Release 0.9.3: what changed and why

In brief

• Each policy indicator is now the mean of two to five criteria, 33 per country, each answered against a written level description and backed by a public source.
• Before release, 679 of the 1,260 answers outside the law indicator were re-checked against public sources, and 195 of them changed. Author rulings set 16 answers that the codebook left open.
• The mean policy score moves from 56.7 to 53.2. Composite scores change by 3.8 points on average; 23 of the 45 countries stay within three ranks of their 0.9.2 position, and the largest move is 11 places.
• All indicators outside the eight policy indicators have the same scores as in 0.9.2, checked indicator by indicator, so every change in the composite scores and ranks comes from the policy coding.

Why the policy coding changed

In version 0.9.2 each policy indicator was a single code with two to four levels. One code had to summarise several distinct facts — whether a strategy exists, whether its measures have responsible bodies and a budget, whether anyone monitors it — so countries with quite different frameworks received the same code, a single judgement could move a country by a third or a half of the scale, and a reader could not see which fact had decided the score. For the EU member states the codes had also started from the 2025 country reports for the review of the EU Social Economy Action Plan, which are not public.

Version 0.9.3 splits each indicator into criteria. Each criterion is one observable fact, answered at one of four levels (0, 33, 67 or 100) against a written description; the law indicator records five legal facts, the adoption stage on four levels and the other four at 0, 50 or 100. The indicator is the mean of its criteria. This does three things. Scores become finer: across the 45 countries the eight indicators now take 83 distinct values between them, against 24 in 0.9.2. Each answer can be checked on its own, because it states the fact, the level description applied and its source. And a difference between two countries can be traced to a named criterion rather than to an overall impression.

Four levels per criterion is a deliberate limit. Each extra level needs a description that two coders can tell apart; finer distinctions come from adding criteria, not levels. The same logic — scoring a set of specific questions against written guidance and aggregating them — underlies other established indices, for instance Freedom House’s Freedom in the World, which scores each indicator from 0 to 4 against guiding questions, and the World Bank’s Women, Business and the Law, which answers yes-or-no questions on the law and aggregates them into indicator scores.

What each policy indicator measures now

• SE strategy / action plan (1.2.1). Version 0.9.2 placed each country on one of four levels by the kind of strategy it had: none, a draft, an SE chapter in a broader strategy or a regional strategy, or a dedicated national strategy. Version 0.9.3 keeps status and dedication as two criteria and adds the strategy’s operational content, its monitoring and the part SE organisations played in drafting or monitoring it. Lowest-scoring criteria across the 45 countries: participation of SE organisations (mean 26) and operational content (mean 36).
• Institutional structure & dialogue (1.2.2). Version 0.9.2 gave the top level to a dedicated structure with structured dialogue, and the middle level to a responsible but fragmented ministry. Version 0.9.3 asks separately about the lead, the breadth of its mandate, a consultative body with SE organisations and inter-ministerial coordination, and credits a body in full only where a dated record shows that it met in 2024–2025. Lowest-scoring criteria across the 45 countries: inter-ministerial coordination (mean 27) and consultative body with SE organisations (mean 36).
• Social clauses in public procurement (2.3.2). Version 0.9.2 combined the legal basis for social clauses, the national push and the evidence of use in one three-level code. Version 0.9.3 scores them separately and adds capacity support for contracting authorities and SEs. Lowest-scoring criteria across the 45 countries: evidence of use (mean 36) and capacity support (mean 39).
• Official SE definition (3.1). Version 0.9.2 recorded only whether an official definition existed in law or an adopted strategy, yes or no. Version 0.9.3 also gives partial credit to a pending draft or a non-binding document and scores what the definition contains: a social aim, participatory governance and limits on profit distribution.
• Legal recognition of SEs (3.2.1). Version 0.9.2 distinguished no law, a partial framework and a general law in force. Version 0.9.3 records five facts — adoption stage, scope, implementing rules, the status or register the law creates, and the bodies it provides for — so a law in force that has not been implemented no longer scores like one that has. The law’s headline level on a scale of 0 to 3 is kept beside the score. Lowest-scoring criteria across the 45 countries: bodies foreseen by the law (mean 40) and status or register (mean 44).
• SE support ecosystem (4.3.6). Version 0.9.2 coded whether a national network of support structures operated. Version 0.9.3 scores the programme’s lead, its territorial coverage, how it is funded, how long it has run and the range of services it offers. Lowest-scoring criteria across the 45 countries: public funding of support (mean 48) and national programme (mean 60).
• EU funds mobilised for SE (5.1.1). Version 0.9.2 judged the scale of EU-fund mobilisation in one code. Version 0.9.3 scores programming, calls and instruments, the channel through which funds reach SEs and published evidence of take-up separately. Lowest-scoring criteria across the 45 countries: evidence of take-up (mean 42) and calls and instruments (mean 56).
• Dedicated national funding (5.2.2). Version 0.9.2 gave the top level to any dedicated national programme or instrument for SE. Version 0.9.3 also asks how many types of instrument are used, for how long, and whether amounts or beneficiaries were published for 2024 or 2025. Lowest-scoring criteria across the 45 countries: instrument types (mean 24) and execution (mean 36).

The level descriptions for every criterion are in the methodology note (section 6) and in the policy-criteria file of the open dataset.

How the answers were coded and checked

The reference date is 31 December 2025. Every answer rests on public documents: laws and official pages first, then public reviews by the European Commission, the OECD, the ILO and UNDP. The non-public country reports served only as a starting point for the EU member states; no answer rests on them.

The codebook was piloted by coding four countries twice, independently: 79% of the 112 answers matched exactly and 96% were within one level (Cohen’s kappa 0.71, quadratic-weighted 0.88). After the codebook was revised, a second coder coded eight countries independently: 77% exact agreement and 98% within one level on 224 answers (kappa 0.68, quadratic-weighted 0.88). Both codings were made with an AI model working from public sources under the authors’ rules, so these figures measure how consistently the codebook can be applied, not agreement between independent experts.

Before release, the answers of all 45 countries were re-checked against public sources in batches: every answer of low confidence, and every criterion of an indicator whose new score differed from its 0.9.2 code by 40 points or more. In all, 679 answers were re-checked and 195 changed. After the check, 297 answers carry high confidence, 870 medium and 93 low. Where a source could not be read in full, the answer keeps its earlier level and its confidence says so.

Rules and author rulings

Some readings of a document are not settled by a level description. The authors decided them as general rules, written into the methodology note, so that the same situation is coded the same way in every country:

• A single measure is not a chapter. A broader strategy reaches the second-highest level of dedication only with an SE objective, section or several SE measures; a single measure counts as a mention. A score for “an SE chapter or pillar with measures” should not rest on one line in a plan about something else.
• Regional strategies count where there is no national one. The level description allows dedicated regional strategies, and in France the 2014 law on the social and solidarity economy makes regions responsible for regional strategies. France keeps a strategy score of 66.7; without the regional strategy’s content it would be 20.0.
• Equivalents of Article 77. A provision that lets buyers reserve health, social or cultural service contracts for non-profit or social-economy organisations counts like Article 77 of Directive 2014/24/EU, also outside the EU (Norway and Georgia).
• Sectoral or programmatic adoption. A strategy or action plan approved by a ministry or a programme body counts as adopted, and the evidence states that the government did not adopt it:
  • Lithuania, SE strategy / action plan (S1): approved by order of the Minister of Economy (No. 4-207, April 2015), not by a government resolution.
  • Slovakia, SE strategy / action plan (S1): approved by the ministry-chaired ESF+ working group of Programme Slovakia (August 2023, update June 2024), not by a government resolution.

Where a rule changed an answer, the answer is marked “author ruling” in the evidence. The rulings in this release, with the indicator score as coded and as ruled:

What the change does to the scores

The countries whose composite changed most, with the policy indicators that moved most. These are not changes in the countries’ frameworks between releases: the reference date is the same, and the differences come from measuring each framework in more detail and from corrections found in the re-check.

Composite and policy scores 0–100, 0.9.2 → 0.9.3; rank among the 45 countries. Every country’s criteria, evidence and sources are on its profile page and in the explorer.

Limits

The criteria make each answer checkable; they do not make the coding infallible. Agreement was measured on samples, both codings used the same AI-assisted process, and 93 answers remain of low confidence. The policy layer measures how complete a country’s framework is on paper and in its documented operation, not its quality or its effects on social enterprises. Scores of 0.9.2 and 0.9.3 should not be read as a time series; the time series in the dataset is recomputed on the 0.9.3 coding.

Files: methodology note · policy criteria (CSV) · open dataset. Later releases keep the 0.9.3 scores of the 45 countries unless their release notes say otherwise.