{"schema_version":1,"research_id":"eoa_inverse_innovation_exp06_external_evaluation_20260803","source_assessment_id":"representation_independent_interface_contract__criminology_forensic:P4:v0","cell_id":"representation_independent_interface_contract__criminology_forensic","search_queries":["site:cjars.org administrative criminal justice data longitudinal harmonized records data documentation","site:ojp.gov criminal justice administrative data quality missing data longitudinal research records","site:niem.gov justice domain model arrests charges case disposition standard","criminal justice administrative data event history observation window no event missing data","site:niem.gov justice domain NIEM criminal justice arrest charge court corrections data model","site:niem.github.io justice domain NIEM arrest charge disposition","site:ohdsi.github.io CommonDataModel observation period absence event data","site:w3.org/TR/prov-o generatedAtTime invalidatedAtTime provenance recommendation","site:learn.microsoft.com SQL Server temporal tables FOR SYSTEM_TIME AS OF history official","site:hhs.gov/ohrp regulations 45 CFR 46 secondary research identifiable private information IRB official","site:bls.gov/oes software developers median annual wage May 2025","site:bls.gov/ooh software developers median pay 2025"],"sources":[{"source_id":"s1","title":"Overview","publisher":"Criminal Justice Administrative Records System, University of Michigan","url":"https://cjars.org/overview/","source_class":"OFFICIAL_ORGANIZATION_DATA","publication_date":"n.d.","accessed_at":"2026-08-03","claims_supported":["CJARS is building a nationally integrated repository following individuals through the justice system.","CJARS reports more than three billion records from 30 states and ongoing work on harmonization, linkage, and coverage.","Researchers, data providers, the federal statistical system, named foundations, NSF, the Census Bureau, and the University of Michigan are identifiable stakeholders or funders seeking a uniform data infrastructure."]},{"source_id":"s2","title":"Criminal Justice Administrative Records System (CJARS): Data Documentation, 2023 Q3","publisher":"Criminal Justice Administrative Records System, University of Michigan","url":"https://cjars.org/wp-content/uploads/CJARS_data_docs_2023q3.pdf","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"2023-Q3","accessed_at":"2026-08-03","claims_supported":["Justice data differ substantially across jurisdictions in coding, processing, and storage, creating barriers to multisource research.","CJARS already harmonizes event dates, offenses, dispositions, sentences, and supervision intervals while retaining original source variables.","CJARS performs entity and episode resolution, but its documentation identifies longitudinal episode linkage as a continuing challenge and, in that release, unfinished work.","CJARS uses data-use agreements, protected identifiers, secure research environments, versioned archives, IRB review, and project-level legal, ethical, and scientific review."]},{"source_id":"s3","title":"Longitudinal linkage of administrative data: design principles and the total error framework","publisher":"UK Office for National Statistics and Government Analysis Function","url":"https://www.gov.uk/government/publications/joined-up-data-in-government-the-future-of-data-linking-methods/longitudinal-linkage-of-administrative-data-design-principles-and-the-total-error-framework","source_class":"GOVERNMENT_OR_REGULATOR","publication_date":"2021-07-16","accessed_at":"2026-08-03","claims_supported":["Longitudinal administrative data are vulnerable to duplication, missing objects, linkage errors, harmonization errors, timing discrepancies, censoring, and retrospective edits.","Compensating aggregate errors can conceal longitudinal defects.","The guidance calls for requirements-driven design, explicit acceptance criteria, uncertainty reporting, source collaboration, and consideration of legal, ethical, technical, and continuity constraints."]},{"source_id":"s4","title":"NIEM 5.0 Justice Domain: j:Arrest","publisher":"National Information Exchange Model","url":"https://niem.github.io/model/5.0/j/Arrest/","source_class":"STANDARD","publication_date":"n.d. (NIEM 5.0 model)","accessed_at":"2026-08-03","claims_supported":["NIEM supplies a machine-readable Justice-domain Arrest element with a defined legal-administrative meaning.","A mature justice information-exchange standard is prior art for common event vocabulary and schema-level representation."]},{"source_id":"s5","title":"OMOP Common Data Model v5.3: Observation Period","publisher":"Observational Health Data Sciences and Informatics","url":"https://ohdsi.github.io/CommonDataModel/cdm53.html","source_class":"STANDARD","publication_date":"n.d. (version 5.3)","accessed_at":"2026-08-03","claims_supported":["OMOP explicitly models periods during which recorded-event absence may be interpreted as event absence.","Outside an observation period, absence of a record cannot be construed as evidence that the event did not occur.","Observation-period derivation is source-dependent and may require explicit assumptions."]},{"source_id":"s6","title":"Temporal tables","publisher":"Microsoft Learn","url":"https://learn.microsoft.com/en-us/sql/relational-databases/tables/temporal-tables?view=sql-server-ver17","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"n.d.; applies to SQL Server 2016 and later","accessed_at":"2026-08-03","claims_supported":["Production database technology already supports retained row histories and point-in-time reconstruction.","System-versioned tables expose AS OF and interval queries, demonstrating technical feasibility for part of the proposed recorded-at history behavior.","Database temporal history alone does not supply justice-event equivalence, observation adequacy, or policy-relative classification semantics."]},{"source_id":"s7","title":"Coded Private Information or Biospecimens Used in Research, Guidance (2018)","publisher":"U.S. Department of Health and Human Services, Office for Human Research Protections","url":"https://www.hhs.gov/ohrp/coded-private-information-or-biospecimens-used-research.html","source_class":"OFFICIAL_GUIDANCE","publication_date":"2018","accessed_at":"2026-08-03","claims_supported":["Coded data can remain identifiable when investigators can readily ascertain identities through a key or other means.","Institutions should designate knowledgeable officials to determine whether secondary research is nonexempt, exempt, or subject to human-subject protections; investigators should not make that determination independently.","Unexpected acquisition of identifiability can change the regulatory status and trigger IRB requirements."]},{"source_id":"s8","title":"Software Developers, Quality Assurance Analysts, and Testers","publisher":"U.S. Bureau of Labor Statistics","url":"https://www.bls.gov/ooh/Computer-and-Information-Technology/Software-developers.htm","source_class":"GOVERNMENT_OR_REGULATOR","publication_date":"2025-08-28","accessed_at":"2026-08-03","claims_supported":["May 2024 median annual wages were $133,080 for software developers and $102,610 for software quality-assurance analysts and testers.","These wage benchmarks support resource-equivalent labor assumptions used in the cost bands, with separate loading assumptions for benefits, overhead, and specialist governance work."]}],"problem_evidence":{"support":"STRONG","rationale":"CJARS directly documents large-scale, multisource justice-event harmonization and substantial jurisdictional differences in coding, processing, and storage. Its unfinished episode-resolution work shows that event identity across justice stages remains difficult. ONS independently documents duplication, timing, linkage, harmonization, censoring, and retrospective-edit errors in longitudinal administrative data and warns that aggregates can conceal them. These sources establish that the general problem exists and can affect longitudinal inference. They do not establish how often a particular study's results change specifically because clients depend on warehouse representation.","source_ids":["s1","s2","s3"]},"stakeholder_evidence":{"support":"MODERATE","rationale":"CJARS is a concrete potential adopter and identifies researchers, data providers, the Census Bureau, the University of Michigan, NSF, and foundations as stakeholders or funders seeking uniform, harmonized longitudinal justice data. Its documentation identifies principal investigators, data-use agreements, IRB oversight, and project review processes capable of authorizing bounded work. No source expresses demand for this exact opaque trajectory contract or commits to a pilot.","source_ids":["s1","s2"]},"prior_art":{"proximity":"ADJACENT_PRIOR_ART","closest_analogues":[{"name":"CJARS integrated and harmonized justice-event infrastructure","similarity":"Very close in domain and objective: it harmonizes dispersed arrest, adjudication, incarceration, probation, and parole records; retains source variables; resolves people and episodes; versions releases; and supports longitudinal research.","remaining_difference":"The opened documentation describes a relational research database and harmonization/linkage pipeline, not a representation-independent behavioral component with a shared black-box oracle, append-only correction law, separate occurrence and recorded-at semantics, and an explicit three-valued window result.","source_ids":["s1","s2"]},{"name":"OMOP Observation Period","similarity":"Close semantic analogue for distinguishing recorded absence within adequate observation from unknown status outside observation.","remaining_difference":"It is a healthcare common-data-model table, not a justice-event state machine; observation-period assumptions remain ETL-defined, and the source does not specify cross-representation conformance, corrections, episode equivalence, or versioned justice classification.","source_ids":["s5"]},{"name":"NIEM Justice-domain information model","similarity":"Established schema-level prior art for representation-independent justice vocabulary and machine-readable arrest semantics.","remaining_difference":"The inspected artifact standardizes message structure and terminology, not longitudinal event identity, as-of knowledge, correction behavior, adequate observation, or substitutability testing.","source_ids":["s4"]},{"name":"SQL Server system-versioned temporal tables","similarity":"Established implementation practice for retaining historical versions and reconstructing state as of a cutoff.","remaining_difference":"System time tracks row versions; it does not define source-assertion truth, occurrence time, event equivalence, observation sufficiency, policy versions, or domain-level abstract equivalence across stores.","source_ids":["s6"]}],"distinctive_claim_remaining":"For one approved longitudinal justice study, a single opaque trajectory contract combining source-attributed assertions, non-erasing corrections, occurrence-versus-recorded-at time, explicit observation intervals, versioned classification rules, a three-valued window answer, and one cross-representation conformance oracle will prevent relational and document implementations from changing predeclared study classifications relative to direct warehouse-query and schema/aggregate-check baselines. This is contrastive and falsifiable but untested; it is not a claim of factual record accuracy, correct linkage, universal ontology, causal validity, or world novelty.","confidence":"MODERATE"},"implementation_evidence":{"support":"MODERATE","rationale":"The component parts are technically credible: CJARS demonstrates large justice-data harmonization, identifiers, episode processing, versioned archives, secure access, and governance; NIEM supplies justice-domain schema terms; OMOP demonstrates explicit observation-period semantics; and temporal databases provide retained history and as-of queries. A synthetic two-store implementation avoids live-data authority barriers. The evidence does not demonstrate that the proposed invariants are jointly coherent, that the oracle is discriminating, or that adapters can preserve study classifications under real source irregularities. For real pseudonymous records, data-use agreements, privacy controls, institutional determinations, and possibly IRB review remain mandatory.","source_ids":["s2","s4","s5","s6","s7"]},"scores":{"meaningful_impact":{"score":4,"rationale":"The documented error modes can alter longitudinal objects, timing, linkage, and classifications, and aggregate checks may miss them. Preventing silent cohort or outcome drift could matter substantially, although realized impact and prevalence in a named study are unmeasured.","source_ids":["s2","s3"]},"stakeholder_pull":{"score":4,"rationale":"CJARS has a large active infrastructure, named governmental and research partners, multiple funders, and an explicit objective of uniform longitudinal justice data. Pull for this exact interface contract has not been expressed.","source_ids":["s1","s2"]},"incremental_advantage":{"score":3,"rationale":"The combined behavioral oracle and three-valued observation result could add protection beyond schemas, harmonized tables, temporal storage, and aggregate migration checks. Most constituent semantics already have strong adjacent precedents, and comparative performance is untested.","source_ids":["s2","s4","s5","s6"]},"distinctiveness_plausibility":{"score":3,"rationale":"No opened source combines all proposed justice-specific behaviors into a representation-independent conformance contract. The distinction is plausible but narrow and based on a bounded search; world novelty, patents, and freedom to operate were not assessed.","source_ids":["s2","s4","s5","s6"]},"technical_implementability":{"score":4,"rationale":"Existing justice harmonization, observation-period modeling, and temporal-history technology make a synthetic prototype feasible. Event-equivalence rules, bitemporal interpretation, generated sequence tests, and leakage auditing still require validation.","source_ids":["s2","s5","s6"]},"adoption_authority_feasibility":{"score":3,"rationale":"A principal investigator, data steward, repository governance process, and IRB or designated institutional official are identifiable authorities. Their willingness to authorize this exact pilot is unknown, and real-data authority varies by agreement and jurisdiction.","source_ids":["s1","s2","s7"]},"evidence_readiness":{"score":3,"rationale":"A synthetic, comparator-based experiment is well bounded and can be executed without identifiable records. There is no inspected client-dependency inventory, mutation-tested oracle, independent implementation result, or study-level outcome comparison.","source_ids":["s2","s3"]},"safety_net_benefit":{"score":3,"rationale":"Returning unknown when observation is inadequate could reduce false absence classifications, a safeguard with a strong analogue in OMOP and support from longitudinal censoring guidance. Its justice-specific calibration and unintended effects are untested.","source_ids":["s3","s5"]},"scalability":{"score":3,"rationale":"A reusable contract and oracle could support multiple storage implementations, but CJARS documents extensive jurisdictional variation and complex manual, rules-based, and machine-learning harmonization. Scaling policy and equivalence rules is likely governance-intensive.","source_ids":["s1","s2"]}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"10K_TO_50K","scope":"Specify 15–25 synthetic trajectories, create a transparent model plus relational and document adapters, seed invalid implementations, and run black-box, generated-sequence, metamorphic, and leakage tests.","confidence":"MODERATE","assumptions":["Approximately 6–12 person-weeks across a data engineer, test engineer, and part-time criminology/data-governance reviewer.","BLS wages are converted to 2026 resource-equivalent cost using roughly 1.5–2.0 times salary for benefits, overhead, contracting, and tooling.","No identifiable records, production security accreditation, or paid data acquisition."],"source_ids":["s2","s8"]},"initial_deployment_startup":{"band_2026_usd":"50K_TO_250K","scope":"Productionize the contract and conformance suite for one bounded study environment, implement one incumbent adapter and one alternative adapter, document governance and versioning, and conduct security and privacy review.","confidence":"LOW","assumptions":["Approximately 0.5–1.5 loaded FTE-years across engineering, analysis, data stewardship, privacy, and project management.","Existing secured research infrastructure and authorized source extracts are reused.","Excludes large-scale remediation of source data and person/episode-linkage model development."],"source_ids":["s2","s7","s8"]},"operational_launch":{"band_2026_usd":"250K_TO_1M","scope":"Integrate the component with one active longitudinal study, independently validate mappings, dual-run against the incumbent pipeline, investigate divergences, train users, approve the contract version, and complete launch assurance.","confidence":"LOW","assumptions":["Approximately 1.5–4 loaded FTE-years plus secure-computing, audit, and validation effort.","Launch includes multiple justice stages and irregular historical extracts but remains one institutional program.","No national standard-setting, new cross-jurisdiction data acquisition, or replacement of upstream linkage systems."],"source_ids":["s1","s2","s7","s8"]},"annual_recurring":{"band_2026_usd":"50K_TO_250K","scope":"Maintain adapters and fixtures, rerun regression and leakage tests, review contract changes, investigate source changes, preserve versioned artifacts, and support privacy and reproducibility audits.","confidence":"LOW","assumptions":["Approximately 0.5–1.5 loaded FTE equivalents annually plus modest secure compute and storage.","Source-policy changes are periodic rather than continuous.","Major new jurisdictions, linkage redesigns, or source remediation are separately funded."],"source_ids":["s2","s8"]}},"verified_pipeline_gates":{"externally_supported_problem":{"status":"YES","reason":"Independent official and first-party sources document jurisdictional representation differences and longitudinal duplication, timing, linkage, censoring, harmonization, and revision errors that can affect research data.","source_ids":["s2","s3"]},"externally_credible_adopter_or_authorizer":{"status":"YES","reason":"CJARS, its University of Michigan principal investigators and repository governance, participating data stewards, the Census Bureau, and institutional review processes are identifiable potential adopters or authorizers with an expressed need for uniform longitudinal justice-data infrastructure. No commitment is established.","source_ids":["s1","s2","s7"]},"distinct_testable_incremental_claim":{"status":"YES","reason":"The incremental claim compares one predeclared behavioral contract and oracle against direct warehouse queries, schema/row-total checks, and existing standardized-data practices, with classification divergence and failure to detect seeded defects as explicit falsifiers.","source_ids":["s2","s4","s5","s6"]},"bounded_next_evidence_step":{"status":"YES","reason":"A 15–25-trajectory synthetic experiment with two deliberately different stores, a transparent reference model, preregistered outputs, baseline comparators, and seeded defects is bounded in scope and cost.","source_ids":["s2","s3","s5","s6","s8"]},"no_unresolved_safety_or_authority_stop":{"status":"YES","reason":"The next step uses only synthetic subjects and excludes operational decisions and identifiable records. Any later real-data work must remain behind data-use, privacy, security, institutional, and IRB determinations; those unresolved requirements do not block the synthetic test.","source_ids":["s2","s7"]},"credible_cost_scope_and_range":{"status":"YES","reason":"All four bands state deliverables, labor assumptions, exclusions, and confidence. BLS wage benchmarks and CJARS-documented engineering, stewardship, security, and review functions make the resource-equivalent ranges credible enough for stage gating, though not procurement estimates.","source_ids":["s2","s8"]}},"next_evidence_step":"With a criminology analyst, data steward, and two independent implementers, preregister 15–25 synthetic trajectories covering duplicated charge rows, consolidated and split episodes, late dispositions, corrections, overlapping and gapped observation, equal occurrence times with different recorded-at times, and policy-version changes. Independently encode them in a relational row store and nested document store. Compare four approaches: direct store-specific queries; schema/row-total/aggregate migration checks; a harmonized-table baseline with OMOP-inspired observation periods; and the proposed opaque contract backed by a transparent reference model. Before execution, fix expected as-of trajectories, three-valued window results, classifications, allowed errors, leakage rules, and at least 12 seeded mutants. Advance only if all three valid implementations have zero material classification or uncertainty-state divergences and the contract suite kills at least 90% of seeded semantic mutants while outperforming baseline checks. Falsify or redesign the intervention if valid implementations passing the suite produce different approved-study classifications, if inadequate observation can become a negative result, if corrections erase earlier knowledge states, or if the proposed suite does not detect more seeded semantic defects than the baselines.","blocking_evidence":["No study-specific inspection establishes the prevalence of undocumented row, null, ordering, or join dependencies.","No identified adopter has requested or committed to the exact contract.","No independent implementation or mutation-tested conformance result shows that the oracle is coherent and discriminating.","No real-data dual run establishes whether the contract preserves cohort membership, timing inputs, or outcomes under actual irregularities.","No institutional or jurisdiction-specific legal determination authorizes use of real pseudonymous justice records under this interface.","Costs are resource-equivalent planning ranges, not vendor quotes or institution-specific budgets."],"research_disposition":"PARTNERED_RESEARCH_PROGRAM","world_novelty_boundary":"This evaluation measured only external support, adjacent prior art, bounded feasibility, and a testable incremental contrast. It did not measure world novelty, patentability, freedom to operate, market size, realized impact, or comprehensive literature and product coverage. No exact match was found in the eight relied-on sources, but the search cannot establish absence elsewhere.","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_version":0,"controller_recommendation":{"action":"STOP_EMPIRICAL_RESEARCH_NEEDED","repairable":false,"material_progress_observed":true,"progress_targets":["Execute the preregistered synthetic three-implementation comparison and publish fixture-level divergence classifications.","Demonstrate zero material divergence among valid implementations and at least 90% detection of preregistered seeded semantic mutants.","Show that the proposed oracle detects more representation-induced semantic defects than direct queries and schema, row-total, and aggregate checks.","Obtain written interest and authority boundaries from a named data steward or research program before any real-data extension.","Before real-data use, obtain the institution's privacy, security, data-use, and human-subject determination and document rollback criteria."],"reason":"The web evidence establishes a real problem, credible stakeholders, substantial adjacent practice, technical plausibility, and a bounded experiment, but it cannot establish the core causal claim that this combined contract detects and prevents representation-induced study-classification drift. That evidence requires implementation and live conformance testing, so further web research cannot close the principal gap."},"proposal_index":4}