Methodology and limitations

Limitations

• The data is not updated automatically: it was pulled on the date shown above. Detailed spending data lags one to two years behind the current debate.
• Cross-country comparisons are distorted by public versus private provision, gross versus net benefits, tax expenditures and GDP anomalies. The relevant traps are shown on each answer.
• Outcomes compared with spending show correlation, not causation, and results lag spending.
• Assessments of announced measures are partial: independent estimates do not exist for every measure, and the tool says so. Federal Planning Bureau baselines assume unchanged policy; they are not forecasts.
• The European Commission is labelled separately: it is not a Belgian independent institution, and it is the referee of the EU fiscal rules.
• Distributional effects (who pays) are not covered.
• The AI can be wrong. Check the assumption card. By design, it never makes policy recommendations.

How the assistant works

1. Language: the assistant replies in French, Dutch or English; other languages get an answer in English.
2. Guard: a small classifier first checks the question. A request for a policy recommendation gets a standard reply, and may be turned into a neutral factual question. Questions outside Belgian public finances are politely declined.
3. Computation: the AI never states a number from memory. It finds series in a curated catalogue, then writes Python code that runs on the stored data, without network access. It can also search a stored index of documents and quote them word for word.
4. Checks: every number must match the output of the code or a quoted passage; the assumption card must be complete; comparisons need at least two framings; wording must not recommend a policy. If the checks fail, the assistant retries up to twice, then says it could not answer reliably.
5. Assumption card: each answer lists the data used, the documents cited, the government level, the period, the framings, the peer group, the transformations, the exclusions, the caveats and the code that ran. It is built from a log of what the code actually did.

The dashboard

The dashboard charts are pre-computed from the same data. They are the trusted tier of the site; chat answers are exploratory.

The walk-through text is written around the chart data: every number in it is computed from the chart's own table, and every comparison is checked against that table when the site is built.

Independence labels

Official statistics

Statistical offices and central banks (Eurostat, National Bank of Belgium, OECD, …) applying European accounting rules (ESA 2010).

Independent Belgian institution

Belgian institutions independent of the government: Federal Planning Bureau, Study Committee on Ageing, High Council of Finance, Court of Audit.

European Commission

The European Commission: not a Belgian independent institution, and the referee of the EU fiscal rules.

Government estimate

Figures published by a government (budget documents). Shown as the government's own estimate.

Sources

Every series and document collection the site uses, with the organisation that publishes it and its independence label. Each title links to the publisher's page.

Peer group

Comparisons use a peer group fixed in advance:

• BE
• NL
• DE
• FR
• AT
• DK
• SE
• FI
• EU27_2020
• EA21

Known traps

Pitfalls the assistant checks for; when one applies, it is listed in the caveats of the answer.

• Benefits are taxed differently across countries

  Spending figures count benefits before tax. Countries that tax benefits heavily (e.g. Denmark, Sweden, the Netherlands) get part of the money back, so their gross spending overstates the net effort compared with countries that pay lower, untaxed benefits.

• Public or private provision differs across countries

  Some countries provide through private schemes what others provide through government. Dutch basic health insurance and occupational pensions, for instance, largely sit outside general government, so lower public spending does not necessarily mean less spending overall.

• Spending through the tax system is not counted as spending

  Support given through tax reductions (company cars, pension savings, housing loans, etc.) lowers revenue instead of raising expenditure. Countries that rely more on such tax expenditures look like lower spenders and lower taxers than they really are.

• Ireland's GDP is distorted

  Irish GDP is inflated by the accounting of multinational companies, so any ratio to GDP (spending, debt, taxes) looks much smaller for Ireland than its economic weight suggests. Per-capita figures or modified gross national income are better comparators.

• Luxembourg's GDP is distorted

  Luxembourg's GDP is produced partly by cross-border workers who live and consume abroad, and its population is small, so ratios to GDP look low and per-capita figures look very high.

• Budget documents differ from national accounts

  Government budgets are mostly drawn up on a cash basis and cover one government's own departments, while the national accounts (ESA 2010) record transactions when they occur and cover all units classified in government. The budget balance announced by a government can therefore differ markedly from its ESA balance.

• "The government" can mean several things

  In Belgium "the government" may mean the federal government, Entity I (federal plus social security) or all levels of government together (general government). Figures differ a lot depending on the level, so the answer states which one it uses.

• Subsector figures must not be added up

  Revenue and expenditure of each level of government include the transfers it pays to or receives from the other levels. Adding them up counts these transfers twice; only balances (net lending/borrowing, primary balance) add up to the general government total.

• Compare government levels only with other federal countries

  How tasks are split between central, regional and local government depends on each country's constitution. Only Germany and Austria have a federated (state) level comparable to Belgium's Communities and Regions, so subsector comparisons with other countries say little.

• Latest figures are provisional

  Some of the figures used are flagged as provisional by the source and may still be revised, sometimes noticeably. Comparisons use by default the latest year with data for every country; the countries whose figures are provisional in that year are named.

• A projection at unchanged policy is not a forecast

  The Federal Planning Bureau and the Study Committee on Ageing project what would happen if current policy stayed as it is. They do not predict what governments will decide, so the projected deficits show the size of the challenge rather than the most likely outcome.

• Amounts in euro are not adjusted for inflation

  Amounts in current euro grow with prices and with the economy, so a rise over time does not by itself mean more real resources. Ratios to GDP or values at constant prices are better for comparing years.

• Structural and nominal balances are different measures

  The nominal balance (net lending/borrowing) is what is actually recorded. The structural balance removes the estimated effect of the business cycle and of one-off measures; it is an estimate that depends on the method and is revised often. EU fiscal rules focus on structural and net-expenditure measures.

• % of GDP and per capita tell different stories

  A ratio to GDP measures effort relative to the size of the economy; an amount per inhabitant in purchasing power standards measures resources per person. A rich country can spend less in % of GDP but more per person, so comparisons show both framings.

• Spending and outcomes - correlation is not causation

  Outcomes such as life expectancy, poverty or school results depend on many factors besides public spending (income, lifestyle, demography, private spending). Putting spending next to outcomes shows an association, not that one causes the other.

• Results follow spending with a delay

  The effect of spending on outcomes can take years - education spending shows up in skills and jobs a decade later, prevention in health much later. Comparing spending and outcomes for the same year can mislead.

• Datasets of different vintages do not match exactly

  Expenditure by function (September 2026) and the main aggregates and NBB accounts (July 2026) come from different releases, so the same total can differ slightly (e.g. Belgian 2024 spending of 335,288 vs 335,100 million euro). Compare within one dataset, or allow a tolerance of about 0.1%.

• Euro amounts before 1999 depend on the conversion used

  Before 1999 Belgian accounts were kept in Belgian francs. Eurostat converts them through the ECU exchange rate of each year, while the NBB uses the irrevocable euro rate, so Belgian amounts in euro for 1995-1998 differ between the two sources by up to 4.6%. Ratios to GDP are not affected.

• Two views of the federal and regional balances

  The NBB publishes the federal and regional balances in pure ESA terms (S1311, S1312) and in a second view (TCR, FDTCR, SFLC, SWLR, SBCR) that includes advance payments of regional surcharges on personal income tax. In 2024 the regions' deficit is 9,420 million euro in the first view and 8,609 million in the second. Use S1311/S1312 by default and never mix the two.

• Use Maastricht debt, not financial-accounts debt

  The NBB financial accounts (DF_FINGOV_BALANCE) value government liabilities at market value, which gives a different total (about 610 billion euro against 644 billion of Maastricht debt). Public debt in the EU sense is consolidated gross debt at face value (DF_CGD, EDP tables).

• Subsector debts do not add up to total debt

  Each level of government can hold debt issued by another level. Total government debt removes these holdings (inter-subsector consolidation), so the subsector debts add up to the total only after subtracting that consolidation.

• Entity I/II balances are published only in % of GDP

  The NBB items for Entity I and Entity II exist only as a percentage of GDP. Amounts in euro must be computed as the sum of the federal and social-security balances (Entity I) and of the Communities and Regions and local balances (Entity II).

• Regional entities add up only with the residual lines

  The individual Communities, Regions and community commissions add up to the Communities and Regions total only together with the interregional units / not allocated line and the statistical discrepancy. Some sources also leave out small entities (e.g. the German-speaking Community).

• Social benefits are not transfers between governments

  Social benefits are paid to households. Money flowing between levels of government is recorded in separate transfer items (transfers of fiscal receipts, other current and capital transfers to other government). The NBB dataflow on social benefits (DF_NFGOVISB_DISS) does not show inter-government balances.

• Transfers between levels of government inflate subsector spending

  COFOG group 01.8 (transfers of a general character between levels of government) is large for each subsector but almost zero for general government. It mostly reflects money passed on to another level, which spends it under another function.

• The EDP table has no provisional flags

  The deficit and debt notification table carries no status flags, even for the latest year notified in April 2026. That year should be treated as provisional, as flagged in the main aggregates table, and may be revised in the October notification.

• EU and euro-area averages start late

  For several tables the EU and euro-area aggregates start much later than the country series (spending by function from 2010, social protection from 2008, life expectancy from 2002 and 2014) and may end a year earlier. Long comparisons with the EU average are therefore shorter than the country data.

• No EU or euro-area average in this source

  This source publishes no EU27 or euro-area aggregate for the series used (at most an OECD or EU25 average). Comparisons use the peer countries, or the alternative noted in the defaults, instead of an EU average.

• The structural balance is in % of potential GDP

  The European Commission expresses the structural balance as a share of potential GDP (the estimated output at normal capacity), not of actual GDP. It cannot be subtracted directly from a nominal balance in % of GDP, and potential GDP is itself an estimate.

• Projected years are inferred from the publication title

  The FPB and ageing-committee workbooks do not mark which years are projected. The catalog flags years from 2026 as projections because the publications cover 2026 onwards; the latest observed years (2025) may themselves be partly estimates.

• Figures quoted from documents are not computed

  Figures quoted word for word from a report were not computed from the snapshot and cannot be checked against it. They are shown with the document and page, and are not used in calculations.

• Government estimate, not an independent assessment

  These figures come from the government's own budget documents or plans. They have not been checked by an independent institution; where available, the assessments of the Court of Audit, the High Council of Finance or the Federal Planning Bureau should be read alongside.

• European Commission figures are not Belgian-independent

  The European Commission is not a Belgian independent institution - it also applies the EU fiscal rules to Belgium. Its forecasts assume no policy change and are labelled separately from the Federal Planning Bureau's projections.

• OECD and ESA tax categories differ

  The OECD classifies taxes differently from the European national accounts (ESA) - for instance, it excludes imputed social contributions and groups some taxes elsewhere. Totals from OECD Revenue Statistics and Eurostat therefore differ slightly; do not mix them in one comparison.

• Poverty figures refer to the previous year's income

  The year shown is the year of the EU-SILC survey. In most countries the income measured is that of the previous calendar year, so policy changes show up in the figures with a delay.

• Indexation follows the smoothed health index, not inflation

  Social benefits and public-sector wages are indexed when the smoothed health index passes a threshold (the pivot index), so they follow prices with a delay and in steps. The health index leaves out alcohol, tobacco and motor fuels, so it can rise less or more than the consumer price index. Many private-sector wages follow their own sectoral rules.

• Tax-return statistics are not government revenue

  These figures are tax computed on returns, by income year and by the taxpayer's place of residence. National-accounts revenue records personal income tax (withholding, prepayments and assessments) when it is due, so the two differ in timing, coverage and level and should not be compared or combined directly.

• Euro-area aggregates with different compositions

  EA21 is the euro area including Bulgaria (from 2026) and EA20 the composition of 2023-2025; both are back-calculated over the whole period with a fixed list of countries. Use EA21 by default and do not mix the two in one comparison.

• Table drafted from a PDF, not yet verified

  This table was drafted from a published PDF by a language model and has not yet been checked row by row by a person. It cannot be loaded for computation until it is verified.

• Statbel and Eurostat count the Belgian population differently

  Statbel and the Federal Planning Bureau use the legal (register) population, which leaves out asylum seekers in the waiting register. Eurostat's population for Belgium includes them from 2011, so it is 0.2-0.5% higher (about 58,000 people on 1 January 2025). Do not mix the two in one ratio or growth rate; use Eurostat for per-capita figures and international comparisons, and Statbel/FPB for age or regional detail.

• Household types come from the population register

  Statbel derives household types from the National Register, not from a survey. "With children" means at least one child of any age in the household, and unmarried couples are only recognised when the register shows the link, so older years understate them. Long-run shifts between married and unmarried couples are partly a registration effect; compare totals and one-person households with more confidence than couple types.

• Brussels GDP per head is inflated by commuters

  GDP is counted where people work, but divided by the people who live there. About half of the jobs in Brussels are held by commuters from Flanders and Wallonia, so Brussels shows the highest GDP per inhabitant in Belgium while the income of its residents is below the national average; the same effect, in reverse, lowers GDP per head in the commuter belt (Halle-Vilvoorde, Walloon Brabant). For living standards by region use household disposable income per inhabitant (regional_household_income), which is measured where people live.

• NACE public administration, education and health are not the government sector

  NACE sections O (public administration and defence), P (education) and Q (health and social work) group establishments by what they do, not by who owns them. They include private and subsidised schools, hospitals and care homes, and leave out government units that do other things (e.g. public transport). Pay and staff in O-Q are therefore not the pay and staff of general government (S13); for government wage spending use compensation of employees paid by S13 (D1PAY in gg_main_aggregates, D1 in gg_exp_cofog).

• Compensation of employees is more than wages

  Compensation of employees (D1) is gross wages and salaries (D11) plus employers' social contributions, including imputed contributions for pension schemes that employers pay directly, such as the pensions of statutory civil servants in Belgium. Employer contribution rates differ by country and between public and private employers, so compare D11 for gross pay and D1 for labour cost.

• Pay per person depends on part-time work

  Employment in persons counts a part-time worker as one person. Where part-time work is common (the Netherlands; education, health and care in Belgium), compensation per person is lower than compensation per full-time equivalent. Divide by hours worked (THS_HW) where available, or state that the figure is per person.

• Theoretical replacement rates assume a full career

  OECD replacement rates are model calculations, not observed pensions. They assume a person who enters the labour market at 22 in 2024, works without interruption at a constant relative wage until the normal pension age, and retires under today's legislated rules (including scheduled future changes). Actual pensions are lower for people with career breaks, part-time work or early exit, and a replacement rate says nothing about how many people receive which pension or at what cost to public finances.

• ODA includes refugee costs spent in the donor country

  Under DAC rules a donor may count the first-year costs of hosting refugees and asylum seekers at home as ODA (measure 1820). These costs rose sharply in 2015-2016 and again from 2022 (Ukraine), so part of the increase in ODA in those years was not spent in developing countries. Show ODA excluding in-donor refugee costs (1010 minus 1820) when the question is about aid to developing countries.

• ODA headline measure changed in 2018

  Since 2018 the official ODA figure and the ODA/GNI ratio are on a grant-equivalent basis (only the grant part of loans counts, FLOW_TYPE 1160, measures 11010 and 11002). Earlier years use net disbursements (FLOW_TYPE 1140, measures 1010 and 2). Do not join the two into one series without saying so; net disbursements are still published after 2018 for comparison.

• The tax wedge is for model households at the average wage

  OECD Taxing Wages computes taxes and contributions for model households (single or couple, 0 or 2 children) earning a multiple of the average full-time wage, using standard tax rules only. It is not the average tax actually paid by workers, and it ignores most deductions, local variation and non-wage income. The tax wedge is in % of labour costs (employer contributions included), while the income-tax and contribution rates are in % of the gross wage, so the components do not add up to the wedge directly.

• Practising, professionally active and licensed staff are not comparable

  Countries report health personnel as practising (seeing patients), professionally active (also in research, administration, industry) or licensed to practise (everyone entitled, including retirees and people working abroad). Licensed counts are much higher. Compare countries on the same status, normally practising; Belgium reports practising and licensed physicians but no professionally active figure, and its practising counts rely on a minimum-activity threshold that changed over time (series breaks flagged b).

• Survey (ILO) unemployment is not the number of registered job-seekers

  ILO unemployment comes from the labour force survey - people without work who searched for a job in the last four weeks and can start within two weeks. Registered unemployed job-seekers (Actiris, VDAB, Forem, ADG) and people receiving unemployment benefits (ONEM/RVA) are administrative counts with other rules; in Belgium they are clearly higher than the ILO figure. Do not mix the two or present one as the other.

• Break in the labour force survey in 2021

  The EU labour force survey was reformed in 2021 (new questionnaire and definitions, Regulation 2019/1700), and several countries have other breaks (b flags; for Belgium also 1999, 2001, 2005, 2011 and 2017). Changes across a break year are partly statistical. For long trends in the 20-64 employment rate use employment_rate, which Eurostat adjusts for the breaks; elsewhere mention the break when a comparison spans it.

• Low-reliability survey estimates

  Cells flagged u rest on few survey respondents (small provinces, young age groups, rare categories) and can move a lot from year to year. Report them with that warning and do not rank regions or countries on small differences between them.

• Inactive "by reason" covers only those who want to work

  inactive_by_reason counts only people outside the labour force who say they would like to work but are not looking (about 4% of Belgians aged 15-64), by their main reason. It is not the breakdown of all inactive people, it has no retirement reason, and "own illness or disability" is not the number of people on sickness or invalidity benefits (INAMI/RIZIV). For the number of inactive people use labour_status (INAC).

• Regional labour rates are by place of residence

  Regional employment and unemployment rates describe the residents of a region, wherever they work. Brussels has many more jobs than employed residents (a large share is filled by commuters from Flanders and Wallonia), so its low employment rate does not mean few jobs in Brussels.

• Pension types count the same person more than once

  Someone who receives, for example, an old-age and a survivor's pension is counted under both types. Only the total counts each beneficiary once, so do not add the types together. Beneficiaries are also not the same as pensions paid, and ESSPROS counts differ from national administrative counts such as the Federal Pension Service's (other schemes, reference date and definitions).

• EU-SILC figures refer to the previous year's income

  The year shown is the year of the EU-SILC survey. In most countries, including Belgium, the income measured is that of the previous calendar year, so policy changes show up with a delay. The survey also has breaks in series (Belgium changed its method in 2019), so compare levels across a break with care.

• Aggregate replacement ratio is not a theoretical replacement rate

  The EU-SILC aggregate replacement ratio compares the median pension of today's 65-74 year-olds with the median earnings of today's 50-59 year-olds, so it reflects past careers and past rules. The OECD and Ageing Report replacement rates are model calculations of what a hypothetical worker with a full career will receive under current rules. The two measure different things and should not be compared directly.

• The national minimum wage is a floor, not the typical low wage

  Belgium's national minimum wage (set by the National Labour Council) is indexed automatically and raised in steps by agreement, and most sectors have their own, higher minimum wages set by joint committees. Few workers are paid exactly the national minimum. Countries without a statutory minimum (Austria, Denmark, Finland, Sweden) have collectively agreed floors that are often higher, so their absence from the data does not mean they have no minimum.

• Net earnings and tax wedges are model calculations

  These figures apply each country's tax and benefit rules to hypothetical households at fixed shares of the average wage. They show what the rules imply, not what real households pay or receive, and they leave out many deductions, local taxes in some countries, and in-work benefits not modelled. The tax wedge also depends on the wage level and family type chosen.

• A population projection is not a forecast

  Eurostat's EUROPOP projection shows what the population would be if its assumptions on fertility, life expectancy and migration held. Only the baseline scenario is kept; the sensitivity variants (lower fertility, higher or lower migration ...) give quite different results far ahead. The Federal Planning Bureau and Statbel publish their own Belgian outlook with other assumptions, so the two do not match exactly.

• EU and euro-area age bands do not add up in some years

  For the EU and euro-area aggregates, Eurostat's 5-year age bands leave out countries with incomplete age detail in 2013-2020 and 2024, so the bands sum to 2-13% less than the total population. Country figures add up exactly. For EU totals use the population series; for EU age structure prefer the shares in the population structure indicators.

• Citizenship is not country of birth

  These figures classify people by their current citizenship, not by where they were born or by their background. People who acquire Belgian nationality move from the foreign to the national group, and nationality rules differ between countries, so the share of foreign citizens is not the share of people born abroad and is hard to compare across countries. Nationals also appear in the migration flows (Belgians moving abroad or returning).

• Net migration includes a statistical adjustment

  Eurostat's net migration is computed as total population change minus natural change (births minus deaths). It therefore also absorbs register corrections and other statistical adjustments, and is not equal to immigrants minus emigrants as counted in the immigration and emigration tables.

• Asylum applications are not people granted protection

  The series counts first-time asylum applicants registered in a year. It is not the number of people granted protection (decisions can be negative and come later), nor the number of people in reception facilities, nor the population of refugees living in the country. A person who applies in several countries can be counted in each.

• No data on asylum decisions, recognition or returns

  The snapshot holds first-time asylum applications only. It has no data on decisions (recognition rates, refusals), on how many applicants stay or leave, on returns or on irregular stays. Do not estimate these from the applications; say that the data does not cover them.

• Migration headcounts carry no cost information

  These tables are headcounts. Public spending in this snapshot is not broken down by citizenship, migration status or asylum procedure, so dividing a spending total by these counts does not give a cost per person in that group. Report the counts as they are.

• Regions and provinces overlap

  The table holds Belgium, its three regions (NUTS1) and the eleven NUTS2 units (Brussels-Capital Region as BE10 plus the ten provinces). Each level adds up to Belgium on its own; mixing levels, or adding a region to its provinces, counts people twice. Regions are territories, not Communities.

• Primary incapacity (first year) is not invalidity (after one year)

  Sickness insurance pays primary incapacity benefits during the first year of work incapacity (after the employer's guaranteed wage for employees) and invalidity benefits once incapacity lasts more than a year and the medical council recognises the person as invalid. inami_incapacity counts only people in invalidity at 31 December; INAMI does not publish a year-end headcount of people in primary incapacity, only benefit days and amounts (inami_benefit_spending, phase PRIMARY, to 2023). Do not present the invalidity count as "all people on sick leave".

• "Long-term sick" in the press usually means invalidity only

  Press figures of about 500,000 "long-term sick" are usually the INAMI invalidity count for employees and unemployed people (537,728 at the end of 2025). Some add the self-employed (38,915) or people in primary incapacity, others count the general scheme only. Say which regime and which phase a number covers; add EMP and SELF for the whole insured population.

• Employees and self-employed are separate INAMI schemes

  Every INAMI row carries a regime, EMP (employees and unemployed, general scheme) or SELF (self-employed and helping spouses). The schemes have different rules (waiting period, benefit levels, the catch-up premium), so their counts and averages are not directly comparable. No row holds both; add EMP and SELF yourself for a national total, and never compare the self-employed figure alone with a national total.

• INAMI breakdowns cover the same people

  In inami_incapacity the breakdowns status_sex, age and residence_region each add up to the same invalidity total, and status ALL / sex T rows are totals of the others. Always filter on breakdown and pick either the total row or its parts; summing across breakdowns or across total and detail rows counts people two or three times. Age bands exist only for 31 December of each release year (2014-2025), and the band under 25 replaced "under 20" and "20 to 24" from the 2022 release.

• Survey estimates and administrative counts measure different things

  Labour force survey and EU-SILC figures are estimates from a sample of households, with definitions harmonised across countries (for example ILO unemployment, or inactive because of illness). Administrative counts (INAMI/RIZIV invalidity, Federal Pension Service pensioners, tax returns) record everyone entitled to a benefit or a status under Belgian rules, on a reference date. The two cover different people and do not match even for the same topic, so one cannot be divided by the other or presented as the other.

• Persons, benefits and payments are different counts

  A person can receive several benefits at once (a retirement and a survivor's pension, pensions from two schemes), and a count of benefit days or payments is not a count of people. Counts of pensions or benefits are therefore higher than the number of beneficiaries, and amounts divided by them give an amount per benefit, not per person. Say which of the three a figure counts and add benefits only when the source says each person is counted once.

• Spending on a function covers more than the group counted

  Dividing a spending total by a head count gives an amount per person of that group only if the spending goes to that group. Functional spending is usually broader - unemployment spending (COFOG 10.5) also pays partial and temporary unemployment and some early-retirement benefits to people who are not ILO-unemployed; old-age spending includes survivors' and other benefits - and the head count may come from a survey with another scope. Present the result as spending divided by the number of people in the group, not as the benefit each person receives.

• Projected years do not simply continue the observed ones

  Observed and projected population figures come from different tables. A projection starts from its own base population and assumptions (fertility, life expectancy, migration), so its first years need not match the latest observations, and the Eurostat (EUROPOP) and Federal Planning Bureau projections differ from each other. Label the projected years, keep each projection with its own observed series (Eurostat with Eurostat, FPB with Statbel), and do not read a jump at the junction as a real change.

• Model-household amounts, not average benefits paid

  These are FPS Social Security calculations for a model household living only on one minimum benefit, net of personal income tax and including holiday pay, child benefit and social supplements. They are not the average amount paid to recipients, not a gross statutory scale, and not the number of people on a minimum. Amounts depend on the household rate (single, cohabiting, with family). 2023 and 2024 shares of the poverty line rest on an estimated threshold (flag e).

• Net benefit amounts vs the gross minimum wage

  The FPS minimum-benefit series are net disposable incomes; Eurostat's minimum_wage is a gross monthly wage before tax and social contributions. Comparing them directly overstates the gap between work and benefits. For a like-for-like comparison of net incomes in and out of work use the OECD series (oecd_min_income_adequacy, net_earnings) or state the gross/net difference explicitly. Belgium also has sector minima above the national minimum wage.

• People in reception are not asylum applications

  Fedasil figures are the number of places and of people accommodated on one day. They are not the number of applications in a year (asylum_applications, a flow), nor decisions or people granted protection. Not every applicant is in the network (some live elsewhere or wait for a place), and the network can also house people whose procedure has ended for a transition period. Capacity is a policy choice; the waiting list depends on it. Report the figures as they are, with the reference date.

• Reception headcounts carry no cost information

  These are counts of places and people. The snapshot has no reception spending per place or per person, so do not derive a cost per place, per person or per applicant by dividing a budget or a spending total by these counts. Report the counts as they are.

• ONEM payment units are not persons

  ONEM counts payment units - the annual average of monthly units, where a partial benefit or a month paid in part counts as a fraction. The number of different people who received a benefit during the year is higher, and the number on a given date can differ. Say "on average X payment units (about X full-time equivalents of benefit) a year", not "X people".

• ONEM categories overlap and have no grand total

  Some items are totals of others (CCI_DE_TOTAL includes the CCI-DE statuses and RCC_DE; RCC_TOTAL = RCC_DE + RCC_NDE; CB_TOTAL, TC_TOTAL, TL_TOTAL are group totals), and the sex, age, household and duration breakdowns are alternative views of the same units. Never sum all rows. Pick one item per concept, filter every breakdown to TOTAL / T unless it is the subject, and do not present a sum across schemes as "ONEM beneficiaries".

• Temporary unemployment and career breaks are not unemployment

  Temporary unemployment is paid to employees who keep their job while work is suspended (economic reasons, weather, force majeure - massively in 2020-2021); career break, time credit and thematic leave are paid to people in work who reduce or stop for a while. ONEM also pays part-time workers, activation subsidies and exempted job-seekers. None of these people are unemployed; only fully unemployed job-seekers (CCI-DE) approach the usual meaning.

• Benefit recipients are not registered job-seekers

  ONEM counts people paid an unemployment benefit. Job-seekers are registered and counted by the regional employment services (VDAB, Forem, Actiris, ADG), which also register people without a benefit (e.g. young people in their waiting period, or people excluded from benefits); those counts are not in this snapshot. Survey (ILO) unemployment is a third, different measure.

• Unemployment benefits limited in time from 2026

  The 2025-2026 reform limits unemployment benefits in time; ONEM excludes people in successive waves from January 2026. Some of them turn to the integration income paid by the CPAS/OCMW (integration_income), others find work or have no income support. The annual series here end in 2025, before the first exclusions, so they do not show the shift; do not extrapolate 2025 trends to 2026.

• Integration income recipients are counted once a year

  The series counts every person paid the integration income at least once in the year. Because many people receive it only for some months, this is well above the number paid in any given month and is not comparable with ONEM annual averages. People who move or change family category during the year are counted in each region or category, so regions and categories do not add up to the total; use the published totals.

• NATO defence spending is not COFOG defence

  NATO's figures are reported by ministries of defence under NATO's own definition. They include military pensions, part of the gendarmerie or other forces trained and equipped for military tasks, peacekeeping and military aid, and count equipment when it is paid for or delivered. COFOG function 02 (defence) in the national accounts excludes pensions (they go to social protection), puts police in public order, and records equipment on an accrual basis. The two can differ by several tenths of a percentage point of GDP. Quote NATO for the 2% / 3.5% targets, and COFOG to compare with other government functions.

• Belgian education is run by the three Communities

  Education in Belgium is a competence of the Flemish, French and German-speaking Communities, each with its own school system, rules and budget. Eurostat publishes only the Belgian total, so pupils, teachers and ratios per Community are not in these tables; spending per pupil from COFOG GF09 is a Belgian average across three systems.

• Pupils per teacher is not class size

  The pupil/teacher ratio divides full-time-equivalent pupils by full-time-equivalent teachers, including teachers for remedial classes, special needs and subjects taught to split groups. Classes are therefore larger than the ratio. The teachers series counts persons, not FTE, so dividing pupils by it gives a different and lower number than the published ratio.

• Prisoners are counted on one day of the year

  The prison population is a stock on a reference day, including people in pre-trial detention; it is not the number of people imprisoned during the year, and it excludes people under electronic monitoring at home. Official capacity is the number of places; an occupancy above 100% means overcrowding. Countries differ in what they count, so compare trends more than levels.