A Country I Cannot See
I tried to read the public register of my own country. I did not expect that this would turn out to be impossible. Public Record, essay two.
I am an artist who has lived in Luxembourg for most of my life. A few months ago I started a project that involves looking at the public list of every company registered in this country and trying to understand what is actually there. I started because I wanted to understand the country I live in.
I did not expect that this would turn out to be impossible.
Last week I published an essay about that project. In it I cited a number, 1500+ companies registered at a single building on Avenue J.F. Kennedy. The number was wrong. The figure included every company that had ever been registered at that address since the register’s records began, not those currently active. Many of the entities in that count had been struck from the register years ago, dissolved, no longer there in any legal sense. You might expect a correction like this to make things look better. The building is less crowded with letterboxes than I implied. The architecture I was complaining about is more modest than my count suggested.
It does not look better. It looks worse.
What I learned through making the correction is what this essay is about. The error was the entry to a finding I did not expect: that the Luxembourg state has, deliberately and across multiple governments, built a piece of infrastructure that prevents its own citizens from understanding their own country. This is not a metaphor. It is what the rest of this essay describes.
The thing the essay needs you to hold from this point forward is small and structural. There are two registers. They contain the same data. One is for paying customers. One is for everyone else. Both are nominally public. Only one is actually useful. Everything else in this essay is either evidence for that statement or its political consequence.
Like most people who read the international press, I have encountered the periodic scandals about Luxembourg, LuxLeaks in 2014, the Panama Papers in 2016, OpenLux in 2021. They arrived. The political class expressed surprise. The news cycle moved on. I never knew what to do with these moments. The stories were always about specific people and specific deals. They never quite told me what was structurally the case in my country. So a few months ago I decided to look myself.
The work itself is small and physical. I have a pen plotter, a sheet of paper, and a list of company names that the register associates with one address. The plotter draws the names, in overlapping ink, on the same single sheet, until the page saturates into a black field. I take the plotter to the address. I plot at the building. The act of drawing happens where the corporate density it documents is materially located. Each sheet is dated, both to the day of plotting and to the day of the data slice it was drawn from. The work operates at the level of corporate-entity names and addresses, the data the register itself was designed to publish. It does not try to identify or expose the human beings behind these companies. The question this essay is asking is publicity, not privacy.
To produce the first sheet I needed a list of names at one address. The LBR’s public interface does not let a citizen ask the register that question. It accepts only individual company-number queries, one at a time, behind a captcha. To get a list of names at an address you have to assemble it from elsewhere.
I assembled it from what is actually publicly downloadable on Luxembourg companies. GLEIF, the global registry of legal-entity identifiers, which Luxembourg-domiciled entities populate when other international regulations require them to. And from RESA, Luxembourg’s official daily journal of corporate filings. I had also asked OpenCorporates, the international registry that aggregates corporate information across jurisdictions, for bulk access. They refused.
From this assembled bulk, the list at 49 Avenue Kennedy had 1,524 names. I plotted those.
What I did not fully take in at the time was what these sources actually were. None of them is up to date. None is complete. None distinguishes companies that are currently active from companies that have been struck from the register. The list of 1,524 names was not a list of currently active companies. It was the cumulative residue of every company that had ever been registered at that address since the records of those external sources began.
This essay is about what happened when I went back to the data to filter for current status.
To know current status, the only route open to a citizen is the LBR’s per-record interface, behind the captcha. I would have to query each of the 1,524 entities individually, then each of Luxembourg’s roughly 169,000 active entities, one record at a time. So I built a small browser script that runs in the background of an ordinary tab, solves the puzzles one by one, and reads the records as they come back. It works at about the speed of a careful human reader, roughly one query every twenty seconds.
The active-or-radiated distinction lives only there. One record at a time, behind the puzzle.
Here is what becomes visible if you sit with that.
By the time you finish querying, the answer has changed. Companies are registered every day. Companies are dissolved every day. The register mutates faster than any citizen can read it through the public interface. A complete current snapshot of Luxembourg’s active companies would take me roughly six months at script speed to produce, working continuously. By the time it was finished, parts of it would already be wrong.
This is what the architecture is. It is not just slow. It is calibrated to a rate at which no citizen, working alone, can ever hold a current picture of the public register. The error in my first essay, citing 1,524 entities at one address without distinguishing active from dissolved, is the kind of error the architecture produces in any citizen who tries to read the register without paying for the architecture’s commercial alternative.
That alternative is the API. The LBR sells programmatic access to professional users at a published price of 5,000 € per year for platform access. With it, the architecture is fast, complete, current. Without it, the architecture is slow, fragmentary, structurally stale. There are two registers. One is for paying customers. One is for everyone else.
The architecture cannot prevent a citizen from looking. It can only make looking slow enough that the looking eventually surfaces what the architecture is.
There is a chronology that was not visible to me when I started, and it changes what the architecture looks like when you see it whole.
Between late 2016 and 2020, journalists at Le Monde and sixteen international media partners scraped the LBR’s public site continuously. They extracted 3.3 million documents totalling 1.3 terabytes of data. The architecture in 2020 had no captcha. There was no anti-automation defence. The data was, as the LBR’s director put it shortly afterward, freely accessible to the public.
This was the investigation that became OpenLux in February 2021. According to OCCRP’s reporting on the 2020 financial data, 55,000 of Luxembourg’s companies had no real economic activity. They collectively managed assets of more than six trillion euros, more than eighty times Luxembourg’s annual GDP at the time.
The Luxembourg government’s response unfolded in two stages. Before the OpenLux investigation published on 8 February 2021, the government had already registered the domain openlux.lu, in order to host an official rebuttal in search results alongside the journalism. Then a year later, on 21 February 2022, Justice Minister Sam Tanson held a press conference to announce a reform of the LBR. Asked specifically about the absence of a feature OpenLux journalists had criticised, the ability to search the register by an individual’s name and see every entity that person is connected to, she replied: “That’s not foreseen.” She continued: “I want to underline that the Luxembourg register is one of the most transparent you can find. I don’t think that you have this functionality in a lot of countries.” The missing analytical capacity was framed as a comparative transparency virtue. The director of the LBR gave an interview to Paperjam in the same period in which he defended the open architecture explicitly: the website was freely accessible, the data was public, the legality of OpenLux’s scraping was not the LBR’s concern.
That was the defensive posture, holding from 2021 through 2022. It did not yet include captchas, paid-API gating, or any architectural reconstruction. The data was open and the openness was defended.
In November 2022 the European Court of Justice ruled on a different register. A man known only as “WM” had asked the LBR to restrict public access to data about him in the Beneficial Ownership Register, because his work required travel to politically unstable countries and creating risk of kidnapping or violence. His privacy concern was legitimate. His case was bundled with another, brought by Sovim SA, a Luxembourg-incorporated company whose lawyers were challenging the public-access regime itself. The Court ruled wider than WM’s case strictly required. It invalidated unrestricted public access to beneficial-ownership data on data-protection grounds. The ruling addressed the Beneficial Ownership Register specifically. It did not address the main Trade and Companies Register, which is a different legal regime.
The architecture’s defenders sometimes invoke this ruling anyway, as if the post-Sovim climate justifies extending opacity into other registers. It does not, and the timing rules it out. The paid API the LBR sells was already operational when the ruling came down. According to i-Hub, a Luxembourg KYC services company that was an early test user, the API was in use by summer 2022, with public availability by October, months before the November ruling. It is a commercial product directed at financial-sector clients with anti-money-laundering and know-your-customer obligations the per-record human-rate interface cannot serve. At 5,000 € HT per year for platform access, it is normal infrastructure for a financial centre. The API would exist if the ruling had never happened. It is not the architectural reconstruction.
The architectural reconstruction came later, and on a different timeline.
EU Regulation 2023/138 entered into force in June 2024. It requires basic company information to be made available free of charge, in machine-readable form, with bulk download and through an API. It mandates the opposite of what an opacity-protecting captcha would deliver.
In August 2025, fourteen months after the regulation’s deadline, the LBR’s web portal was redesigned with a captcha system whose stated purpose, in the launch coverage, was to prevent automated extraction. Large professional users, the same launch coverage said, were invited to use the paid API.
The captcha was not a response to the 2022 ruling. The ruling addressed a different register and was three years in the past. The captcha was deployed fourteen months after a regulation requiring the opposite of what it does. The architectural reconstruction is this captcha, not the API that preceded it. The API was, and is, a normal commercial product in the financial-services landscape. The captcha is what makes the public path slow enough to be unusable for citizens at scale, while the API remains fast for those who pay and qualify. The reconstruction’s effect is to push anyone serious toward the API, not by making the API better, but by making the public path worse.
Same institution. Same director. Opposite stated positions, four years apart. In 2021: the data is freely accessible, scraping is legal. In 2025: the captcha is necessary, professionals are invited to the paid alternative.
The architecture I have been describing is not legacy. The captcha is a 2025 institutional choice, made fourteen months after a 2024 regulation that required the opposite, and defended in 2026 by routing citizen requests away from the regulation toward sector approval.
EU Regulation 2023/138 does not just require a portal listing a dataset. It requires the data itself to be made available free of charge, in machine-readable form, with bulk download and through an API. It does not exempt institutions that fund themselves through fees. Annex 5 names, for the "Companies and Company Ownership" category, two specific things: basic company information (names, registration numbers, addresses, legal forms, status, registration and de-registration dates) and documents and accounts submitted to the company register.
The Luxembourg state filed compliance reporting under this regulation in February 2025. The filing is publicly downloadable on data.public.lu. Under the Companies and Company Ownership category, Luxembourg reports thirteen datasets to the EU. Every one of them is titled, identically except for the year, ”HVD - Annex 5 Companies and Company Ownership - Annual deposits”, 2012 through 2024. They are the financial-statement deposits companies are required to file each year. They are openly published. They have API endpoints. They have bulk download URLs.
The basic company information is not in the filing. Not as a separate dataset. Not as an annex. Not as a footnote. The LBR is not named in the filing as the responsible body for any high-value dataset. The half of the obligation that would require the LBR’s per-record-only register to be opened in bulk is treated, in the state’s own reporting to the EU, as if it were not part of the obligation. The captcha-protected register and the compliance filing coexist on the state’s own infrastructure. The filing reports what the architecture is willing to release. It does not report what the architecture is built to withhold.
Estonia, an EU member state subject to the same regulation, transitioned its business register to fully open data on 1 October 2022 with no economic damage. The technical infrastructure existed. The political will existed. The regulatory framework was the same as Luxembourg’s.
The architecture’s defenders will say that the LBR is not a state agency. It is a groupement d’intérêt économique, a hybrid Luxembourg legal form whose three members are the Luxembourg State, the Chamber of Commerce, and the Chamber of Trades. It is, technically, self-financing: it receives no government money, it funds itself through filing fees, document orders, certificate sales, and the paid API. The director emphasises this often. “Taxpayers are not involved,” he says.
The phrase suggests the relationship is commercial: the citizen who wants the data may pay for it, like any customer. But the institution’s actual response to access requests tells a different story. The 5000 € annual access fee is published in the LBR’s own tarifs document, version 14.0, available on lbr.lu. Anyone willing to read the price list can find it. When I formally requested bulk access, citing EU Regulation 2023/138, the institution’s reply did not quote that figure. It asked which professional sector I belong to. The published price is the floor of access, not the gate. The gate is sector approval. A citizen offering 5,000 € is not in the same position as a fiduciary offering 5,000 €. The second is a normal customer. The first is an applicant with an unusual purpose that the institution has not yet decided whether to recognise.
This transformation has no legal foundation. The publicity obligation exists because the data is public, not because the state pays for the storage, and not because the citizen has presented an acceptable professional reason for asking. EU law is clear on this and the state has filed its own paperwork acknowledging it. But the framing operates rhetorically rather than legally. It produces a public perception in which the citizen who cannot satisfy the institution’s framing of legitimate purpose is not entitled to expect access EU law requires.
I declined to answer in the institution’s frame and restated the citizen ground. The clock is now running on a regulatory non-response.
I am not the first to ask. Luxembourg’s own open-data portal maintains a public tracker of citizen requests for dataset publication. It records seven distinct requests for LBR or RCS data since July 2022, the most recent filed in March 2026, six weeks before mine. Every one is marked as no-response or not-published. The architecture’s silence is not new. It is procedural.
I want to engage the strongest version of the defence, because the weak version is easy to dismiss and the strong version is what a Luxembourg lawyer would actually present. It runs roughly as follows.
The 2022 ECJ ruling left the post-Sovim access regime genuinely ambiguous, and Luxembourg’s response has been conservative rather than instrumental. The professional users who pay for the API have legitimate anti-money-laundering and due-diligence obligations that the per-record human-rate public interface cannot serve. The GIE is a normal Luxembourg legal form, used across multiple sectors, and its self-financing model is not unusual. No formal EU Commission infringement procedure has been opened against Luxembourg under EU 2023/138. The architecture’s continuation is therefore not yet the subject of any external legal finding.
Each of these is true. None of them changes the architectural fact.
The ECJ ambiguity does not require Luxembourg to extend opacity into a register the ruling did not concern. The professional-users argument explains why a paid API exists; it does not explain why the unpaid public path was rebuilt, between 2022 and 2025, to be slower and more opaque than it was before OpenLux. The GIE structure is unremarkable in Luxembourg, but not all GIEs operate state-mandated public registers under EU regulations whose obligations cannot be delegated by reorganisation. And the absence of a Commission infringement procedure means only that no formal enforcement has been initiated; it does not change the regulation’s text, the deadline of June 2024, or the state’s own filing acknowledging what is owed. The architecture is not yet the subject of formal enforcement. It is the subject of a citizen pointing at it.
Some aggregates are published. STATEC, the national statistics office, publishes structural business statistics with one to two year lag, in pre-aggregated formats the user cannot re-cut. The LBR publishes annual figures in its press releases. Commercial providers like Bureau van Dijk sell aggregated and analysed datasets to paying clients. These figures are partial, often delayed, and unverifiable. The methodologies are not published. The underlying data is not released.
The architecture does not produce total opacity. It produces something more sophisticated. It produces mediated visibility: visibility that depends on trusting intermediaries whose interests align with the architecture they purport to describe. STATEC is a state agency. The LBR is a state-managed register. Commercial providers profit from the asymmetric access regime. Each presents figures the citizen cannot verify. The citizen who wants analytical knowledge of their own country is offered, in place of access, the choice of which institutional intermediary to trust.
The leaks that have intermittently exposed Luxembourg’s corporate landscape, LuxLeaks in 2014 and OpenLux in 2021, were necessary because the architecture had foreclosed every legitimate route to public knowledge. The leaks were not the failure. The architecture that made the leaks the only available form of public knowledge was the failure. We had to learn about our own country from documents stolen or scraped at scale, because the country’s own register, the one nominally open to us, was calibrated to prevent us from learning anything we did not already know how to ask.
A government that requires its citizens to depend on leaks to understand their own civic context has substituted exposure for transparency. Exposure is contingent. Transparency is structural. The architecture maintains the first by ensuring the second cannot exist.
I have been describing the architecture as if its purpose were primarily to exclude citizens. The exclusion is a side effect. What the architecture is actually doing is producing a specific commercial product that the Luxembourg state sells.
The product is plausible state ignorance.
A corporate client choosing where to domicile a complex structure does not just want low taxes or favourable legal forms. They want a regulatory environment in which the state can credibly claim, to itself and to others, not to know what is happening in its own jurisdiction at structural level. This credible state ignorance is institutionally valuable. It allows the state to host activity it could not openly endorse.
LuxLeaks, in 2014, made this product visible briefly. The leaked documents showed hundreds of private tax rulings, signed by the Luxembourg tax administration on behalf of multinational corporations, granting bespoke arrangements that reduced effective tax rates to near zero. The administration knew. The administration was an active counterparty to each of those rulings. But because the rulings were private, because the bilateral conversation between the multinational’s tax advisors and the state was structurally invisible to the public, the state could plausibly claim, in any general discourse, not to know the cumulative scale of what it was authorising. That structural invisibility is what corporate clients pay for. LuxLeaks did not show illegality. It showed the state’s ignorance-as-product breaking surface for one news cycle before reasserting itself.
The architecture I have been describing in this essay is the same product, in a different layer. The LBR’s per-record-only access is to citizens what private tax rulings were to the general public: the structural invisibility of the cumulative picture, designed in. Keller Easterling, in Extrastatecraft, calls this infrastructure as political medium: the spatial and technical specifications through which governance happens without ever being legislated. The LBR is exactly that. It is policy in the form of a per-record query rate.
The arrangement is stable because all parties benefit, except the citizenry, whose democratic capacity to understand the country we live in is the cost of the product. We have not been able to see the architecture clearly, so we have not yet recognised that we are paying this cost.
The architecture’s defenders will say that full transparency would damage Luxembourg’s economy. This is overstated. Other jurisdictions that have moved toward greater transparency under international pressure have experienced managed transitions, not collapses. The choice between opacity and transparency is not a choice between economic survival and economic destruction; it is a choice between maintaining a specific economic model and accepting a managed transition to a different one.
The architecture is a deliberate political-economic choice, made by the Luxembourg state, repeatedly, across multiple governments of different ideological compositions, because the commercial value of opacity has been judged to exceed the cost of its democratic violations. The choice has been invisible because the architecture has prevented anyone from naming it as a choice.
I am naming it as a choice now.
A democratically elected government that subordinates its democratic obligations to the production of an opacity-as-service economy is operating against the premise on which its legitimacy rests. The citizenry that elected this government has been excluded, by deliberate architectural design, from the capacity to understand the country it nominally governs through its electoral participation.
The architecture’s most effective defence has been the absence of public attention to its specific properties. The defence is contingent. It can fail. If enough people see clearly what the architecture does, the architecture’s continuation becomes a political choice the state has to defend in public, not a default it can maintain in silence. What was inherited becomes argued. What was natural becomes contested.
The plotter will keep going to the addresses. The procedural archive will keep recording the institution's responses. Whether the institution changes or does not, that record continues to be made. The work was made inside this architecture. It is being read inside this architecture still.
The register is a fact. The country is a fact. The architecture between them is a choice. Every day that goes by without the choice being named is another day the choice is permitted to operate without being defended. Once it is named, it has to be defended. Defending it requires saying out loud what has, until now, gone without saying: that the Luxembourg state has decided that the commercial value of preventing its citizens from understanding their own country exceeds the democratic cost of doing so.
I would like the state to be required to say this out loud.
This essay is my small contribution toward that requirement. The rest of the contribution is on the paper.
