Tetra Laval, paragraph 43. “Thus, the prospective analysis consists of an examination of how a concentration might alter the factors determining the state of competition on a given market in order to establish whether it would give rise to a serious impediment to effective competition. Such an analysis makes it necessary to envisage various chains of cause and effect with a view to ascertaining which of them are the most likely.”
Commission v Tetra Laval, paragraph 44. “The analysis of a ‘conglomerate-type’ concentration is a prospective analysis in which, first, the consideration of a lengthy period of time in the future and, secondly, the leveraging necessary to give rise to a significant impediment to effective competition mean that the chains of cause and effect are dimly discernible, uncertain and difficult to establish. That being so, the quality of the evidence produced by the Commission in order to establish that it is necessary to adopt a decision declaring the concentration incompatible with the common market is particularly important, since that evidence must support the Commission’s conclusion that, if such a decision were not adopted, the economic development envisaged by it would be plausible.”
Why it matters. The Booking theory of harm is a leveraging chain running from the flight market to the hotel market over a horizon to 2025 and beyond. Paragraph 44 is written for that chain. The judgment never names the case. It reaches paragraph 44 only through paragraph 77 of CK Telecoms, and it states the review standard at paragraph 158 in words that originate in paragraph 39 of the same judgment, omitting the last sentence of paragraph 39, that such a review “is all the more necessary in the case of a prospective analysis required when examining a planned merger with conglomerate effect”.
What a citation would have changed. On the result, probably nothing, since paragraph 77 of CK Telecoms and paragraph 51 of Impala hold that the quality requirement adds no higher standard. The reasoning, one paragraph the judgment does not contain, saying how the quality of the evidence was “particularly important” in a case where the only quantified step failed. Paragraph 439 would have had to be written with that paragraph in view.
Tetra Laval, T‑5/02, paragraphs 155 and 162. The General Court’s own conglomerate standard, in a judgment it cites twice for other paragraphs.
Tetra Laval (T‑5/02), paragraph 155 (extract). “Thus the Commission’s analysis of a merger transaction which is expected to have an anti-competitive conglomerate effect calls for a particularly close examination of the circumstances which are relevant for an assessment of that effect on the conditions of competition in the reference market. As the Court has already held, where the Commission takes the view that a merger should be prohibited because it will create or strengthen a dominant position within a foreseeable period, it is incumbent upon it to produce convincing evidence thereof (Airtours v Commission, paragraph 63).”
Tetra Laval (T‑5/02), paragraph 162 (extract). “In addition, since the anticipated dominant position would only emerge after a certain lapse of time, by 2005 according to the Commission, its analysis of the future position must, whilst allowing for a certain margin of discretion, be particularly plausible.”
Why it matters. Paragraph 41 of Commission v Tetra Laval says that paragraph 155 adds no condition on the standard of proof and merely describes the function of evidence. That is the reading which reconciles Booking with the case law.
What a citation would have changed. The Court uses “plausibility” as its test three times (paragraphs 237, 240 and 449). Paragraph 162 would have required it to say that the analysis must be “particularly plausible” where the dominant position emerges over time, which is the case here, with a horizon to 2025 and 2026. The word “particularly” is the difference.
General Electric, T‑210/01, paragraphs 64 to 66 and 69. The application of Commission v Tetra Laval to a conglomerate prohibition by the General Court.
General Electric, paragraph 69 (opening). “In those circumstances, the Commission had the onus to provide convincing evidence to support its conclusion that the merged entity would probably behave in the way foreseen.”
Why it matters. The Booking chain depends on how Booking would behave after the deal, and on how hotels and rival agencies would respond.
What a citation would have changed. Little in outcome, since the Court found Booking’s intended conduct in its own internal documents (paragraphs 293 to 295). In form, the judgment would have had to call that evidence “convincing”, a word it avoids.
Continental Can, 6/72, paragraph 26, and Towercast, C‑449/21, paragraph 52. Invoked by the applicant at paragraph 455 and not engaged.
Continental Can, paragraph 26 (closing, capitalisation as in the official English text). “ABUSE MAY THEREFORE OCCUR IF AN UNDERTAKING IN A DOMINANT POSITION STRENGTHENS SUCH POSITION IN SUCH A WAY THAT THE DEGREE OF DOMINANCE REACHED SUBSTANTIALLY FETTERS COMPETITION, I . E . THAT ONLY UNDERTAKINGS REMAIN IN THE MARKET WHOSE BEHAVIOUR DEPENDS ON THE DOMINANT ONE .”
Towercast, paragraph 52 (extract). “In that regard, the mere finding that an undertaking’s position has been strengthened is not sufficient for a finding of abuse, since it must be established that the degree of dominance thus reached would substantially impede competition, that is to say, that only undertakings whose behaviour depends on the dominant undertaking would remain in the market” (extract ends before the citation).
Why it matters. Both are Article 102 judgments, and the Court could have answered that the Merger Regulation’s test is its own. It did not say so. It answered with paragraphs 463 and 464, which concede the applicant’s premise, and with paragraphs 467 to 471, which replace “substantially fetters” with “consolidated and perpetuated”.
What a citation would have changed. This is the one omission that could have changed the result. The Continental Can test asks whether, after the acquisition, “only undertakings remain in the market whose behaviour depends on the dominant one”. The judgment’s own findings show Expedia with a share of 10 to 20 per cent on the business side of the market (paragraph 19) and four rivals, Trip.com, Lastminute, TUI and eDreams Odigeo, whose expansion the Court thought important enough to protect (paragraph 439). A court applying Continental Can would have had to explain how a market with those rivals meets the dependence test, or hold that the test does not govern merger control. The judgment does neither, and an appeal will ask which.
Supportive omissions.
Impala, C‑413/06 P, paragraphs 48 and 51. The judgment which says there is no presumption either way and no higher standard for prohibitions.
Impala, paragraph 48. “Contrary to what the appellants submit, it cannot therefore be inferred from the Regulation that there is a general presumption that a notified concentration is compatible with, or incompatible with, the common market.”
Impala, paragraph 51 (extract). “However, it cannot be deduced from that that the Commission must, particularly where it pursues a theory of collective dominance, comply with a higher standard of proof in relation to decisions prohibiting concentrations than in relation to decisions approving them. That case-law merely reflects the essential function of evidence, which is to establish convincingly the merits of an argument or, as in the case of the control of concentrations, to support the conclusions underpinning the Commission’s decisions (see, to that effect, Commission v Tetra Laval, paragraphs 41 and 44).”
Why it matters. Paragraph 48 is the Court of Justice’s answer to the word “presumption” at paragraph 455, and the judgment does not use it. Paragraph 51 is reached indirectly, through paragraph 77 of CK Telecoms, and it is the authority behind paragraph 439.
What a citation would have changed. Paragraph 48 would have disposed of the presumption argument by name, at no cost. Paragraph 51 cuts both ways, since it also says that the “inherent complexity of a theory of competitive harm” counts in assessing plausibility, a sentence the Court of Justice repeated at paragraph 78 of CK Telecoms. Citing it would have strengthened paragraph 439 and obliged the Court to weigh the complexity of a five‑step chain. Neither paragraph 51 nor paragraph 78 is cited.
General Electric, paragraph 65. Conglomerate mergers may have anticompetitive effects, and where dominance would in all likelihood be strengthened in the relatively near future, the Commission must prohibit. What a citation would have changed. Nothing in outcome. It would have given the outcome a conglomerate pedigree the judgment now lacks.
Exhibit 1 above is open to every reader. It is the part of this audit that only a mapped corpus produces, and it is where the appeal lives. What follows is the rest, written to be dropped into a brief.
