We make the law readable for computers.
So you have time for what only people can do.
What we're building
We work from one official source: the Basiswettenbestand (BWB). That's the official register of Dutch laws, maintained by the government. The same source used by lawyers and the Dutch Tax Authority. No blogs, no forums, no training data from the internet.
We don't store the law as one long text, but as structured data. We keep every version as it stood on a specific day. Each article, each paragraph and each clause stays separately addressable, even after the law later changes.
On top of that structure, we add a layer of enrichments: amounts, percentages, references between articles, formulas. Each one is linked to the exact spot where it appears in the source text.
What this enables
We make tax law understandable for consumers and actionable for professionals.
Understandable for consumers
Tax questions answered in plain language, with a traceable source down to the paragraph level. Clickable links to wetten.overheid.nl let you verify every claim yourself.
Actionable for professionals
Accountants, tax professionals and accounting-software companies get the law as data: searchable down to the paragraph level, with references between articles as a network. As a tax expert, you validate a short list of recurring legal phrases once. The system then recognises every next occurrence automatically.
This work doesn't stand alone. The Netherlands has had years of initiatives to make regulation computer-executable, including ALEF and RegelSpraak. The broader Rules as Code field (the idea of writing laws and rules as code so computers can execute them) brings more parties in Europe together on the same question. We follow their research and know their work. We develop our own approach independently.
How this is different
The law as a network, not flat text
For most systems, a law is one long text. For us, it's a network: each article and each paragraph is a node with its own officially unique number. That number is a JCI (Juriconnect Identifier: a unique number for each piece of law, like an ISBN for books). References don't run along the text, but along lines between nodes.
Executing rules, not predicting them
AI guesses each next word based on probability. We don't work that way. We build an execution layer that follows rules in the order the law itself prescribes. We apply conditions, amounts and exceptions in the same order as in the law. The same question gives the same answer.
Source first, answer second
Every number and every conclusion is linked to a law article with a date. If the law changes later, the link still points to the version it was about. No summary without a source.
Alongside, not against
Interpreting, advising, deciding on a specific situation: that's work that requires human judgement. We're not building anything to replace it. We build a layer underneath: the law as structured data, available for anyone who works with it. An addition to the existing landscape, not a replacement.
We don't yet cover all Dutch tax laws, and the execution layer is still in design.
Working on something this infrastructure could fit?
We'd like to talk to accountants, tax professionals, accounting-software companies and anyone working with the law in their day-to-day.