Method
How Citoyen stands behind its figures
Every figure comes from an official source and passes checks before it is published. Here is how, and what we do when we get something wrong.
Four commitments
Official sources only
Every figure comes from a statistics office, a central bank or an international organisation. The producer is named on every card, with a link to the original series. The few values from a research centre or a private source carry a badge that says so.
Checked before publication
Every new value goes through automatic checks. A suspicious value is set aside and examined, never silently corrected or rounded.
Compared on the same definition
Comparisons and rankings across countries read one harmonised source (Eurostat, OECD, IMF, World Bank…). Two different national definitions are never drawn on the same chart.
AI helps, it does not decide
AI agents collect, check and write a first draft. A person approves every new indicator and every publication. No figure is invented: missing data stays missing.
The journey of a data point, from source to card
Every figure shown goes through these four steps.
- The sources
- AI
- The human
Official sources publish
Statistical institutes, central banks, international organisations: INSEE, Eurostat, OECD, IMF, World Bank, ONS, Destatis, Statistics Canada… The figures come from official sources; the rare items entered by hand are sourced and reviewed.
SourcesAutomated collection and checks
Programs fetch new releases and check that they are consistent. An abnormal value is set aside automatically, never silently corrected.
AI and automationHuman review
Every new indicator is first tried on a test version, reviewed, then approved by hand. Updates to existing series are automatic; their anomalies are examined one by one.
HumanTransparent publication
Every card states its source, definition, scope and the date of the latest data point. A link leads to the source, and to the original figure where the source allows it.
You
The automatic checks
They apply to every series, at every update. A series that fails is quarantined: it is not published until the cause is understood.
| Check | What is checked | Example |
|---|---|---|
| Unit | The series arrives in the expected unit. | An unemployment rate must arrive in %, not in thousands of people. |
| Bounds | The value stays within a plausible range. | An unemployment rate outside 0 to 60 % is set aside. |
| Jump | The value does not leap from one period to the next without a known reason. | An unemployment rate moving by more than 15 points in one period is blocked. Real shocks are named one by one, such as Brazil's hyperinflation before 1995. |
| Freshness | The producer still updates the series. | A discontinued series gives way to a more recent source for the same concept. |
| Forecasts | A forecast is never shown as a fact. | Years the IMF projects are dropped: only actual years are published. |
| Repetition | A quarterly value is not passed off as monthly. | A monthly series that changes only one month per quarter is set aside. |
What AI never does
- Invent a figure the sources don't contain. The projections shown are the sources' own, labelled as such.
- Take sides for a government, a party or a policy.
- Publish a new indicator or deep dive on its own.
Neutral by design
- The same indicators and the same method for every country.
- Neutral colours, never a party's.
- Each side's arguments attributed, without taking a position.
When we get it wrong
A corrected error is recorded in the corrections log, with its date and how it was found, and the card concerned says so for 30 days. Producers' normal revisions (a quarter revised by a statistics office, for example) are not errors and are not listed.
Latest corrections
- Wrong series
France · Part-time employment
The card showed the unemployment rate of women under 25: the Insee series had been linked by mistake. It now reads the share of part-time work among people aged 15-64 (Eurostat), and the old values were deleted.
Found by an internal review. · See the card ›
- Forecast shown as a fact
South Korea · Current account balance
The 2025 value was an IMF forecast (3.5 % of GDP), not the actual figure (6.5 %). Forecasts were removed from the 13 cards affected.
Found by the rankings audit. · See the card ›
- Wrong series
Brazil · 10-year bond yield
The series used was a public lending rate (the BNDES TJLP), not a government bond yield. It was removed, and an automatic check now blocks this case.
Found by the rankings audit. · See the card ›
Does a figure look wrong?
The flag at the bottom of every card sends it to us with its context. For a text, a ranking or anything else, write to us.