{
  "record": {
    "id": "5b1a989c-3cc2-4967-9316-46967264a146",
    "code": "PAI-5B1A989C",
    "force": "Departament de Justícia i Qualitat Democràtica (Generalitat de Catalunya), prison services",
    "country": "Spain",
    "country_code": "es",
    "is_country": true,
    "region": "Catalonia",
    "technology": "RisCanvi prison risk assessment",
    "technology_slug": "offender-risk-assessment",
    "vendor": "",
    "vendor_slug": "",
    "programme": "",
    "status": "operational",
    "stage": "in-use",
    "evidence": "journalism",
    "evidence_is_claim": false,
    "since": "",
    "legal_basis": "",
    "oversight": "External audits: an adversarial audit by Eticas and an audit by Dribia using Justice Department data, both reported by Euronews in August 2024; the Justice Department also cites an external audit it commissioned in 2023 (Ara). The sources do not say whether the Dribia audit and the 2023 audit are the same.",
    "notes": "RisCanvi is the risk-assessment protocol used in Catalan prisons to estimate the likelihood that a prisoner will reoffend or commit violence; a screening version with 10 risk factors and a full version with 43 exist. Ara (22 February 2025) reports that the Justice Department commissioned an external audit in 2023 which concluded that the tool has \"una capacidad predictiva robusta\", that government sources put its accuracy at about 75 to 80 per cent, and that a recalibration was expected to be in operation during 2025. The same article reports a study by Antonio Andrés-Pueyo, one of RisCanvi's creators, and David Férez-Mangas, covering incidents from 2017 to 2022, which found that 5,785 of 7,302 prisoners involved in violent incidents (79.22 per cent) had no valid assessment at the time; assessments are valid for six months.\n\nThe Generalitat's 2026 budget programme reports (programme 213, prison services and alternative penal measures) use as performance indicators the percentage of classified prisoners, and of pre-trial prisoners, with a current RisCanvi assessment (\"amb l'avaluació de Riscanvi vigent\").\n\nAlgorithmWatch (25 May 2021) describes RisCanvi as a deterministic statistical model rather than machine learning, introduced about 2010, with justice professionals making the final decision. Euronews (8 August 2024) reported an adversarial audit by Eticas that found \"substantial deficiencies\" in its reliability, and a separate audit by Dribia that found no serious discriminatory bias by age, sex or nationality but found that the tool underestimated some risks, including self-inflicted violence among young people and women. Ara (20 June 2025) reports that, in its response to a compensation claim by the family of a cook killed by a prisoner at Mas d'Enric prison in 2024, the Justice Department concluded that RisCanvi had been used in an \"incorrecta y deficitaria\" way.",
    "source": "Ara, 22 February 2025",
    "source_url": "https://es.ara.cat/sociedad/justicia/estudio-alerta-80-presos-implicados-incidentes-violentos-no-tenian-valoracion-riesgo_1_5293742.amp.html",
    "last_checked": "2026-11-06",
    "record_url": "https://www.policeainews.com/tracker/record/5b1a989c-3cc2-4967-9316-46967264a146"
  },
  "licence": "https://www.policeainews.com/data#licence"
}