Notation des allocataires : l'indécence des pratiques de la CAF désormais indéniable

Notation des allocataires : l'indécence des pratiques de la CAF désormais indéniable ("Scoring of welfare beneficiaries: the indecency of CAF's algorithm now undeniable") is the 27 November 2023 investigative publication of La Quadrature du Net (LQDN), produced with the French anti-precarity collectives Stop Contrôles and Changer de Cap, that releases and substantively analyses two leaked versions of the algorithmic risk-scoring system operated since 2010 by the Caisse nationale des Allocations Familiales (CNAF) — the French national family-benefits agency — to assign "scores de suspicion" (suspicion scores) to roughly 13 million beneficiary households covering some 32 million people. The publication has a three-part shape — a substantive narrative analysis published bilingually in French and English; a methodological annex documenting the algorithm's variable list and weightings; and a public GitLab release of the 2010-2014 and 2014-2018 source-code versions obtained from the Commission d'accès aux documents administratifs (CADA) after some eighteen months of administrative-law effort. It is the corpus's first French-language Publication anchor, the lead-file investigative report inside LQDN's wider France Contrôle multi-year campaign on algorithmic welfare-control, and the evidentiary base on which the October 2024 Conseil d'État filing by the original fifteen-organisation French civil-society coalition rests.

Argument and central framing

The report's substantive working argument is that the CNAF algorithm is not a broken anti-fraud tool but a successfully operating discrimination system: that the algorithm does not malfunction when it disproportionately targets low-income, single-parent, disabled, and unemployed beneficiaries for administrative checks, but rather operates as designed — and that this is the substantive accusation the source-code disclosure makes undeniable. The report's launch banner, "L'indécence des pratiques de la CAF désormais indéniable" (the indecency of CAF's practices now undeniable), reframes the file from a contested technical debate about algorithmic variables to a documented administrative-and-political register: the algorithm is shown to score the very populations the family-benefits agency exists to support, intensifying surveillance precisely during the moments of life instability the welfare system is built to cushion. The rhetorical anchor "algorithme de la honte" (algorithm of shame) sits inside the report's substantive register alongside the working "double peine" (double penalty) frame — that the same beneficiary populations are the targets of welfare provision and of administrative-control suspicion at once — and the canonical campaign-register phrase "algorithme de notation des allocataires" (beneficiary scoring algorithm) that the paired campaign framing entry is built around.

Source-code disclosure and methodological annex

The report's evidentiary spine is the public release of the algorithm's source code, secured by LQDN through a sustained administrative-law effort. The Commission d'accès aux documents administratifs (CADA), the French administrative authority charged with arbitrating access-to-public-documents requests, ordered the CNAF to release the algorithm after LQDN's freedom-of-information requests had been repeatedly stonewalled or partially redacted across the eighteen-month 2022 working period — itself documented in the LQDN follow-up "Notation des allocataires : fébrile, la CAF s'enferme dans l'opacité" (23 December 2022, updated 12 January 2023), which records the CNAF's heavily redacted initial response and the misleading public communications the agency mounted against the disclosure file. The report's GitLab companion, opened on the same day as the publication under the Etalab Open License 2.0, hosts the two released algorithm versions (2010-2014 and 2014-2018) — the CNAF withheld the current version — along with the methodological annex through which LQDN reconstructs the variable list, the weighting register, and the methodology used to construct the profile-types the algorithm scores against. The substantive move of the methodological annex is to make the algorithm's discrimination operational and inspectable: the document is the working evidentiary base on which all subsequent civil-society and legal arguments against the system rest.

Discriminatory variables and profile simulations

The report's discrimination-criteria analysis anchors on a substantive socio-economic register: the variables that drive the suspicion score most heavily are precisely the markers of precarity — low income, unemployment, receipt of the Revenu de Solidarité Active (RSA) minimum-income benefit, residence in disadvantaged neighbourhoods, disability status while employed, single-parent household composition, and high rent-to-income ratios. The report's profile-simulation graphs render these effects in operational terms by plotting the suspicion-score curves for representative beneficiary profiles, showing the substantive threshold effects at which the algorithm flags households for administrative control. The substantive register the simulations build is that the algorithm operates as a structurally discriminatory selection mechanism rather than as an individually-calibrated risk model — that the population subject to enhanced administrative-control suspicion is, in practice, the population the welfare system is designed to support. The substantive argument the report's Amnesty International translation carries forward into the Anglosphere civil-society register — that the system "operates in direct opposition to human rights standards, violating the right to equality and non-discrimination" — sits directly on the report's variable-and-simulation evidentiary base.

Authorship and prior-art register

The report is institutionally authored: there are no individual bylines. The named co-producers are LQDN as the lead organisation and the Stop Contrôles and Changer de Cap collectives — French anti-precarity civic coalitions whose substantive working register on welfare-control practice supplies the on-the-ground experiential base alongside LQDN's technical-and-legal register. The report's prior-art literature on which the substantive analysis rests includes Vincent Dubois's 2016 (with Morgane Paris and Pierre-Édouard Weil) sociological study of CAF control and fraud-policy practice and his 2021 chapter on "controlling the assisted", Bénédicte Collinet's 2013 study of data-mining at the CAF, the 2010 CNIL official opinion on the algorithm, and the Défenseur des droits' 2017 report on welfare-fraud prevention. The institutional-authorship and prior-art register together situate the report inside the sustained French civil-society and sociological-research conversation on welfare-administration practice that the report's source-code release operationalises through algorithmic-system inspection.

Travel into the litigation and the European coalition

The report's most consequential downstream artefact is the 16 October 2024 Conseil d'État filing by the original fifteen-organisation French civil-society coalition that LQDN coordinated — anti-poverty, anti-discrimination, disability-rights, social-work, immigrant-defence, basic-income, civic-tech, and lawyers' organisations together translating the report's evidentiary base into a substantive legal case before France's highest administrative court. LQDN's 2 December 2024 English-language summary of its arguments renders the substantive legal-arguments scaffolding along three axes — French and European anti-discrimination law on equal treatment in administrative decision-making; GDPR Article 22 on the right not to be subjected to decisions based solely on automated processing where the decision produces legal or significantly similar effects; and French administrative-law proportionality on the necessity-and-proportionality test for state interference with the rights of beneficiaries — and shows the report's substantive findings carrying directly into each axis. The March 2024 escalation publication on the CNIL's approval of CNAF access to the DRM (Dispositif de Ressources Mensuelles) monthly-income database — covering 12 million beneficiaries in real time — is the report's working evidence that the November 2023 disclosure had not deterred the agency's continuing operational expansion, and the substantive working datum the LQDN follow-up cites against the agency's efficiency-rebuttal — that recovered overpayments amount to only 0.2% of total CAF disbursements — translates the report's substantive accusation into a quantitative-policy register.

The January 2026 European-coalition expansion brought the case from a fifteen-organisation French civil-society file to a twenty-five-organisation French-and-European one, with European Digital Rights (EDRi), AlgorithmWatch, the Panoptykon Foundation, the European Network Against Racism, and Data for Good among the joining organisations alongside the French national trade-union confederations — the paired campaign framing entry records the cross-Continental coalition shape in detail. The report's downstream working life inside the European civil-society field is the substantive evidentiary base on which the cross-national convergence on welfare-algorithm contestation has been built.

Posture within the corpus

Within the corpus, Notation des allocataires : l'indécence des pratiques de la CAF désormais indéniable is the publication-side anchor of La Quadrature du Net on the French / Continental-European publications slate, the first French-language Publication entry, and the lead-file artefact of the wider France Contrôle investigative-campaign portfolio that includes LQDN's ongoing work on the France Travail unemployment-agency algorithm, the Caisse nationale d'Assurance Maladie health-insurance algorithm, the French tax-administration risk-scoring system, and the retirement-authorities algorithms. The org-side body identifies the algorithmic-public-administration litigation portfolio as LQDN's most substantively distinctive contribution, and this report sits at its conceptual and evidentiary centre.

In the corpus's publications slate, the report is the welfare-algorithm-contestation foundational artefact, alongside the existing AlgorithmWatch-anchored Automating Society Report 2020 Continental-European multi-country ADM mapping, the Panoptykon Foundation-anchored Black-Boxed Politics: Opacity is a Choice in AI Systems Polish / Central-Eastern European foundational essay on algorithmic-opacity-as-political-choice, and the Derechos Digitales-anchored Decisiones automatizadas en la función pública en América Latina Spanish-language Latin American multi-country ADM mapping. The four anchor the corpus's emerging Continental and Global-South publication-register on automated-decision-making contestation: AlgorithmWatch maps the European empirical register, Panoptykon names the algorithmic-opacity-as-political-choice conceptual frame, Derechos Digitales maps the Latin American comparative register, and LQDN supplies the source-code-disclosure-and-litigation evidentiary register that ties the French national welfare-algorithm file to the wider European civil-society conversation.

The report's lasting movement-side contribution is that it converts the substantive debate about welfare-control algorithms from a contested technical question into a documented administrative-and-political record on which civil-society legal challenge, European institutional advocacy, and cross-national coalition expansion have since been built. In the corpus's terms it is the load-bearing French-language Publication anchor on the algorithmic-public-administration contestation line and the foundational investigative artefact on which the wider Continental European welfare-algorithm-contestation register is built.