Its title. Which ministry is sponsoring it. What kind of instrument it is. When it transposes one, which EU directive. Which Luxembourg laws its title says it amends. And which European regulations its title cites. That is enough to be useful, and it is nowhere near enough — which is why the rest of this page is honest about where the line is.
Bodane reads Luxembourg’s legislative register directly and keeps its own copy: every draft bill and draft grand-ducal regulation, every event recorded against them, and the manifest of documents filed with them.
Counted from the ingested database in September 2026. The corpus grows; these are readings, not marketing figures.
Not a saved search. A topic is a small standing description of an area you are responsible for, and you can have as many as your practice has areas — there is no per-topic charge, because charging per topic would teach you to watch less.
A topic definition never leaves your deployment. It is not sent anywhere for analysis, it is not pooled with other firms’ topics, and it is not used to improve anything outside your own account. If you ask Bodane to propose your topics from a few sentences describing your practice, those sentences are sent out to be classified — the data page says exactly how — but the topics you then keep are not.
The terms of art you would search for if you had the time — bénéficiaires effectifs, financement du terrorisme, dépositaire — in the language the register publishes in.
Finance, Justice, and whoever else legislates into your world.
Named by their European identifier, so a Luxembourg bill transposing one is recognised as such without depending on the wording of its title.
Named as laws, not as words — the loi du 12 novembre 2004, the Code du travail. Amendment titles cite the law they change, so a subscription to the law catches them exactly. 1,658 distinct acts are named across the register; the most-amended is the Code du travail, at 208 bills, and the loi du 4 décembre 1967 on income tax is close behind at 203.
Kept separate from directives, because they behave differently: a regulation is never transposed, so there is no link to follow and it can only be recognised where a bill cites it by number. 161 bills do, across 194 regulations. For most regulations the picker can therefore only offer one Luxembourg has already legislated about at least once. Six are offered regardless, because for them the first bill is the one that matters: AMLR, AMLA, the transfer-of-funds regulation, DORA, MiCA and the GDPR.
A grand-ducal regulation and a private member’s bill do not carry the same risk, and a topic can say so.
The failure mode of every alerting product is the same: it cries wolf until you stop reading it. Bodane’s answer is not a cleverer score but a visible split — an exact identifier or two agreeing signals in one list, one signal in another, and a coarse filter kept behind a fold where it cannot dilute either.
One of four things is true. The bill transposes a directive the topic names; its title says it amends a law the topic follows; or its title cites a European regulation the topic follows — each a fact about the bill, not a guess about its wording. Or, fourth, two independent signals agree: a keyword from the topic appears in the title and the sponsoring ministry is one the topic watches.
One exception keeps the first kind honest. A law is named by its date, and Luxembourg sometimes passes several unrelated laws on one day; when the date a title cites is one of those, the bill goes to review with that reason given, rather than being matched on a citation that could mean another law.
This is a short list on purpose. On a test topic covering anti-money-laundering, four bills matched over twelve months.
One signal only: the keyword is in the title, but no ministry you watch is behind it. These are ordered by how rare the keyword is across all 7,753 titles, so a term that appears in thirty titles outranks one that appears in three hundred. A term in more than one title in twenty is treated as noise and not listed at all. A private member’s bill ranks lower than a government one, because most of them are withdrawn.
Each row takes a verdict — relevant, or not. The verdicts are kept, and they become the labelled set against which the next tier of matching is measured. It is the one part of the product that asks something of you, and it is the part that makes it yours.
Everything from a ministry you watch with no keyword agreement at all — 1,275 bills on the test topic above, with no time limit, against four matches in twelve months. Collapsed, and labelled in the product itself as a filter rather than evidence. It is there for a deliberate trawl: nothing in it alerts you unless you open it and mark a bill relevant.
The exposé des motifs says why a bill exists. The fiche d’évaluation d’impact says who carries the cost. The commentaire des articles says, article by article, what the law becomes. Matching a topic against what a bill does, rather than what it is called, is the whole reason Bodane exists.
380 bills have now been read end to end — 15% of the 2,528 that carry readable text. Every one of the 3,540 subject claims kept was located verbatim in our own stored copy of the document; a quote we cannot find again is discarded rather than shown.
Nothing described as working above depends on it, and a match made from a reading is shown as something to review rather than as a settled answer, because those matches have not yet been calibrated against a practitioner’s verdicts.
Following a law tells you the law is touched. A template does not rest on a law; it rests on l’article 5, paragraphe 2 of it, and a bill that rewrites article 12 is no business of a clause citing article 5. So Bodane reads citations at article level — in the text of bills, and in documents you give it.
You give Bodane a template clause, an engagement letter, a standard form. It reads out the provisions the document rests on — one entry per provision, however often the document cites it — and shows you each one with the sentence it was read from. You confirm the list. Until you have, nothing about that document alerts: a mis-read citation nobody checked would become an alert nobody could explain.
The text of the document is not stored. The provisions and their evidence sentences are, with the title you give the document and a fingerprint that recognises the same file twice, and none of it can be turned back into the document. Nothing from it is sent to the provider that reads bills.
Word files in .docx format and plain text are read. The older .doc format is refused, and so is PDF, rather than attempted: a scanned PDF yields no text, and no text looks exactly like a document that cites no articles, so a format whose failure cannot be told from an ordinary empty answer is declined instead of guessed at.
When a bill’s text has been extracted, every provision it names is keyed and compared with your list. A hit comes back at one of three tiers: the bill amends the article — a drafting formula follows the citation and the citation is the subject of its sentence; the bill mentions the article; or the bill’s title names the act and its text does not say which of the act’s articles it amends. The tiers rank the alert. None of them suppresses it.
What it finds reaches you rather than waiting to be found: a bill that reaches an article one of your documents rests on is listed in the monthly briefing and in the digest, and it is listed there even when none of your topics caught the bill at all. A bill that names only the act stays on the clause alerts page, folded under the others, rather than becoming an item in the briefing.
The key is the article, deliberately not the paragraph: a bill that rewrites article 5 entire must reach a clause resting on article 5, paragraph 2. And a clause resting on articles 32 à 36 hears about a bill touching any one of the five.
For an amended article, the version in force before the bill and the version it produced, side by side, with the difference marked at word level. The State publishes each consolidated version with the date it began to apply, and the pair is chosen on that date: opened from an alert, it is the pair either side of the day that bill came into force. A bill not yet in force — or in force after the newest version published — has changed nothing in the text held, and the page says so and shows the article as it reads today, rather than somebody else’s amendment.
Measured before it was built, on the legislative corpus itself: 13,007 article-level citations bound to a dated act across 4,236 documents; 4,067 distinct act–article pairs over 805 acts; 21% carrying a paragraph or alinéa; 99.6% resolving to an act Bodane holds. Enumerations — les articles 5 et 6 de la loi du… — add about 11% more bindings and were missed by the first version. Recounted in October 2026, once the reading also followed de la même loi, bare articles listed under la loi … est modifiée comme suit and hyphenated numbers such as article 710-6: 20,712 citations across 4,260 documents, 7,923 distinct act–article pairs, and 4,438 amendments — about one citation in five, against one in twenty-two before. What is not measured is a firm’s own citation style, and that needs one firm’s documents.
A keyword filter over titles is the first thing anyone tries. It fails for reasons that are specific and measurable, not vague.
46% of titles are pure amendment — they name the law being changed and nothing else. The subject is whatever that older law was about, which the title does not say. This is the one on the list Bodane has since answered: the citation is useless as a description and exact as an identifier, so the law itself became something you can subscribe to. The rest of this list still stands.
The register files bills under a subject vocabulary at enactment, and 6,417 published acts carry codes. Draft legislation carries none: of the 672 bills still moving through consultation, parliament or a pending vote, not one has a code. There is no category to subscribe to while there is still time to act on it.
Thirty-two bills in the whole corpus have blanchiment in the title. Anti-money-laundering obligations move far more often than thirty-two times.
The median is 196 characters. The tail is worse.
Titles on this page reproduced from Légilux, Service central de législation, Ministère d’État — licensed CC BY 4.0.
A monitoring tool you have to remember to open is a monitoring tool you stop opening. What Bodane finds goes where the work already happens — an inbox, a calendar, a draft to a client — and the application is where you go when you want to look closer.
What moved on your topics since the last one, grouped by how far each bill has got — because a bill out for opinion and a bill already in force need different things from you. Plain text, no tracking pixel, and it links back into the product rather than restating it.
Two events do not keep until the digest. Those go out on their own the next morning, capped at eight bills — because an alert that arrives every week is a digest with a worse name.
Transposition deadlines the Union publishes, commencement dates the Grand Duchy publishes, and dates written into the body of a bill — one forward list, and a private address you paste into Outlook, Google or Apple Calendar. With the e-mailed digest, it is how the product reaches a partner who never signs in.
It is not called a list of deadlines, and that is deliberate. The corpus contains au plus tard le 10 juillet 2027, a real compliance deadline, beside trois arbres après le 1er janvier 2030. Every row shows the sentence its date came from and names the source: a date read out of a bill’s text is labelled as a machine reading, and only what the Union and the Grand Duchy publish as official is presented that way.
Knowing about a bill is half the job. Telling forty clients about it is the other half. For any bill that has been read, Bodane assembles a draft note — what the bill does, who it lands on, which dates matter — that you edit, put your own name on, and send.
Assembled, not written fresh: every sentence of substance comes from the machine reading of that bill’s own documents and is stored with the quote it came from, so drafting a note introduces no claim that the reading did not already make — and the note says to check it against the published text before it is sent. It is in French because the client reads it; the product itself can be used in French or in English. It is marked a draft on the page and in the file, and it is not advice.
Partners do not think in topics. They ask each other what is happening to the 1967 law. So each act has a page: the bills in the corpus that touch it, the most recently moved first, with what each one did and where it got to — and a button to follow the act from there. Where there are more than sixty, the page lists the sixty most recent and says how many there are in all: 203 on the income tax law alone.
Every decision reached, the rejections included, with its date and the reason it gave, in a record that cannot be edited or backdated afterwards. Downloadable as a spreadsheet.
Bodane tells you a bill is moving and why it thinks it concerns you. Whether it does, and what you should do about it, is your judgement and your professional responsibility.
Matching on titles and ministries misses bills. That is the stated reason for the review queue, and for reading the documents — which has run on 15% of the bills that carry readable text and keeps going. Any product in this space that claims to catch everything is describing an ambition.
Nothing is auto-dismissed on your behalf. The unconfident list is shown to you as unconfident rather than filtered away, which is more work and the only defensible design.
When a bill amends an article your template rests on, Bodane tells you which article and which template. Whether the clause still does what you meant it to is drafting, and drafting is yours.
Luxembourg legislation, in the register’s own language. EU directives appear where a Luxembourg bill transposes one. It is not a European legislative tracker and does not pretend to be.
Not a prepared example — your topics, against the real Luxembourg register, including what it misses. The reading tier gets sharper as practitioners mark what comes back, so what you tell us is wrong is worth as much to us as what you tell us is right. The demo shows a fixed snapshot of that register, taken when it was last updated. It fetches no new bills and sends no digests or alerts; a topic you define there is matched against the snapshot as you save it, and a document you add to the clause library is read, but raises no alerts until the register moves — which on the demo it does not.