Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

Serbia Social Card (Socijalna karta)

Cut off by a data match: a social-assistance registry

A rules-based registry cross-links roughly 130 to 135 data fields from other state registers to check who still qualifies for social assistance and to flag suspected income, and a matched flag flows into a benefit cut and a three-month reapplication lockout before anyone reconciles it against the real world. Modeled on Serbia's Social Card (Socijalna karta). Watch where correction is supposed to live: the informal, cash-based reality of the poorest collapses into stale snapshots, and the people cut off can rarely reach the channel that would fix it.

Stylized model of a documented deploymentPublic benefits & eligibility

Open this example in PAN Lab v0.1 to apply pressures and levers and watch what the system does.

What this models

This example runs on the Social-Card-class cross-registry data-matching registry network: 5 components and 12 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 3 assumed · 3 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

  • assumed

    This models the cross-registry data-matching pattern documented in the Serbia Social Card case file - not a reconstruction of the actual registry or its undisclosed rules.

  • baseline

    The system is best modeled as rules-based cross-registry data-matching, not a predictive model: the underlying rules are undisclosed and there is no evidence it uses machine learning, so a better classifier is not the leverage point here.

  • assumed

    Peer pathways are authored on both signs: deference to the system spreads worker to worker and one matching ruleset's blind spot is shared across every household, while the internal second-look pathway starts closed - workers report they cannot change what the system recorded.

  • baseline

    The cross-registry-feeds node carries a real inflow into the model - roughly 130 to 135 fields the engine matches against, often from stale snapshots - so it enters the dynamics, not just the picture. What that linkage reaches (data collected for other purposes, covering relatives and connected persons) and whom it exposes are documented in the case file, not computed here.

  • baseline

    The defining feature is an absence of in-system correction: A11's World Bank Inspection Panel request alleges that, because the system is semi-automated, social workers cannot correct errors recorded in it, so the record-side reconciliation edge starts closed at baseline. A 15-day appeal window, a mandatory three-month reapplication lockout, and only 361 appeals against more than 100,000 notifications describe a correction channel the people cut off could rarely reach.

  • assumed

    Roma are reported among the most affected because informal earnings are misclassified as income, but the registry records no ethnicity, so this exposure is inferred from benefit reliance and case documentation, not officially disaggregated. This Lab models institutional workflow propagation, not demographics, and estimates no differential harm to served people; that exposure and the individual harms are documented in the case file and measured outside any diagram like this one.

What this example does not show

  • Roma are reported among the most affected because informal earnings are misclassified as income, but the registry records no ethnicity, so this exposure is inferred from benefit reliance and case documentation, not officially disaggregated. The Lab models institutional workflow propagation, not demographics, and estimates no differential harm to served people; that exposure and the individual harms are documented in the case file and measured outside any diagram like this one.
  • The caseload figures are a moving, method-dependent range — about 35,000 fewer recipients by August 2023, at least 44,000 by early 2024, and over 60,000 by October 2025 — and are largely net caseload declines, not audited counts of system-caused removals; the government attributes part of the fall to a stronger economy. The Lab uses the case's shape, not calibrated rates.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

  • Serbia's Social Card (Socijalna karta) registry, given a statutory basis by the Law on the Social Card in force from 1 March 2022 and financed in part by an 82.6 million euro World Bank public-sector loan, cross-links roughly 130 to 135 categories of data from other state registers to verify social-assistance eligibility and flag suspected undeclared income or assets. After the law, named sources report the caseload falling by a range of tens of thousands: government figures cited by Amnesty International show about 35,000 fewer recipients by August 2023, A11 counts at least 44,000 people having lost assistance by early 2024, and the UN Working Group on Business and Human Rights reported over 60,000 without assistance by October 2025. These are largely net caseload declines rather than audited counts of system-caused removals, and the government attributes part of the fall to a stronger economy. Roma are reported among the most affected because informal earnings are misclassified as income, but the registry records no ethnicity, so this is inferred rather than officially disaggregated. As of the latest reporting, Constitutional Court, World Bank Inspection Panel, and UN scrutiny were pending or active, with no court or panel yet ordering changes.

    empirical
    • Advocacy A11 - Initiative for Economic and Social Rights, Two Years of the Social Card Law: Fair Distribution of Financial Social Assistance Remains Out of Reach (2024) https://www.a11initiative.org/en/two-years-of-the-social-card-law-fair-distribution-of-financial-social-assistance-remains-out-of-reach-law-should-be-abolished/
    • Advocacy Amnesty International, Trapped by Automation: Poverty and discrimination in Serbia's welfare state (2023) https://www.amnesty.org/en/latest/research/2023/12/trapped-by-automation-poverty-and-discrimination-in-serbias-welfare-state/
    • Investigative Context / Thomson Reuters Foundation, As Serbia adopts digital welfare system, the poorest miss out (2023) https://www.context.news/digital-rights/as-serbia-adopts-digital-welfare-system-the-poorest-miss-out
    • Government UN Working Group on Business and Human Rights, End of Mission Statement, Serbia visit 6-15 October 2025 (2025) https://www.ohchr.org/sites/default/files/documents/issues/business/workinggroupbusiness/2025-10-15-eom-wgbhr-serbia-en.pdf
    • Government World Bank Inspection Panel, Panel Registers the Request for Inspection from Serbia Public Sector Efficiency and Green Recovery Program (2024) https://www.inspectionpanel.org/news/panel-registers-request-inspection-serbia-public-sector-efficiency-and-green-recovery-program
    • Reference China-CEE Institute, Serbia political briefing: Two years of the implementation of the Law on social card (2024) https://china-cee.eu/2024/04/10/serbia-political-briefing-two-years-of-the-implementation-of-the-law-on-social-card/

  • Under Serbia's Social Card system, a removed beneficiary has 15 days to appeal and must wait three months to reapply regardless of changed circumstances, and removal letters frequently reference only unspecified data from the electronic database. A11's Request for Inspection to the World Bank Inspection Panel alleges that, because the system is semi-automated, social workers cannot correct errors recorded in it. Over roughly two years the Ministry processed more than 100,000 notifications of suspected income or asset increases, while beneficiaries filed only 361 appeals against Centers for Social Work rulings; because the two figures cover different populations, the gap illustrates how rarely flags were contested rather than a measured appeal rate. Documented misclassifications include a one-off funeral donation read as income and long-scrapped cars still counted as assets.

    empirical
    • Advocacy Amnesty International, Trapped by Automation: Poverty and discrimination in Serbia's welfare state (2023) https://www.amnesty.org/en/latest/research/2023/12/trapped-by-automation-poverty-and-discrimination-in-serbias-welfare-state/
    • Advocacy A11 - Initiative for Economic and Social Rights, Two Years of the Social Card Law: Fair Distribution of Financial Social Assistance Remains Out of Reach (2024) https://www.a11initiative.org/en/two-years-of-the-social-card-law-fair-distribution-of-financial-social-assistance-remains-out-of-reach-law-should-be-abolished/
    • Government World Bank Inspection Panel, Panel Registers the Request for Inspection from Serbia Public Sector Efficiency and Green Recovery Program (2024) https://www.inspectionpanel.org/news/panel-registers-request-inspection-serbia-public-sector-efficiency-and-green-recovery-program
    • Reference China-CEE Institute, Serbia political briefing: Two years of the implementation of the Law on social card (2024) https://china-cee.eu/2024/04/10/serbia-political-briefing-two-years-of-the-implementation-of-the-law-on-social-card/
    • Investigative Context / Thomson Reuters Foundation, As Serbia adopts digital welfare system, the poorest miss out (2023) https://www.context.news/digital-rights/as-serbia-adopts-digital-welfare-system-the-poorest-miss-out

Where this connects

Institutional pressures in this domain

  • Austerity & recovery incentives — Cost-cutting and overpayment-recovery targets tilt the system toward denial and enforcement errors.
  • Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
  • Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.
  • Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.

All of them in context on the Public benefits & eligibility domain page.

Levers available here and the patterns behind them

Documented case histories