The first review of the Digital Markets Act (DMA) reads less like an evaluation than a wellness check performed by the patient’s proud parent. The pulse is strong. The color is good. Any lingering symptoms? Too early to tell.
On April 28, 2026, the European Commission published the review required under Article 53 and declared the DMA “fit for purpose.” The report credits the law with “a tangible positive impact” and sees no need to revise the list of core platform services. In the Commission’s telling, the machinery is working; where the evidence remains thin, time will supposedly fill the gaps.
I do not doubt the Commission’s sincerity. I doubt that the exercise could ever have produced a different result.
The DMA’s evaluative architecture is designed in a way that makes failure effectively impossible to demonstrate. Not because the Commission is uniquely stubborn or populated by self-serving sycophants. Rather, the regime’s foundational premises determine what counts as evidence in the first place. In a new white paper, I describe this phenomenon as autopoietic regulation: a system that reproduces its own assumptions rather than testing them. The DMA’s first review confirms that diagnosis precisely because it passes.
The Review That Couldn’t Fail
Start with a simple question: What finding would have led the Commission to conclude that the DMA, or some core part of it, was a mistake?
Not “needs more time.” Not “compliance is incomplete.” A genuinely negative verdict: that an intervention’s costs exceeded its benefits; that a prohibition targeted conduct that was making consumers better off; or that the regulation’s underlying premise was wrong. Any review worthy of the name should be able to answer that question. An evaluation that can return only “working” or “working, but not yet finished” is not really an evaluation at all.
The DMA’s review cannot return such a verdict, and I argue that the reason is structural. The regulation’s recitals present its obligations as simultaneously good for end users, business users, and innovation. In the DMA’s self-conception, there are no tradeoffs (except gatekeepers’ losses, which, as I have argued, are the point).
No recital seriously entertains the possibility that mandated interoperability could reduce quality, that a prescribed fee structure could distort investment incentives, or that redistributing rents from platforms to business users could leave consumers—or other business users—worse off. If the statute does not acknowledge such costs, the review designed to assess the statute has no category in which to record them.
So watch what happens when those costs show up anyway.
When Bad News Doesn’t Count
The review did not lack for adverse evidence. The accompanying staff working document notes that advertisers, despite gaining new data-access rights, still face “persistent issues in obtaining sufficiently granular and comparable data.” It reports that alternative app stores and payment methods have produced “mixed results,” hampered by technical and contractual barriers. Gatekeepers and others also warned that interoperability mandates carry high compliance costs and “can harm customers, stifle innovation, and raise security and privacy concerns.”
In an ordinary oversight regime, observations like these would mark the beginning of an inquiry. Has the intervention made its intended beneficiaries better off? Are the benefits worth the costs? Under the DMA, they function largely as consultation inputs: recorded, acknowledged, and then set aside.
The recurring move is to treat disappointing results not as evidence that an obligation may be misconceived, but as evidence that it has not yet had time to work. App-distribution rules underperforming? Too soon to judge. Messaging interoperability barely adopted? Too soon to draw conclusions. The asymmetry is striking: success counts immediately, while failure is always provisional.
The clearest illustration comes from the one company that does not fit the DMA’s template. Amazon—the only retailer designated under a framework built largely for search engines, app stores, and operating systems—used the review’s publication to argue publicly that the assessment “fails to address the fundamental design flaws that make the DMA unfit for retail,” and that compliance has “stalled investment, damaged the customer experience, and not made services more affordable.”
One need not take Amazon’s word for any of this. The point is what the review does with such claims. It records them as stakeholder grievances. The possibility that a 40%-50% decline in a retailer’s organic traffic might reflect a defect in the regulation, rather than transitional friction on the way to a better equilibrium, lies outside the frame the Commission brought to the exercise.
A Machine That Eats the Evidence
Here is where the autopoiesis point—big word, I know—connects to the architecture of the Digital Markets Act (DMA), and to two ideas I develop at length in the paper: the structural-unfairness axiom and the conflation of public and private gain.
The DMA does not ask whether a gatekeeper’s conduct harms competition to the detriment of consumers. It begins from the axiom that designated markets are structurally unfair—that gatekeepers, by virtue of their position, “obtain a disproportionate advantage.” Unfairness is not something the Commission must demonstrate case by case. It is a premise the regime carries in.
That kind of premise is self-sealing. Whenever an asymmetry between gatekeepers and business users persists—whenever rivals have not gained share, or users have not migrated—the axiom supplies a ready explanation: The unfairness has not yet been corrected. There is no result the premise cannot absorb, because the premise was never stated in a form that data could contradict. As I argued in the companion post on consumer choice, stubborn loyalty to a gatekeeper’s product is read not as a verdict to respect, but as a wall yet to be torn down.
The second idea explains why the review hears the costs it does and misses the ones it does not. The DMA equates business-user grievances with the public interest. Its objectives are framed around benefits to business users—overcoming barriers to entry and redressing the “imbalance” between gatekeepers and their business users—and its enforcement metrics ask whether those imbalances persist.
That is a structural choice. It erases a distinction competition law has worked hard to preserve: the difference between harm to competitors and harm to competition. Once business-user satisfaction becomes a proxy for the public interest, the review will systematically overweight the voices of organized complainants who want compliance never to be declared complete. It will underweight the millions of consumers who benefit from integrated products and have no standing committee at the Commission’s compliance workshops. The feedback the regime is wired to receive is precisely the feedback that confirms it.
Put the two together, and the result is a regulation impervious to disconfirmation. Unfairness is structural, so it is always present. After all, when will rights and obligations be in perfect balance? Exactly: never. Gatekeepers are always the beneficiaries of a lopsided distribution, so their losses never count as costs. And because unfairness is structural and axiomatic, there is no counterfactual against which to judge whether the distribution of rights and obligations is “fair.” Compared to what?
Given the conflation of public and private interests, the public interest is likely to become whatever organized business users say it is. Given the assumption that any obligation imposed by the DMA is achievable, any failure to reach that benchmark will be read as either an infringement or malicious compliance.
A regime constituted on those premises is not self-evidently “working” just because its review says so, any more than a bus headed for a cliff is working because it is still moving. As I have repeatedly stated in tweets I now cannot unearth: Movement is not progress.
The One-Way Ratchet
None of this would be alarming if the DMA were cheap to get wrong. It is not. The conduct it prohibits per se is precisely the category for which the error-cost literature counsels caution, because the welfare effects are context-dependent and false positives are difficult to reverse. A wrongly acquitted monopolist may still be disciplined by entry. A wrongly condemned product design is abandoned, and the foregone innovation is never observed.
A regime that cannot perceive its own false positives will keep producing them. Worse, it will mistake the absence of complaint from the people it has overridden for evidence of success.
The Commission describes the DMA as an “iterative process,” a reassuring phrase that implies learning. But iteration only produces learning if the loop is closed—if some outcome can send the signal: stop, reverse, this was wrong.
The DMA’s loop runs in one direction. Each iteration adds obligations, monitors compliance, and treats shortfalls as grounds for more intervention. The first review was supposed to be the moment when the architecture faced an external test. Instead, the enforcer reviewed its own enforcement, declared the results positive, and concluded that what the regime needs is not greater scrutiny, but more vigorous application of the same approach—with cloud computing and artificial intelligence next in line.
That is the real significance of the review. An instrument confident enough to expand before demonstrating that it can succeed is not being disciplined by evidence. It is being driven by an axiom.
And the question the DMA’s evaluative architecture is structurally incapable of asking—whether the costs of intervention exceed its benefits—remains unanswered because it cannot be asked. The possibility that the premise itself is wrong has been designed out of the system.
A review that cannot find error is not evidence that the policy works. It is evidence that the policy has stopped looking.
