Everything in this publication is built to be checked. This page says how, so that a reader who disagrees with a conclusion can find the exact point where we part company.
How the record is made
Every news and opinion item in the record starts as a draft written by an AI system, Claude (Anthropic), from the one source the item cites. The system reads the source and writes the headline, the summary and the boundary line, then proposes the grade and the index reading.
Before an item goes live the pipeline checks it by machine. It fetches the source again and looks for the figures and names the item rests on. A second model reads the item against the source. The record is searched for a twin. The rules on this page are applied: an opinion carries weight 0, a Reported claim counts half. An item that passes goes live; an item that does not is held for the editor.
The editor reads every held item, every Reported item in a month of the archive being deepened, every correction a reader sends, and the monthly newsletter, which is written and edited by hand. The editor does not read every daily item before it goes live: the daily run publishes twice a day without a person reading first.
So each item says which it was. AI-drafted means it was published without a person reading it first. Editor-read means the editor read its substance before it went live, and the item gives the date. The same fact travels with the open data (the ai field on every item), through the connector, and in the page's own head. This is the disclosure that Article 50(4) of the EU AI Act asks of a publisher whose text an AI system generated. The Monitor makes it item by item rather than claiming an exemption it has not earned.
What the AI system does and does not do. It summarises one named source, classifies the item (news or opinion, Europe or abroad, which areas), proposes the grade and proposes the index reading. Each of those is an interpretation, and the checks above and the editor's corrections are where interpretation is tested. It argues no position of its own and draws no conclusion about Europe: every item is a summary of one source, graded, with its boundary stated, and the reader draws the insight. The Monitor collects news and positions; its dataset is open for the reader's own research, and the analysis is the reader's.
Editorial responsibility for everything published here rests with RB Consult AB, Reidar Balstad, editor. Corrections are made by the editor with the original claim quoted. The editor keeps up with what the drafting system can and cannot do, and this section is updated when the process changes.
Four questions
Sovereignty is not one property. It is at least four, and they come apart constantly.
Who owns the thing. Who operates it. Whose law governs it. Who made the parts.
Almost every European instrument tests the first three and stops before the fourth. That is not an oversight to be indignant about; it is a design choice with consequences, and the consequences are what this publication tracks. When a sovereignty claim sounds strong, the useful question is not whether it is true. It is which of the four it answers, and which it leaves alone.
Verified and Reported
Every factual claim carries one of two gradings, and the grading is part of the claim.
Verified means the primary document was opened and read at the passage asserting it. Not the press release about the regulation. Not a summary of the recital. The article, the recital or the annex, at the provision cited.
Reported means the primary text could not be reached, and the claim rests on a named secondary source. Sometimes a document sits behind a wall, sometimes a European database truncates, sometimes a provision lives in a Council working document that is not published as such. When that happens we say so and name what we did read. A Reported claim is weaker than a Verified one and should be treated that way.
Figures and links drawn from CORDIS (the ledger's Horizon grants view, and map partner links labelled "In the same Horizon Europe consortium") are Verified: CORDIS is the Commission's own record of every signed grant agreement, and each figure links to the project page it comes from. They are read in the ledger's vocabulary, contracted rather than disbursed, and they carry their own limits: a grant is not a product, CORDIS sees the grant channel only, and sitting in one consortium is co-membership rather than a partnership. Organisations are matched by their CORDIS identifier, never by name, and the entity list behind the matching is published with the data.
Where a term is not defined anywhere, we say that too, because the absence is usually the point. An obligation to pursue something the same instrument declines to define is a fact about the instrument, not a gap in our reading.
What it does not test
Every entry states the boundary of the claim as well as its content.
This is the column nobody else publishes, and it is close to the reason this publication exists. A cloud estate can be certified at the highest available sovereignty level while running entirely on imported accelerators, because the level tests do not reach the silicon. A headquarters test is satisfied by a subsidiary. An establishment and management test says nothing about who ultimately owns the parent. None of that is a scandal. All of it is invisible if you only read what an instrument does test.
The markers
Each development in the newsletter carries one marker. The markers were the whole scoring system of the newsletter's first issues: an unweighted tally with no editorial index behind it. Since September 2026 the daily record also carries the weighted reading described under The index below; the two are different measures and are never added together.
▲ Towards autonomy. The development moves Europe towards controlling something it did not control before.
▼ Towards dependence. It moves the other way.
◆ Both directions. It genuinely does both, and splitting it into two entries would misrepresent it. A European datacentre running American software is the recurring shape.
■ No movement. Announced, proposed, consulted on, or otherwise not yet real. A great deal of sovereignty news lives here, and marking it honestly is most of the work.
A section's net is its up count minus its down count. Nothing else enters the arithmetic. Both-directions and no-movement are counted and reported but do not move the net, because a development that points two ways and a development that has not happened yet are not the same thing as a zero, and averaging them into one would hide both.
Issue zero is the worked example. Twenty-nine developments: fourteen towards autonomy, six towards dependence, three in both directions, six with no movement. Net plus eight. Underneath that, policy and law ran at minus one while defence and space ran at plus five, which is the finding, and it is a finding the headline number would have buried. That is why every section carries its own count and the total is never quoted alone.
Corrections
Corrections with a primary source attached are applied and credited.
That sentence is the policy in full. Send the document and the provision, and if it says what you say it says, the entry changes and your name goes on the change unless you ask otherwise. Disagreements about interpretation are welcome too, and are treated differently: they may become an added line rather than a correction, because two people can read one recital honestly and differently.
Where an entry is graded Reported, it is provisional by construction. Reaching the primary text is an upgrade, not an embarrassment, and it happens most often because a reader had access we did not.
What this is not
It is not investment advice, and nothing here is a view on any company as an investment.
It is not a vendor evaluation. Suppliers appear when an instrument or a claim makes them relevant, and their own words are quoted as their own words.
It is not a scorecard of countries or firms. The markers score developments, one at a time, against what changed. They do not aggregate into a ranking of anybody, and any attempt to read them that way is reading them wrongly.
The index
Since September 2026 every published news item carries a reading for the European Sovereignty Index, shown on the item under "How this item counts". The reading has five parts: the direction the development pushes Europe's dependence (towards autonomy, towards dependence, both ways, or none); which of the four questions above it changes the answer to; how binding it is (announced, agreed or funded, in force or operating); how big it is (small, medium, large, with money breaks at EUR 100 million and 1 billion, or 500 million and 5 billion in defence); and how sure we are, which is the grade (Verified counts in full, Reported half). The weight is the product of the four, at most 36; a both-ways item is two parts summed with their signs. Two rules of reading: operation alone is not sovereignty when the owner and the governing law stay outside Europe, and Europe is the geography while the EU's own targets are the yardstick.
The pipeline proposes the reading. An item the checks hold gets its reading approved by the editor with the grade; an item that passes the checks goes live with the proposed reading, and its badge says so. The monthly flow per pillar is the sum of the signed weights divided by 36, so one maximal item equals 1.0; a running total accumulates from 1 January 2026, a date that moves back as older items enter the archive. The four pillars are combined with a published key (compute 30, policy 25, defence 25, industry 20) that a reader can change on the index page. The flow does not say how much sovereignty Europe has; that is the level layer, twenty indicators scored against the EU's own targets, which follows. What the flow does not test: the developments not picked; a quiet month scores 0 whatever the state; two equal items in opposite directions cancel; and the weekly picks lean towards European moves, so the flow leans positive.