{
  "schema": 1,
  "version": "1.0",
  "published_at": "2026-10-05",
  "title": "ARGOS — AI Evaluation & Continuous Assurance",
  "status": "VOCE public continuous-assurance framework; not a certification scheme",
  "purpose": "Keep institutional AI and agent systems within an evidence-backed operating envelope as models, prompts, tools, permissions, data, suppliers and human conditions change.",
  "assurance_cycle": [
    {
      "id": "baseline",
      "order": 1,
      "name": "Baseline",
      "question": "What exact model, prompt, tool, permission, data and policy configuration is in production?",
      "evidence": [
        "Version manifest",
        "Agent/tool inventory",
        "Policy/configuration snapshot"
      ],
      "i18n": {
        "fr": {
          "name": "Référence initiale",
          "question": "Quel modèle, prompt, outil, jeu de permissions, donnée et politique exacts sont en production?"
        },
        "it": {
          "name": "Baseline",
          "question": "Quali modello, prompt, strumento, permessi, dati e policy esatti sono in produzione?"
        }
      }
    },
    {
      "id": "observe",
      "order": 2,
      "name": "Observe",
      "question": "What is the deployed system actually doing across tasks, tools, costs, errors, approvals and human interventions?",
      "evidence": [
        "Runtime traces",
        "Task outcomes",
        "Approval/override logs",
        "Cost telemetry"
      ],
      "i18n": {
        "fr": {
          "name": "Observer",
          "question": "Que fait réellement le système déployé sur les tâches, outils, coûts, erreurs, validations et interventions humaines?"
        },
        "it": {
          "name": "Osservare",
          "question": "Che cosa fa realmente il sistema implementato tra task, strumenti, costi, errori, approvazioni e interventi umani?"
        }
      }
    },
    {
      "id": "detect",
      "order": 3,
      "name": "Detect change",
      "question": "Which material change, drift, incident or supplier event invalidates the previous assurance state?",
      "evidence": [
        "Change event",
        "Drift signal",
        "Incident record",
        "Supplier notice"
      ],
      "i18n": {
        "fr": {
          "name": "Détecter le changement",
          "question": "Quel changement matériel, dérive, incident ou événement fournisseur invalide l’état d’assurance précédent?"
        },
        "it": {
          "name": "Rilevare il cambiamento",
          "question": "Quale modifica sostanziale, drift, incidente o evento del fornitore invalida lo stato di assurance precedente?"
        }
      }
    },
    {
      "id": "reevaluate",
      "order": 4,
      "name": "Re-evaluate",
      "question": "Which evaluations, security tests, conformance controls and cognitive controls must be repeated after the change?",
      "evidence": [
        "Triggered test plan",
        "Regression suite",
        "Control selection rationale"
      ],
      "i18n": {
        "fr": {
          "name": "Réévaluer",
          "question": "Quelles évaluations, tests de sécurité, contrôles Conformance et contrôles cognitifs doivent être répétés après le changement?"
        },
        "it": {
          "name": "Rivalutare",
          "question": "Quali valutazioni, test di sicurezza, controlli Conformance e controlli cognitivi devono essere ripetuti dopo il cambiamento?"
        }
      }
    },
    {
      "id": "compare",
      "order": 5,
      "name": "Compare",
      "question": "How does the new system state differ from the accepted baseline on quality, safety, authority, cost and human oversight?",
      "evidence": [
        "Baseline delta",
        "Performance delta",
        "Risk delta",
        "Cost delta"
      ],
      "i18n": {
        "fr": {
          "name": "Comparer",
          "question": "Comment le nouvel état diffère-t-il de la référence acceptée sur qualité, sécurité, autorité, coût et supervision humaine?"
        },
        "it": {
          "name": "Confrontare",
          "question": "In che modo il nuovo stato differisce dalla baseline accettata su qualità, sicurezza, autorità, costo e supervisione umana?"
        }
      }
    },
    {
      "id": "decide",
      "order": 6,
      "name": "Decide",
      "question": "Can the changed system remain in service, require restrictions, rollback, additional approval or withdrawal?",
      "evidence": [
        "Decision record",
        "Accountable owner",
        "Conditions/limits",
        "Rollback decision"
      ],
      "i18n": {
        "fr": {
          "name": "Décider",
          "question": "Le système modifié peut-il rester en service, doit-il être restreint, revenir en arrière, exiger une validation supplémentaire ou être retiré?"
        },
        "it": {
          "name": "Decidere",
          "question": "Il sistema modificato può restare in servizio, deve essere limitato, riportato indietro, richiedere un’approvazione aggiuntiva o essere ritirato?"
        }
      }
    },
    {
      "id": "retain",
      "order": 7,
      "name": "Retain evidence",
      "question": "Can the institution reconstruct why the system was considered acceptable at that point in time?",
      "evidence": [
        "Dated evidence pack",
        "Reviewer record",
        "Versioned test outputs",
        "Decision history"
      ],
      "i18n": {
        "fr": {
          "name": "Conserver les preuves",
          "question": "L’institution peut-elle reconstruire pourquoi le système était considéré comme acceptable à cet instant?"
        },
        "it": {
          "name": "Conservare le evidenze",
          "question": "L’istituzione può ricostruire perché il sistema era considerato accettabile in quel momento?"
        }
      }
    },
    {
      "id": "reopen",
      "order": 8,
      "name": "Re-open",
      "question": "What future event automatically reopens assurance rather than waiting for a periodic audit?",
      "evidence": [
        "Trigger rules",
        "Monitoring rule",
        "Scheduled review",
        "Event-driven control"
      ],
      "i18n": {
        "fr": {
          "name": "Rouvrir",
          "question": "Quel événement futur rouvre automatiquement l’assurance au lieu d’attendre un audit périodique?"
        },
        "it": {
          "name": "Riaprire",
          "question": "Quale evento futuro riapre automaticamente l’assurance invece di attendere un audit periodico?"
        }
      }
    }
  ],
  "triggers": [
    {
      "id": "ARG-T01",
      "name": "Foundation-model or major model-version change",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Changement de modèle de fondation ou de version majeure"
        },
        "it": {
          "name": "Cambio di foundation model o di versione principale"
        }
      }
    },
    {
      "id": "ARG-T02",
      "name": "Prompt, system instruction or orchestration-policy change",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Changement de prompt, d’instruction système ou de politique d’orchestration"
        },
        "it": {
          "name": "Modifica di prompt, istruzione di sistema o policy di orchestrazione"
        }
      }
    },
    {
      "id": "ARG-T03",
      "name": "New tool, API, MCP server or A2A peer",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Nouvel outil, API, serveur MCP ou pair A2A"
        },
        "it": {
          "name": "Nuovo strumento, API, server MCP o peer A2A"
        }
      }
    },
    {
      "id": "ARG-T04",
      "name": "Permission, transaction-limit or approval-threshold change",
      "severity": "critical",
      "i18n": {
        "fr": {
          "name": "Changement de permissions, limites transactionnelles ou seuils d’approbation"
        },
        "it": {
          "name": "Modifica di permessi, limiti transazionali o soglie di approvazione"
        }
      }
    },
    {
      "id": "ARG-T05",
      "name": "New data source, RAG index, memory mechanism or processing region",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Nouvelle source de données, index RAG, mécanisme de mémoire ou région de traitement"
        },
        "it": {
          "name": "Nuova fonte dati, indice RAG, meccanismo di memoria o regione di elaborazione"
        }
      }
    },
    {
      "id": "ARG-T06",
      "name": "Material degradation in task quality or error distribution",
      "severity": "critical",
      "i18n": {
        "fr": {
          "name": "Dégradation matérielle de la qualité des tâches ou de la distribution des erreurs"
        },
        "it": {
          "name": "Degrado sostanziale della qualità dei task o della distribuzione degli errori"
        }
      }
    },
    {
      "id": "ARG-T07",
      "name": "Security incident, prompt-injection event or material near miss",
      "severity": "critical",
      "i18n": {
        "fr": {
          "name": "Incident de sécurité, injection de prompt ou quasi-incident matériel"
        },
        "it": {
          "name": "Incidente di sicurezza, prompt injection o near miss rilevante"
        }
      }
    },
    {
      "id": "ARG-T08",
      "name": "Supplier, licence, pricing or availability change",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Changement de fournisseur, licence, prix ou disponibilité"
        },
        "it": {
          "name": "Cambio di fornitore, licenza, prezzo o disponibilità"
        }
      }
    },
    {
      "id": "ARG-T09",
      "name": "Regulatory role, classification or sector obligation change",
      "severity": "critical",
      "i18n": {
        "fr": {
          "name": "Changement de rôle réglementaire, classification ou obligation sectorielle"
        },
        "it": {
          "name": "Cambio di ruolo regolatorio, classificazione o obbligo settoriale"
        }
      }
    },
    {
      "id": "ARG-T10",
      "name": "Material increase in cost, retries or resource consumption",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Hausse matérielle des coûts, retries ou consommation de ressources"
        },
        "it": {
          "name": "Aumento rilevante di costi, retry o consumo di risorse"
        }
      }
    },
    {
      "id": "ARG-T11",
      "name": "Rise in human overrides, missed alerts, decision fatigue or cognitive-control failures",
      "severity": "critical",
      "i18n": {
        "fr": {
          "name": "Hausse des overrides humains, alertes manquées, fatigue décisionnelle ou échecs des contrôles cognitifs"
        },
        "it": {
          "name": "Aumento di override umani, avvisi mancati, affaticamento decisionale o fallimenti dei controlli cognitivi"
        }
      }
    },
    {
      "id": "ARG-T12",
      "name": "Significant organisational or staffing change affecting human oversight",
      "severity": "material",
      "i18n": {
        "fr": {
          "name": "Changement organisationnel ou d’effectifs affectant matériellement la supervision humaine"
        },
        "it": {
          "name": "Cambiamento organizzativo o di personale che incide materialmente sulla supervisione umana"
        }
      }
    }
  ],
  "evaluation_planes": [
    {
      "id": "capability",
      "name": "Capability & task performance",
      "measures": [
        "Task success",
        "Failure modes",
        "Regression",
        "Edge cases"
      ],
      "i18n": {
        "fr": {
          "name": "Capacité & performance des tâches"
        },
        "it": {
          "name": "Capacità e performance dei task"
        }
      }
    },
    {
      "id": "authority",
      "name": "Authority & permissions",
      "measures": [
        "Tool access",
        "Data scope",
        "Transaction limits",
        "Approval bypass"
      ],
      "i18n": {
        "fr": {
          "name": "Autorité & permissions"
        },
        "it": {
          "name": "Autorità e permessi"
        }
      }
    },
    {
      "id": "security",
      "name": "Security & abuse",
      "measures": [
        "Prompt injection",
        "Tool manipulation",
        "Credential misuse",
        "Cross-scope access"
      ],
      "i18n": {
        "fr": {
          "name": "Sécurité & abus"
        },
        "it": {
          "name": "Sicurezza e abuso"
        }
      }
    },
    {
      "id": "observability",
      "name": "Observability & audit",
      "measures": [
        "Trace completeness",
        "Version traceability",
        "Approval evidence",
        "Incident reconstruction"
      ],
      "i18n": {
        "fr": {
          "name": "Observabilité & audit"
        },
        "it": {
          "name": "Osservabilità e audit"
        }
      }
    },
    {
      "id": "economics",
      "name": "Economics & concentration",
      "measures": [
        "Cost per completed task",
        "Retry spend",
        "Supplier concentration",
        "Exit readiness"
      ],
      "i18n": {
        "fr": {
          "name": "Économie & concentration"
        },
        "it": {
          "name": "Economia e concentrazione"
        }
      }
    },
    {
      "id": "cognitive",
      "name": "Human cognitive assurance",
      "measures": [
        "Alert burden",
        "Verification quality",
        "Takeover comprehension",
        "Decision fatigue"
      ],
      "i18n": {
        "fr": {
          "name": "Assurance cognitive humaine"
        },
        "it": {
          "name": "Assurance cognitiva umana"
        }
      }
    }
  ],
  "assurance_states": [
    {
      "id": "open",
      "name": "Open",
      "definition": "A material trigger has occurred and the previous assurance conclusion cannot be relied upon without re-evaluation.",
      "i18n": {
        "fr": {
          "name": "Ouverte",
          "definition": "Un déclencheur matériel est survenu et la conclusion précédente ne peut plus être utilisée sans réévaluation."
        },
        "it": {
          "name": "Aperta",
          "definition": "Si è verificato un trigger materiale e la precedente conclusione di assurance non può più essere utilizzata senza rivalutazione."
        }
      }
    },
    {
      "id": "restricted",
      "name": "Restricted",
      "definition": "The system may operate only under explicit temporary limits while evidence gaps are closed.",
      "i18n": {
        "fr": {
          "name": "Restreinte",
          "definition": "Le système ne peut fonctionner que sous limites temporaires explicites jusqu’à fermeture des lacunes de preuve."
        },
        "it": {
          "name": "Limitata",
          "definition": "Il sistema può operare solo sotto limiti temporanei espliciti finché le lacune di evidenza non vengono chiuse."
        }
      }
    },
    {
      "id": "accepted",
      "name": "Accepted",
      "definition": "The defined evaluation and control set supports continued operation within the stated scope and conditions.",
      "i18n": {
        "fr": {
          "name": "Acceptée",
          "definition": "L’ensemble défini d’évaluations et de contrôles soutient la poursuite de l’exploitation dans le périmètre et les conditions déclarés."
        },
        "it": {
          "name": "Accettata",
          "definition": "L’insieme definito di valutazioni e controlli sostiene la prosecuzione dell’operatività nello scope e nelle condizioni dichiarate."
        }
      }
    },
    {
      "id": "withdrawn",
      "name": "Withdrawn",
      "definition": "The system or capability is removed from service because the evidence no longer supports continued operation.",
      "i18n": {
        "fr": {
          "name": "Retirée",
          "definition": "Le système ou la capacité est retiré du service parce que les preuves ne soutiennent plus son maintien en exploitation."
        },
        "it": {
          "name": "Ritirata",
          "definition": "Il sistema o la capacità viene rimosso dal servizio perché le evidenze non supportano più la prosecuzione dell’operatività."
        }
      }
    }
  ],
  "integrations": [
    {
      "id": "infrastructure",
      "name": "Agent Infrastructure",
      "url": "https://voce.life/agent-infrastructure",
      "role": "Defines the technical architecture and control surfaces."
    },
    {
      "id": "conformance",
      "name": "Agent Conformance",
      "url": "https://voce.life/agent-conformance",
      "role": "Defines the evidence-based operational controls."
    },
    {
      "id": "cognitive",
      "name": "Human Cognitive Assurance",
      "url": "https://voce.life/human-cognitive-assurance",
      "role": "Defines the human cognitive and neuropsychology assurance layer."
    }
  ]
}