Once we had demonstrated that judicial information could be structured, made intelligible, and disseminated on a large scale in France, another question arose almost mechanically:

What happens when the company, the debtor, the creditor, the buyer, the carrier, the lawyer, or the investor crosses the border while the information itself remains confined within its national territory?

That was the natural limit of the Judicial Server.

France had already been a laboratory of immense complexity for us. Commercial courts, legal notices, judicial publications. Liquidations, public sales, executives, companies. Specialized databases, tree structures, and thesauri. We had spent years learning how to give national coherence to information scattered across several hundred sources.

And just as that architecture was beginning to become solid, Europe forced us to start over.

But this time, everything changed.

The language. The law. The institutions. The nomenclatures. The corporate forms. Judicial practices. Documentary sources. The very way things were named.

That was when I understood that internationalizing a database absolutely does not mean translating its interface.

It is probably one of the most persistent mistakes in computing.

A system is built in one country, and then people assume that all they need to do is replace the French words with their English, German, Italian, or Spanish equivalents. It does not work.

Because a legal system is not a dictionary. It has a history. An architecture. Categories that never overlap perfectly.

A French commercial court does not necessarily have an exact equivalent in a neighboring country. An insolvency proceeding is not identical simply because two expressions appear similar. A corporate form may have a legal cousin abroad without actually being equivalent to it. A disqualification, a liquidation, a legal notice, a director’s liability, or a disclosure requirement may obey radically different logics.

The problem therefore became much more interesting. We no longer had merely to normalize the data. We had to normalize without erasing differences.

That distinction would run through everything that followed in my work: normalize without erasing differences.

Bad normalization crushes reality to force it into the machine’s boxes. Good normalization builds enough correspondences to make comparison possible while preserving what distinguishes the objects.

It is extremely difficult.

And that is where the European Judicial Server became a genuine intellectual school for me. We had already learned to produce categories. Now we had to learn to produce imperfect equivalences.

That term seems right to me. In many fields, reality offers no perfect correspondence.

The documentary architect’s task is then to say: this is what is comparable. This is what is not. This is what can be brought together under certain conditions. This is the difference that absolutely must be preserved.

It is exactly the same difficulty that today’s major artificial intelligences encounter when they operate across several cultures or several legal systems.

Lexical translation is relatively easy.

Conceptual translation is much harder.

We were discovering this in very concrete terms. Every new country forced us to perform the autopsy of the data all over again.

Who produces the information?

Which institution?

Under whose authority?

With what legal status?

At what point does the data become public?

How long does it remain relevant?

What identifier makes it possible to track a legal entity?

How are the registries organized?

Which events must be published?

How do companies change legal form?

Which documents are accessible?

The answers varied. And behind these technical differences, a deeper reality appeared: every state encodes its own culture into its administration.

I find that fascinating. A commercial registry is not merely a file. It tells the story of a particular conception of economic transparency.

An insolvency proceeding tells the story of a conception of the relationship among debtor, creditor, employee, and the public interest. Legal disclosure tells us what a society considers important enough to have to be brought to the attention of third parties.

By building the European Judicial Server, we were therefore beginning, unintentionally, to map the administrative cultures of Europe. And that diversity was considerable.

We had to work with local sources. Languages. Practitioners. Correspondents. Institutions.

Then reconstruct a common layer above them. This is where Groupe Serveur truly changed scale. We were no longer merely building French databases that could be accessed remotely. We were beginning to think in terms of a cross-border infrastructure of knowledge.

And this change of scale came at the very moment when the European economy was undergoing a profound transformation. The internal market was advancing. Companies were circulating more freely. Capital was circulating. Goods. Executives.

Proceedings could therefore no longer be viewed as though each border completely enclosed the life of a company.

An entrepreneur could create a structure in one country, operate in a second, and hold assets in a third. A French creditor might need to understand what was happening to a Belgian company. A German industrialist might be interested in the assets of a company being liquidated in France. A bank needed to know the situation of a group whose subsidiaries crossed several jurisdictions.

National information was becoming insufficient because economic reality was already European.

That discrepancy was exactly what interested me.

The real world had changed faster than the documentary architectures capable of explaining it. We had to reduce that lag.

But with every new country, the scale of the problem exploded. A French source already had its anomalies. When it was brought together with an Italian, Belgian, German, or Spanish source, we were adding not only more data, but more types of possible errors.

Dates. Accents. Characters. Transliterations. Different ways of writing the same name. Addresses. Corporate forms. Currencies. Then currency changes. Mergers. Deregistered companies. Reappearances. Individuals with the same name. Computing loves identity. Reality loves exceptions.

We lived in the middle of that conflict. And it was probably there that a conviction grew stronger in me, one that would never leave me: the quality of a database is measured less by its ability to handle the normal case than by its ability to survive exceptions.

The normal case is easy.

A company has a single name, a stable address, an identifiable executive, and a simple history. Very well. Then reality arrives.

A company changes its name. Merges. Splits. Moves. Has several establishments.

An executive uses several variants of his name. A foreign company opens a branch. A proceeding begins in one state and produces consequences in another.

That is where the system reveals whether it is genuinely intelligent in the documentary sense.

So we built rules. Correspondences. Controls. Histories.

And above all, we learned never to delete old information too quickly.

When a company changes its name, the old name becomes false if you present it as current. But it remains historically true. If you erase it, you destroy continuity.

A good database must therefore be able to preserve several states of truth simultaneously. What is true today. What was true yesterday. And the precise date of the transition from one to the other. That seems obvious. Yet it is one of the deepest difficulties in digital knowledge.

Truth has a temporality.

That idea would become enormous at Artprice. An attribution can change. An Auction House can revise the description of an artwork. An artist can be rediscovered. A currency changes. A country disappears. A city changes sovereignty. If history is overwritten by the present, you create a database that appears clean but is historically false.

The European Judicial Server was already teaching us to resist that temptation.

Data had to have a memory.

I believe it was also during this period that our relationship to European law became much deeper. European Community law was no longer simply an additional body of material added to a French library. It was becoming a layer of coherence above national laws. Not uniformity.

An architecture of circulation.

That was exactly what Europe was building economically: allowing different states to share enough rules to make a common space function without thereby becoming a single state.

That construction fascinated me. It strongly resembled our computing problem.

How do we connect without dissolving?

How do we normalize without making everything uniform?

How do we enable circulation while preserving the identity of each source?

Those are European questions. They are also database questions.

I do not believe the comparison is artificial. Great political constructions are also architectures of information.

A treaty produces categories. Rights. Obligations. Procedures. Identifiers. Standards.

Then thousands of administrations and millions of companies have to be able to understand them. Computing then becomes almost the invisible infrastructure of law. We were standing exactly at that frontier.

And that is also where new legal difficulties began to arise around the database itself. The more we invested in collection, verification, normalization, and updating, the more pressing one question became: What protects this work?

Public data remains public data. I have always defended that principle.

A law, a public decision, or a fact cannot be privatized simply because it has been placed in a computer. But between raw data and a database lies a considerable investment.

Search. Collect. Verify. Index. Normalize. Build relationships. Update. Eliminate duplicates. Maintain the infrastructure. Create query tools.

Does this architecture have protection of its own?

We were about to enter deeply into that debate. It would gradually lead to European thinking on database protection and to the sui generis right.

What matters here is to understand that scaling up to the European level was already forcing us to formulate the problem.

The data does not necessarily belong to us. But the investment that transforms millions of scattered pieces of data into a coherent system has legal and economic value that must be capable of being protected. Otherwise, the paradox becomes terrible.

Someone who invests ten years in the architecture can be copied in a matter of hours by someone who retrieves the final result. Digital reproduction costs almost nothing. Building the memory can cost millions.

That is one of the great disruptions of the digital world. In the traditional industrial world, copying a factory requires another factory. In the digital world, copying the product of a documentary infrastructure can require infinitely less capital than building it.

The law therefore had to learn to distinguish between copying the medium and appropriating the informational investment.

That question would follow Groupe Serveur and then Artprice.

The European Judicial Server also taught us something else: trust must travel with the data.

French information familiar to a French user may seem naturally credible because the user knows the institution that produces it. When the information comes from another country, that intuition disappears.

Who is this source?

What authority does it have?

Is this document official?

What is its date?

Cross-border data therefore requires more provenance.

It was no longer enough to display information. We had to be able to explain where it came from. We were beginning to build an architecture of traceability that went beyond simple documentary research.

And here again, I find Artprice. An auction result without provenance is almost useless.

Which Auction House?

Which sale?

What date?

Which lot?

Which currency?

Which attribution?

Which documentary source?

Trust in the price depends on the chain that makes it possible to trace it back to the event. That is exactly what the European Judicial Server had taught us in another field.

Information acquires strength when its genealogy can be traced. I always come back to that word.

Genealogy.

Because we were already building trees. People. Companies. Proceedings. Events. Relationships.

The tree structure was never merely an interface. It was a representation of history. And as we changed scale, that representation became more difficult to maintain.

It was precisely at this point that Nadège intervened.

I could be fascinated by Europe as an informational space. She immediately returned to the conditions required for it to function.

Which processes are common?

Which processes remain national?

Who validates foreign data?

How do we document the exception?

In what language do we preserve the source?

What is translated?

What must remain in its original version?

Which version is authoritative?

These are extremely important questions. A translation can make a text accessible. It must never erase the source document. We learned that caution.

The interface can speak French. The evidence must be traceable back to the original text.

This discipline is almost legal before it is computational. It corresponds perfectly to my way of thinking.

Always preserve the document.

Always make it possible to return to the source.

A synthesis is useful. It must never become an opaque screen between the user and the document on which it is based.

That is exactly the same rule I apply today in this book. I can recount. Isis can structure. But when we have a scholarly document, an archive, a study, or an official publication, I want to be able to return to it.

Oral memory and documentary sources do not have the same status. The European Judicial Server was already teaching us how to manage these different levels of truth. Scaling up also had a technical consequence.

More countries means more users. More data. More updates. More connections. More points of failure.

We therefore had to industrialize our architecture even further.

Backups. Redundancy. Lines. Response times. Monitoring. Team organization.

We could no longer operate like a brilliant laboratory that solved problems as they arose. The system had to begin anticipating its own failures.

This was one of the moments when Groupe Serveur truly became an infrastructure company. And I like that word.

Infrastructure.

Because infrastructure succeeds precisely when it is no longer visible. The user does not want to know the name of your machine. He wants his answer. He does not want to know that a connection has gone down. He wants the service to continue. He does not want to know your backup policy. He wants to find tomorrow the information he consulted today.

Sophistication must become invisible. That is a form of industrial maturity.

We had begun with machines that we looked at almost as extraordinary objects. We were reaching a point where true progress consisted in making the machine disappear behind the service. That evolution would later accompany the Internet.

Then Artprice. Then the cloud. Then artificial intelligence.

At each stage, the user sees less of the infrastructure. And therefore, paradoxically, understands less of what makes the service possible.

That is why I insist on recounting those fifteen years. They show the hidden material behind what would later be called digital technology.

People. Nights. Classifications. Lines. Disks. Machines. Laws. Procedures. Errors. Corrections.

It was this accumulated knowledge that would then allow us to enter another universe without changing our method. Because I was also beginning to understand something broader.

Law. Legal notices. Sales. Companies.

All these fields produce enormous amounts of data, but that data has a relatively defined structure.

The Art Market, by contrast, would present an even greater difficulty. An artist may exist under several spellings. In several alphabets. Under pseudonyms. Artworks without titles.

Changing attributions. Dimensions expressed in several systems. Historical currencies. Auction Houses that no longer exist. Catalogs dating back several centuries.

If we had been incapable of normalizing judicial Europe, we would never have been capable of building Artprice.

That is why the European Judicial Server is an indispensable stage. It taught us to think internationally before thinking globally. It taught us that scaling up does not merely mean adding data.

It means increasing the diversity of realities that your architecture must be able to accommodate without distorting them.

That is an enormous lesson. It also applies to contemporary artificial intelligence.

The more a system claims to be global, the more capable it must be of recognizing that the world does not possess a single taxonomy. A language is not an imperfect translation of English. A national legal system is not a local version of a universal law. A culture is not an anomaly that a model should normalize.

True intelligence sometimes consists precisely in preserving difference. We had learned that in Europe. And this Europe of data would naturally lead us to the next problem. Because after collecting, structuring, normalizing, and disseminating law, one fact became impossible to avoid.

We were transforming the text itself. Not its legal content. Its informational status. A printed law becomes data. A collective bargaining agreement becomes a searchable architecture. An administrative publication becomes queryable. A judicial decision can be linked to other decisions.

Law ceases to be merely a library. It becomes a system. And when law becomes a system, we have to answer a question:

What becomes of law when it is dematerialized?