Methodology
Kyber EU assesses proposals for EU legislation under the ordinary legislative procedure, once, at the moment the European Commission publishes them. This page states the model behind the assessments, the value choice it makes, how the verdicts are produced and checked, and where the method is weak.
The model
The assessments apply the Polish School of Cybernetics, the social cybernetics of Marian Mazur and Józef Kossecki. It describes societies and states as autonomous systems: systems that steer themselves, keep their own internal balance (homeostasis) and spend energy on two things, keeping themselves working (idle power, Pj) and acting on their environment (disposable power, Pd), with P = Pj + Pd.
Applied to the EU: each member state is a primary autonomous system, and the EU is the members’ voluntary coupling. Every bill renegotiates an exchange. Members give up part of their own steering. In return they may get lower friction between them (one rulebook instead of 27, less cross-border cost) or joint capacity they could not build alone (pooled resources, shared infrastructure, joint bargaining weight). The central question for each bill is: is what the members get back worth the steering they give up?
The value choice, stated openly
Putting the member states first is a choice, not a neutral fact. It means that a transfer of competence to the EU always counts as a cost that has to be justified by what it gives back, and that every member state counts equally: Malta weighs as much as Germany in defining what a “typical member” is. Neither more nor less EU is presumed good. EU membership itself is not treated as a loss of autonomy; each bill is judged on its own exchange.
One consequence readers should know: EU minimum standards that each member could set for itself (for example in worker protection) tend to come out negative under this model, unless one member’s weaker standard harms the others. The model asks whether the coupling adds something; it does not ask whether the standard is desirable in itself.
How a verdict is produced
Sources. The proposal and the Commission’s impact assessment come from the EU Publications Office (CELLAR); the legal basis and national-parliament documents from the European Parliament’s Legislative Observatory; country indicators from Eurostat. The impact assessment is the Commission arguing for its own proposal, so its figures are treated as claims to weigh, not as facts.
EU level. A language model reads the proposal and the impact assessment summary and writes the analysis. It does not choose the key values freely: it answers fixed yes/no questions tied to provisions of the bill (does it create capacity members could not build alone? does it replace divergent national rules in a cross-border activity? does it give the Commission new decision powers? does it bind national measures?), and code derives the coupling benefit and the competence transfer from the answers. Code also rejects verdicts that contradict them, for example a positive verdict where competence moves up and little comes back. The analysis runs at least twice; if the runs disagree, a third run decides by majority, and without a majority nothing is published.
Member states. Differences between countries come from data and code, not from the model’s knowledge. The EU analysis names what makes the bill’s effect differ between countries (for example, the size of a regulated industry); code compares each country with the simple average of the members on the matching Eurostat indicators. Then the model extracts statements about each country from the Commission’s impact assessment, and code keeps only those whose quotation appears word for word in the source. Each country’s note must cite these inputs, may not move against what the data shows, and may use only numbers from the inputs.
Without evidence of a difference, a country takes the EU verdict. The report marks those countries, because taking the EU verdict is not a finding about them. Countries that do not take part in an act (for example Denmark in parts of justice and home affairs) are shown as not applicable.
Divergence measures how far the researched countries’ verdicts are from the EU verdict (same, one step, opposite), averaged with every country counting equally. It is shown only when at least nine countries have country-specific evidence. The Council signal is hypothetical: it says whether the countries assessed as negative would form a blocking minority if governments voted according to these assessments. It is not a forecast of how governments will vote.
National parliaments. Reasoned opinions on subsidiarity are shown as political signals of a parliamentary majority. They do not change any score, and silence is not read as approval.
Terms used in the reports
- EU-level verdict positive · neutral · negative
- Whether, for the member states together, what they get back from the bill outweighs the steering they give up. Neutral means the two are roughly balanced for a reason the analysis states, not that the bill does little.
- Risk to the coupling 1–10
- How much the bill threatens the stability of the members’ voluntary coupling: rules that steer more than members would freely accept, likely non-compliance, burdens that fall very unevenly, or a loss of legitimacy. 1 is no risk, 10 severe.
- Coupling benefit High · Medium · Low · None
- What members gain together that they could not gain alone. It is derived in code from fixed questions: High when the bill creates capacity members could not build individually (pooled resources, a joint body, shared infrastructure, joint bargaining), or replaces divergent national rules in a cross-border activity while also solving a cross-border problem; Medium when it does one of the latter two; Low when it does so only partly; None otherwise. A bill whose gains are mainly procedural (clarifying or restating existing law) is at most Low. A gain each member could achieve alone counts only where one member’s shortfall harms the others.
- Competence transfer Increase · Neutral · Decrease
- The steering members give up. Increase when the bill gives the Commission or an EU agency new powers to decide later (delegated or implementing acts, authorisations, operational decisions), or binds national measures where members decided freely before. Neutral when it only updates values or lists inside an existing EU framework, by that framework’s own procedure. Decrease when it returns decisions to the members. It is always a cost that the benefit has to justify.
- Subsidiarity fit Good · Questionable · Poor
- Whether the matter is decided at the right level. Steering far from where problems arise loses information, and one rule for very different situations risks over-steering, unless uniformity is itself the point (for example, one product standard for one market). A directive leaves members more room than a regulation.
- Heterogeneity High · Medium · Low
- How different the members’ situations are in the area the bill regulates. High heterogeneity makes a uniform rule harder to fit.
- Friction and joint capacity
- The two quantities behind every verdict. Friction is what members spend to keep the coupling working: compliance cost, administrative burden, legal uncertainty. A single rulebook can lower it; new obligations raise it. Joint capacity is what members can achieve together. Moving powers to the EU institutions does not by itself add joint capacity.
- Divergence among member states High · Medium · Low · not enough evidence
- How far the researched countries’ verdicts are from the EU verdict: 0 for the same verdict, 1 for one step, 2 for the opposite, averaged with every country counting equally. High from 0.7, Medium from 0.4, otherwise Low. Countries that take the EU verdict for lack of evidence, or do not take part in the act, are left out; with fewer than nine researched countries no level is given.
- Council signal
- A hypothetical: if governments voted according to these assessments, would the countries assessed as negative form a blocking minority in the Council (at least four members with more than 35% of the participating members’ population)? It is not a forecast of how governments will vote.
- Country verdict
- The exchange seen from one member state: whether it gives up more, or gets back more, than a typical member. It may move away from the EU verdict only for a reason in that country’s data or evidence.
- Exposure far less … far more exposed than a typical member
- How strongly the features that make the bill’s effect differ (for example, the size of a regulated industry) apply to a country, compared with the simple average of the members on Eurostat indicators. It is relative: “about as exposed as a typical member” does not mean the bill is good or bad for that country.
- Takes the EU verdict
- No country-specific evidence was found, so the country is assumed to be affected like the members as a whole. Shown faded; it is not a finding about that country.
- Not applicable
- The country does not take part in the act (for example under an opt-out stated in the proposal).
- No stable result
- The repeated assessments of this country disagreed, so no score is published.
Limits
The assessments are produced by language models under fixed rules and checks in code; they can still be wrong, and the same bill assessed twice can land differently, which is why majorities are required. The country indicators cover only some policy areas so far, so on many bills most countries take the EU verdict for lack of evidence. The impact assessment is the main source of country evidence, and the Commission chooses what it covers. Bills are assessed once, at the Commission proposal; changes made by the Parliament and the Council are not re-assessed. The rules are reviewed against published results and changed openly.