{ "vercy": "1.0-draft", "publication": { "status": "research-draft", "adjudicationStatus": "unreviewed", "publishableCanonical": false, "generatedAt": "2026-09-12T09:00:48Z", "providers": [ "Codex" ], "breadth": "recalled by Codex without web access - no source was read", "missingProviders": [], "pass": 2, "cost": { "grok": { "seconds": 27.7, "error": "Reading additional input from stdin...\nOpenAI Codex v0.153.4\n--------\nworkdir: R:\\02_PROJECTS\\02_Meta_Models_Platforms\\Ver.cy\\current\\thing-registry-backlog\nmodel: gpt-6-astra\nprovider: openai\napproval: never\nsandbox: read-only\nreasoning effort: none\nreasoning summaries: none\nsession id: 01a094d7-ffae-77f2-8253-403452f8d08b\n--------\nuser\nDescribe what is already known about one registered thing. Answer as JSON only, no prose around it.\n\nThing: data anonymization\nSense to describe: (none recorded)\nDomain code: ACT.ACT\nAlso known as: (none)\n\n\nContext for this batch of 1001 things:\n# Batch 043: 1", "usd": 0.0, "recall": true }, "codex": { "seconds": 69.9, "error": "Reading additional input from stdin...\nOpenAI Codex v0.153.4\n--------\nworkdir: R:\\02_PROJECTS\\02_Meta_Models_Platforms\\Ver.cy\\current\\thing-registry-backlog\nmodel: gpt-6-astra\nprovider: openai\napproval: never\nsandbox: read-only\nreasoning effort: none\nreasoning summaries: none\nsession id: 01a094d7-ffa5-7ce3-9c28-ea173870e843\n--------\nuser\nYou are drafting a Vercy meta-model for one registered thing. Answer as JSON only, no prose around it.\n\nThing: data anonymization\nRegistry id: vr.tr.data-anonymization\nPlane / domain: ACT / ACT.ACT\nRegistry definition: (none recorded)\nNames folded into this en" } }, "enrichment": { "pass": 3, "engine": "claude", "producedAt": "2026-10-06T12:00:00Z", "ingestedAt": "2026-10-06T08:46:12+00:00", "fields": [ "agent_conduct", "ethics", "owners", "standards_and_regulation" ] }, "gate": { "passed": true, "version": "1.0", "at": "2026-10-07T10:10:26+00:00", "evidence": [ "completeness 1.00", "sense: Wikidata description", "2 live sources", "review by codex: pass" ] } }, "metaModel": { "id": "THING-Q17006654", "registryId": "vr.tr.data-anonymization", "name": "data anonymization", "version": "0.1.0-research.1", "entryKind": "thing", "family": "Thing Registry", "domain": [ "ACT.ACT" ], "status": "research-draft" }, "canonicalUrl": "https://ver.cy/models/thing/q17006654/", "model": { "registry_id": "vr.tr.data-anonymization", "name": "data anonymization", "purpose": "Enable an AI agent to assess a data anonymization activity, record the evidence supporting its claimed protection, and decide whether its outputs are suitable for a specified use and access setting.", "definition": "Data anonymization is the process of modifying or releasing data so that individuals cannot reasonably be identified from it, including through linkage with other available information, with adequacy depending on the disclosure context and assumed attacker capabilities.", "scope_statement": "This model owns the activity of transforming data and evaluating its outputs against an explicit anonymity criterion, including identification threats, utility requirements, release conditions, and reassessment triggers; it does not assume that removing identifiers establishes anonymity.", "in_scope": [ "The input data, intended use, affected individuals or groups, and proposed output.", "The anonymity definition, assessment unit, threat assumptions, and acceptance criteria adopted for the activity.", "Selection and application of transformations that address direct and indirect identification.", "Evaluation of identification risk, information loss, and evidence supporting the claimed protection.", "Output release conditions, cumulative disclosure, and reassessment when assumptions change." ], "out_of_scope": [ "General information security and access administration beyond their role in the assessed release setting.", "Pseudonymization and identity-key management as standalone activities.", "Determining the authority or legal basis for collecting and processing the original data.", "General data cleaning, integration, and quality management unrelated to anonymization effects.", "Downstream analysis and model development beyond evaluating utility and disclosure through their outputs." ], "distinguishing_features": [ "The activity must specify what identification capability it intends to prevent; a transformation justified only by confidentiality or formatting is insufficient to characterize anonymization.", "Replacing names with stable tokens does not establish anonymity: the assessment must examine retained mappings and linkage through remaining attributes.", "Encryption protects data through restricted access to a decryption capability; an anonymization claim requires assessment of what recipients can infer from the accessible output.", "Masking or redacting direct identifiers is assessed alongside combinations of indirect identifiers, free text, metadata, and external linkage opportunities.", "Aggregation or synthetic generation identifies a technique, while an anonymization claim additionally requires evidence under stated assumptions." ], "characteristics": [ { "name": "Adopted anonymity criterion", "kind": "category", "unit_or_values": "Named definition or framework, version, interpretation, and acceptance rule", "why_it_matters": "Different criteria may support different conclusions about the same output." }, { "name": "Protection unit", "kind": "category", "unit_or_values": "Person, household, event, organization, or explicitly defined group", "why_it_matters": "Protecting a record or event may leave other records concerning the same person linkable." }, { "name": "Input identifiability profile", "kind": "relation", "unit_or_values": "Links to direct identifiers, quasi-identifiers, sensitive attributes, unstructured content, and metadata", "why_it_matters": "The transformation must address identification paths across the actual data representation." }, { "name": "Recipient and adversary capabilities", "kind": "relation", "unit_or_values": "Specified actors, auxiliary data, access rights, resources, and prior knowledge", "why_it_matters": "Risk estimates depend on whose capabilities and knowledge are considered." }, { "name": "Transformation configuration", "kind": "relation", "unit_or_values": "Ordered techniques, parameters, implementation versions, and retained mappings", "why_it_matters": "The evaluated output must be traceable to the process that produced it." }, { "name": "Disclosure assessment results", "kind": "measurement", "unit_or_values": "Named metrics with their units, uncertainty, evaluation population, and attack conditions", "why_it_matters": "A risk value cannot be interpreted independently of its measurement method and assumptions." }, { "name": "Task-specific utility loss", "kind": "measurement", "unit_or_values": "Change in specified analytical metrics relative to an agreed baseline", "why_it_matters": "An output may reduce identification risk while no longer supporting its intended use." }, { "name": "Release setting", "kind": "category", "unit_or_values": "Public download, restricted recipient transfer, controlled environment, query interface, or specified alternative", "why_it_matters": "Access and interaction conditions determine which disclosure paths require assessment." }, { "name": "Assessment state", "kind": "state", "unit_or_values": "Unassessed, under evaluation, criteria met for stated setting, criteria not met, or reassessment required", "why_it_matters": "It separates completion of a transformation from evidence supporting a release decision." } ], "affordances": [ "Identify direct, indirect, longitudinal, and cross-dataset identification paths.", "Compare candidate transformations against explicit protection and utility criteria.", "Evaluate a candidate output using documented attacks, metrics, and applicable formal guarantees.", "Recommend release, revision, or withholding based on evidence for the specified setting.", "Trace a released output to its input version, transformation configuration, and assessment.", "Trigger reassessment when new releases, auxiliary data, recipients, or vulnerabilities change the original assumptions." ], "agent_conduct": { "may": [ "Remove or generalize identifiers in data the agent is authorized to process, following the controller's policy.", "Assess and report the risk that released data could be re-identified.", "Explain the difference between anonymized and pseudonymized data." ], "must_not": [ "Call data anonymous when people can still reasonably be identified from it.", "Try to re-identify people in data released as anonymous.", "Release a dataset publicly without the data controller's approval.", "Link anonymized data with other sources in a way that defeats its protection." ], "requires_human": [ "Approving a dataset as anonymous for publication or sharing." ] }, "ethics": { "considerations": [ "Failed anonymization exposes people who were promised privacy.", "Too much distortion can make data misleading for research that serves the public." ], "affected_parties": [ "Data subjects", "Researchers", "Data controllers" ] }, "owners": { "steward": "The data controller answers for each dataset; data protection authorities oversee it.", "master_systems": [] }, "standards_and_regulation": [ "General Data Protection Regulation, European Union, which places truly anonymous data outside its scope.", "ISO/IEC 20889, Privacy enhancing data de-identification terminology and classification of techniques, ISO and IEC." ], "what_it_is_for": [ "Data anonymization is used when personal data must support research, analysis or publication while limiting the identification of individuals.", "It is assessed in relation to the intended recipients, available external information and conditions of access." ], "failure_modes_and_hazards": [ "Linkage with other information can reveal identities despite the removal of direct identifiers.", "Excessive modification can undermine analytical accuracy and produce misleading conclusions.", "Unsupported claims of anonymity can expose individuals to privacy loss and organizations to legal liability.", "Changes in available data or analytical capabilities can weaken protection over time." ], "facets_not_applicable": { "recognition": "Data anonymization is a procedure whose effectiveness cannot be established from visual appearance." } }, "sources": [ { "id": "wikidata", "title": "Wikidata item Q17006654: data anonymization", "url": "https://www.wikidata.org/wiki/Q17006654", "url_status": "live", "what_it_supports": "identity and sense of the item", "checked_by": "wikidata-api" }, { "id": "wikipedia-en", "title": "Wikipedia: Data anonymization", "url": "https://en.wikipedia.org/wiki/Data_anonymization", "url_status": "live", "what_it_supports": "general description of the item", "checked_by": "wikidata-api sitelink" } ], "structure": { "bundles": [ { "id": "anonymity-claim-and-boundaries", "name": "Anonymity claim and boundaries", "description": "Defines the meaning, subject, and limits of the protection being claimed.", "rationale": "An agent cannot evaluate anonymization without knowing which criterion applies and precisely what the claim covers.", "layers": [ { "id": "definitions-and-acceptance", "name": "Definitions and acceptance", "description": "Records competing interpretations and the criterion selected for this activity.", "findings": [ { "id": "adopted-anonymity-definition", "name": "Adopted anonymity definition", "description": "Record whose definition is adopted, how it is operationalized, and which alternative interpretations remain relevant.", "questions": [ { "text": "Which authority, research framework, or organizational policy supplies the anonymity definition, and which version was adopted?", "kind": "provenance", "id": "adopted-anonymity-definition-q01" }, { "text": "What evidence would satisfy that definition, and how would another relevant interpretation change the conclusion?", "kind": "definition", "id": "adopted-anonymity-definition-q02" } ] } ] }, { "id": "protected-unit-and-output", "name": "Protected unit and output", "description": "Bounds the protected entity and the complete artifact or interface being assessed.", "findings": [ { "id": "claim-coverage", "name": "Claim coverage", "description": "Record whether protection concerns people, events, households, or groups, and which outputs and supporting files the claim includes.", "questions": [ { "text": "Does the protection unit encompass all records about one person, including repeated events and linked tables?", "kind": "boundary", "id": "claim-coverage-q01" }, { "text": "Which tables, metadata, documentation, generated examples, and query responses are included in the assessed output?", "kind": "boundary", "id": "claim-coverage-q02" } ] } ] } ] }, { "id": "identification-threats", "name": "Identification threats", "description": "Models how the proposed output could expose identities, relationships, membership, or sensitive attributes.", "rationale": "Identifier removal alone does not describe the identification opportunities present in the data and its release context.", "layers": [ { "id": "identifying-data-patterns", "name": "Identifying data patterns", "description": "Examines direct identifiers and distinctive combinations across structured and unstructured data.", "findings": [ { "id": "identification-path-inventory", "name": "Identification path inventory", "description": "Record candidate identification paths involving rare combinations, trajectories, relationships, text, and embedded metadata.", "questions": [ { "text": "Which attribute combinations, timestamps, locations, or relationship patterns could distinguish a protected unit?", "kind": "measurement", "id": "identification-path-inventory-q01" }, { "text": "Where might identifiers survive in free text, images, filenames, document properties, or linked records?", "kind": "boundary", "id": "identification-path-inventory-q02" } ] } ] }, { "id": "adversaries-and-auxiliary-information", "name": "Adversaries and auxiliary information", "description": "Specifies plausible recipients, attackers, and external knowledge used in the assessment.", "findings": [ { "id": "attack-capability-assumptions", "name": "Attack capability assumptions", "description": "Record the capabilities and auxiliary information against which linkage and inference are evaluated.", "questions": [ { "text": "Which external datasets, personal knowledge, retained originals, or lookup mappings could each assessed actor access?", "kind": "provenance", "id": "attack-capability-assumptions-q01" }, { "text": "Which singling-out, linkage, membership, and attribute-inference attacks fall within the adopted criterion and threat model?", "kind": "boundary", "id": "attack-capability-assumptions-q02" } ] } ] } ] }, { "id": "transformation-design-and-execution", "name": "Transformation design and execution", "description": "Connects identified threats to the transformations and execution details that produce the candidate output.", "rationale": "Technique names do not establish what protection a particular configuration achieves.", "layers": [ { "id": "techniques-and-parameters", "name": "Techniques and parameters", "description": "Records why transformations were selected and how they are configured.", "findings": [ { "id": "threat-to-transformation-mapping", "name": "Threat-to-transformation mapping", "description": "Associate suppression, generalization, perturbation, aggregation, synthetic generation, or other techniques with the identification paths they address.", "questions": [ { "text": "Which transformation addresses each identified disclosure path, and which paths remain untreated?", "kind": "action", "id": "threat-to-transformation-mapping-q01" }, { "text": "What parameters govern the transformation, including any applicable privacy budget, clipping bounds, grouping rules, or suppression thresholds?", "kind": "measurement", "id": "threat-to-transformation-mapping-q02" } ] } ] }, { "id": "execution-and-retained-linkage", "name": "Execution and retained linkage", "description": "Tracks the executed process and artifacts that could preserve or recreate identity links.", "findings": [ { "id": "output-lineage-and-linkage-artifacts", "name": "Output lineage and linkage artifacts", "description": "Record the exact input and implementation versions, transformation order, and custody of originals, mappings, and intermediate outputs.", "questions": [ { "text": "Which input snapshot, code version, parameter set, and execution record produced the candidate being evaluated?", "kind": "provenance", "id": "output-lineage-and-linkage-artifacts-q01" }, { "text": "Who retains originals, token mappings, intermediate files, or other reconstruction aids, and how does that affect the anonymity claim?", "kind": "boundary", "id": "output-lineage-and-linkage-artifacts-q02" } ] } ] } ] }, { "id": "protection-and-utility-evidence", "name": "Protection and utility evidence", "description": "Records what evaluation establishes about disclosure and fitness for the intended use.", "rationale": "A usable decision requires both protection evidence and an account of the information lost or distorted.", "layers": [ { "id": "disclosure-evaluation", "name": "Disclosure evaluation", "description": "Distinguishes empirical attack results, structural metrics, and formal guarantees.", "findings": [ { "id": "protection-evidence-and-limitations", "name": "Protection evidence and limitations", "description": "Record evaluation methods, results, assumptions, and the limits of what each result supports.", "questions": [ { "text": "Which attacks and metrics were applied, with what success rates, uncertainty, baselines, and coverage of rare or vulnerable subgroups?", "kind": "measurement", "id": "protection-evidence-and-limitations-q01" }, { "text": "If a formal guarantee is claimed, what mechanism, protection unit, assumptions, and composition accounting connect it to the actual output?", "kind": "provenance", "id": "protection-evidence-and-limitations-q02" } ] } ] }, { "id": "analytical-utility", "name": "Analytical utility", "description": "Assesses whether transformation preserves the information required for named tasks.", "findings": [ { "id": "utility-loss-and-distributional-effects", "name": "Utility loss and distributional effects", "description": "Record task performance changes and distortions affecting small groups, unusual cases, and relationships between variables.", "questions": [ { "text": "Which intended analyses were evaluated, and how much did their results change against the agreed baseline?", "kind": "measurement", "id": "utility-loss-and-distributional-effects-q01" }, { "text": "Which populations or phenomena become underrepresented or misleading, and should this require another transformation or a use restriction?", "kind": "action", "id": "utility-loss-and-distributional-effects-q02" } ] } ] } ] }, { "id": "release-and-reassessment", "name": "Release and reassessment", "description": "Connects the evaluated candidate to its actual release conditions and subsequent disclosure context.", "rationale": "Evidence applies to a particular output and setting, while additional releases and new information can change the assessment.", "layers": [ { "id": "release-decision", "name": "Release decision", "description": "Records the evidence and conditions supporting release, revision, or withholding.", "findings": [ { "id": "setting-specific-disposition", "name": "Setting-specific disposition", "description": "Bind the decision to the evaluated artifact, recipients, access mechanism, and unresolved limitations.", "questions": [ { "text": "Does the candidate meet the adopted protection and utility criteria for the actual recipients and access setting?", "kind": "measurement", "id": "setting-specific-disposition-q01" }, { "text": "Should this version be released, revised, or withheld, and which evidence and unresolved limitations support that decision?", "kind": "action", "id": "setting-specific-disposition-q02" } ] } ] }, { "id": "cumulative-disclosure-and-change", "name": "Cumulative disclosure and change", "description": "Tracks interactions with earlier outputs and changes that undermine assessment assumptions.", "findings": [ { "id": "composition-and-reassessment-triggers", "name": "Composition and reassessment triggers", "description": "Record related releases, repeated queries, new linkage opportunities, and practical responses when protection becomes doubtful.", "questions": [ { "text": "How do previous releases, overlapping populations, differencing opportunities, or repeated queries affect the combined disclosure assessment?", "kind": "measurement", "id": "composition-and-reassessment-triggers-q01" }, { "text": "Which new auxiliary data, demonstrated attacks, or recipient changes trigger reassessment, and what response remains possible after distribution?", "kind": "action", "id": "composition-and-reassessment-triggers-q02" } ] } ] } ] }, { "id": "evidence-and-external-alignment", "name": "Evidence and external alignment", "description": "What the world already says about this thing, gathered so the model can be checked against it.", "rationale": "A model that cannot be lined up against existing standards, identifiers and practice cannot be adopted by anyone who already uses them.", "layers": [ { "id": "reported-evidence", "name": "Reported evidence", "description": "Findings from the breadth pass, kept separate from the structural claims.", "findings": [ { "id": "evidence-confidence-notes", "name": "Check these first", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ "This is recalled knowledge, not source-verified research; check the current editions and applicability of the named instruments.", "Terminology and acceptable identification risk vary across legal, statistical and technical communities; no single numerical threshold establishes anonymity everywhere.", "The listed techniques are candidate approaches, not guarantees: adequacy requires evaluating auxiliary information, release conditions, attacker capabilities and utility." ], "questions": [ { "id": "evidence-confidence-notes-q01", "text": "Which of these check these first hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-kinds", "name": "Kinds and varieties", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ "Generalization of identifying attributes into broader categories", "Suppression of identifying values or records", "Aggregation into group-level statistics", "Perturbation through noise addition or value modification", "Release of synthetic data assessed for disclosure risk", "Differentially private releases using a formal privacy mechanism" ], "questions": [ { "id": "evidence-kinds-q01", "text": "Which of these kinds and varieties hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-standards-and-regulation", "name": "Standards and regulation", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ "ISO/IEC 20889, issued by ISO and IEC, addresses privacy-enhancing data de-identification terminology and classification of techniques.", "The European Union's General Data Protection Regulation, particularly Recital 26, distinguishes anonymous information from identifiable personal data.", "The US HIPAA Privacy Rule, administered by the Department of Health and Human Services, provides Safe Harbor and Expert Determination routes for de-identifying protected health information." ], "questions": [ { "id": "evidence-standards-and-regulation-q01", "text": "Which of these standards and regulation hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-real-world-use", "name": "Real-world use", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ "Sharing research datasets while reducing participant identification risk", "Publishing official statistics and public-use microdata", "Preparing health data for secondary analysis", "Producing customer and service analytics", "Providing development and testing datasets derived from operational records" ], "questions": [ { "id": "evidence-real-world-use-q01", "text": "Which of these real-world use hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-measurements", "name": "Typical measurements", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ { "quantity": "Minimum equivalence-class size in k-anonymity", "typical_range": "No universal target; k is a positive integer selected for the release context", "unit": "records" }, { "quantity": "Differential privacy parameter epsilon", "typical_range": "Nonnegative; no universally accepted operational range", "unit": "dimensionless" }, { "quantity": "Fraction of records or values suppressed", "typical_range": "0-100; acceptable levels depend on analytical requirements", "unit": "percent" } ], "questions": [ { "id": "evidence-measurements-q01", "text": "Which of these typical measurements hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-failure-modes-and-hazards", "name": "Failure modes and hazards", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ "Removing direct identifiers while leaving combinations of quasi-identifiers that permit identification", "Linking released data with external datasets to identify individuals", "Inferring sensitive attributes despite preventing direct identification", "Combining repeated releases or queries to weaken privacy protection", "Destroying analytical usefulness or introducing bias through excessive transformation" ], "questions": [ { "id": "evidence-failure-modes-and-hazards-q01", "text": "Which of these failure modes and hazards hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-regional-variation", "name": "Regional variation", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ "EU data-protection terminology distinguishes anonymization from pseudonymization; pseudonymized personal data remain personal data.", "US HIPAA de-identification is a sector-specific framework whose criteria should not be treated as interchangeable with EU anonymization criteria." ], "questions": [ { "id": "evidence-regional-variation-q01", "text": "Which of these regional variation hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] }, { "id": "evidence-neighbours", "name": "Neighbouring kinds and how to tell them apart", "description": "Recalled without web access and unsourced; every item is a lead to verify.", "evidence": [ { "name": "Pseudonymization", "difference": "Replaces or separates identifiers while allowing attribution using additional information; anonymization aims to prevent reasonable identification in the relevant context." }, { "name": "De-identification", "difference": "Often describes a broader removal or reduction of identifying information; whether it entails anonymization depends on the governing framework." }, { "name": "Encryption", "difference": "Protects data through key-controlled access and is ordinarily reversible with the key; encryption alone does not anonymize data." }, { "name": "Data masking", "difference": "Conceals or substitutes selected values, often for operational use, without necessarily preventing identification through remaining attributes." }, { "name": "Differential privacy", "difference": "Provides a mathematical bound on how an individual's participation affects output probabilities; it is a formal privacy property rather than a synonym for anonymization." }, { "name": "Synthetic data", "difference": "Consists of generated records; generation alone does not establish anonymity because outputs can reproduce or reveal information about source individuals." } ], "questions": [ { "id": "evidence-neighbours-q01", "text": "Which of these neighbouring kinds and how to tell them apart hold for the sense of data anonymization this model covers, and on what evidence?", "kind": "provenance" } ] } ] } ] } ] }, "openQuestions": [ "Which competing technical, organizational, and jurisdiction-specific definitions should this model represent, and where do they yield incompatible anonymity judgments?", "Which assessment methods and acceptance thresholds are defensible for high-dimensional, longitudinal, relational, and unstructured data?", "How should the model represent uncertainty about future auxiliary information and adversary capabilities without implying permanent anonymity?", "When do synthetic outputs or trained models require their own linked models, and what evidence must the anonymization activity retain about their disclosure behavior?", "How should individual anonymity assessments connect to group-level harms that can persist even when identification of individuals is constrained?" ], "statistics": { "bundles": 6, "layers": 11, "findings": 18, "questions": 28 } }