{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","cell_id":"representation_independent_interface_contract__criminology_forensic","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"observation_bounded_justice_trajectory_v0","proposal_index":4,"version":0,"title":"Observation-Bounded Justice Event Trajectory Contract","problem":"Longitudinal criminology analyses can become coupled to how an administrative dataset represents arrests, bookings, charges, cases, dispositions, supervision episodes, corrections, and observation gaps. Analysts may treat rows as events, null fields as absence, current values as historically known values, or warehouse-specific joins as the definition of a trajectory. When the warehouse, linkage output, or study data mart is reimplemented, nominally equivalent queries can change cohort membership or recorded outcomes because no representation-independent component defines what an observed justice event, correction, or period of observable absence means.","actors":["Criminology researchers","Study statisticians and analysts","Administrative-data stewards","Research data engineers","Principal investigators","Institutional privacy and ethics reviewers","Authorized reproducibility auditors"],"observable_state":"Given the same synthetic source assertions, alternative data representations can return different numbers or kinds of events, different as-of histories, or different answers to whether a qualifying event was recorded during a study window. Divergence can arise from duplicated charge rows, case aggregation, late-entered dispositions, corrections, overlapping source coverage, null handling, default ordering, or the loss of distinctions between no recorded event and no adequate observation.","consequence":"A storage migration or rewritten data-access layer can silently alter eligibility classifications, event counts, time-to-recorded-event inputs, or reproducibility outputs even though the study definition and source snapshot are intended to remain fixed. Analysts may then attribute a representation-induced change to the studied trajectory or fail to recognize that the available records do not support a definite absence claim.","affected_objective":"Preserve predeclared longitudinal event, observation-window, correction, and as-of query semantics across changes in the representation of a bounded criminology research dataset.","intervention":"Define an Observation-Bounded Justice Event Trajectory as an opaque abstract state machine for a pseudonymous study subject. Its abstract state contains source-attributed event assertions, event-equivalence links, occurrence intervals, recorded-at times, supersession or correction links, explicit source-observation intervals, and a versioned study-classification policy. Public operations append an authorized source assertion, declare two assertions equivalent under a stated rule, append a correction without erasing the earlier assertion, open or close a source-observation interval, retrieve the trajectory as known at a cutoff, classify qualifying recorded events under a named policy version, and ask whether a window contains a qualifying event. The last operation returns a three-way result: recorded event, no recorded event within contractually adequate observation, or unknown because observation is insufficient. Invariants require source attribution for every assertion; preservation of superseded assertions; acyclic correction chains; symmetric event equivalence; separation of occurrence time from recorded-at time; explicit policy versions; and unchanged state after rejected operations. Row layout, table normalization, document nesting, warehouse keys, indexes, caches, and join strategy remain hidden. An implementation is substitutable only if the same black-box and generated-sequence oracle produces equivalent abstract trajectories, classifications, uncertainty states, and allowed errors from equivalent synthetic inputs.","structural_mapping":[{"archetype_element":"Abstract component","domain_realization":"One pseudonymous, observation-bounded Justice Event Trajectory used within an approved longitudinal study."},{"archetype_element":"Observable operation surface","domain_realization":"Append source assertion, relate equivalent assertions, append correction, register observation interval, query an as-of trajectory, classify events under a named policy, and evaluate a study window with a three-way observation result."},{"archetype_element":"Abstract state model","domain_realization":"Source-attributed event assertions, occurrence and recorded-at times, equivalence classes, correction chains, observation intervals, policy versions, and pseudonymous subject scope."},{"archetype_element":"Behavioral laws and invariants","domain_realization":"Corrections do not erase prior assertions; as-of queries exclude information unavailable at the cutoff; equivalent source assertions do not become multiple abstract events under the selected rule; inadequate observation yields unknown rather than absence; query operations do not mutate state."},{"archetype_element":"Hidden representation","domain_realization":"Charge rows, booking tables, case documents, warehouse keys, nested records, indexes, physical ordering, caches, and concrete join plans."},{"archetype_element":"Abstraction function","domain_realization":"Each valid concrete dataset maps to the contract-visible trajectory of event assertions, equivalence relations, corrections, observation intervals, and policy-relative classifications it denotes."},{"archetype_element":"Error and side-effect semantics","domain_realization":"Unauthorized source additions, invalid intervals, cyclic corrections, unknown policy versions, and out-of-scope subject access return specified error categories and leave abstract state unchanged; reads emit no unapproved disclosure."},{"archetype_element":"Conformance oracle and substitutability rule","domain_realization":"One parameterized suite compares abstract results across relational, document, or graph implementations; matching schemas or row counts alone do not establish acceptance."}],"mechanism_mapping":[{"mechanism_slug":"abstract_data_type_specification","role":"Defines trajectories independently of rows and supplies representation invariants and an abstraction mapping from each concrete research store to event, correction, observation, and classification meanings.","counterfactual_removal":"Without it, the intervention would be a data dictionary and could not decide whether differently normalized datasets denote the same longitudinal trajectory."},{"mechanism_slug":"design_by_contract_clause","role":"Specifies preconditions, postconditions, error categories, and unchanged-state guarantees for assertions, equivalence declarations, corrections, observation intervals, and queries.","counterfactual_removal":"Without it, invalid intervals, conflicting corrections, missing policy versions, and insufficient observation could receive implementation-specific treatment."},{"mechanism_slug":"opaque_type_or_module_boundary","role":"Exposes pseudonymous trajectory handles and sanctioned research operations while withholding warehouse keys, source tables, and concrete join paths.","counterfactual_removal":"Without it, analysis code could continue depending directly on row structures and warehouse identifiers, defeating representation independence."},{"mechanism_slug":"black_box_contract_test_suite","role":"Runs one contract-derived fixture suite against every implementation and compares as-of histories, event equivalence, observation results, classifications, errors, and authorized outputs.","counterfactual_removal":"Without it, migration acceptance would depend on schema comparison, aggregate spot checks, or code review rather than a shared behavioral rule."},{"mechanism_slug":"property_based_conformance_test","role":"Generates event, correction, coverage, and query sequences to test correction acyclicity, equivalence laws, temporal cutoffs, policy-version stability, and the distinction between absent and unknown.","counterfactual_removal":"Without it, unusual combinations of late entry, overlapping observation, duplication, and correction could escape hand-authored examples."},{"mechanism_slug":"metamorphic_behavior_test","role":"Checks relations such as extending an adequately observed event-free window preserving an earlier event-free subwindow, adding a duplicate equivalent assertion not increasing abstract event multiplicity, and moving the knowledge cutoff earlier never revealing later-recorded assertions.","counterfactual_removal":"Without it, the oracle would lose representation-independent checks for related histories where enumerating every correct trajectory output is impractical."},{"mechanism_slug":"representation_leakage_probe","role":"Audits outputs for warehouse row IDs, source-specific null conventions, implicit ordering, table names, join multiplicity, and timing behavior that clients could mistake for study semantics.","counterfactual_removal":"Without it, an implementation could pass promised queries while continuing to expose accidental warehouse features that analysis code would make contractual."},{"mechanism_slug":"semantic_versioning_and_deprecation_gate","role":"Governs changes to event-equivalence rules, adequate-observation criteria, classification policies, error categories, and authorized projections.","counterfactual_removal":"Without it, a data-layer update could change study classifications while presenting itself as a compatible implementation revision."}],"causal_chain":["Administrative research data encode justice events and observation coverage through implementation-specific rows, nesting, nulls, and joins.","Analysis clients consequently turn representation details into implicit rules for event identity, temporal knowledge, correction, and absence.","Reimplementing the store or query layer can then change study inputs even when the source snapshot and declared research question remain fixed.","The intervention defines a trajectory through abstract event assertions, observation intervals, corrections, equivalence laws, versioned classification policies, and three-way window results.","An opaque boundary prevents analysis code from reaching warehouse internals, while the abstraction mapping pins each valid concrete state to the same contract-visible trajectory meaning.","Shared black-box, property-based, and metamorphic checks reject implementations that change as-of histories, multiplicity, classifications, uncertainty states, or errors.","A conforming implementation can replace the bounded trajectory component without making its row organization an unofficial part of the study definition."],"baseline":"The comparison baseline is a fixed study analysis executed directly against a particular warehouse or data-mart representation, with migration checked through schema mapping, row totals, selected records, and aggregate output comparison. The bounded evaluation would first inventory which representational assumptions the analysis actually makes rather than presume their prevalence.","nearest_rivals":["A common administrative-data schema standardizes tables and field names but does not by itself define event equivalence, as-of knowledge, correction behavior, or adequate observation.","A statistical analysis plan defines models and variables but may leave the data component’s state transitions, missingness semantics, and representation invariants implicit.","A study-specific ETL pipeline materializes one analysis table but can freeze source-layout assumptions and offers no general substitutability test for an alternative implementation.","A record-linkage system estimates which records concern the same person or event, whereas this contract begins with authorized pseudonymous scope and governs how supplied linkage assertions behave downstream.","An immutable dataset snapshot preserves bytes but does not ensure that two query engines abstract duplicated rows, corrections, or observation gaps identically.","Manual record validation can investigate selected discrepancies but does not create a reusable black-box behavioral surface for every research client."],"remaining_contrastive_claim":"The proposal’s limited claim is architectural: an observation-bounded trajectory component can make event identity, temporal knowledge, correction, and insufficient-observation semantics explicit while hiding warehouse representation. It does not determine whether an administrative assertion is factually true, solve person linkage, define a universal criminal-justice ontology, estimate causal effects, or authorize operational decisions about individuals.","authority_safety":{"decision_authority":"The approved study’s principal investigator and data steward, subject to applicable privacy and ethics oversight, may authorize a synthetic conformance exercise and approve contract versions for the research environment. Interpretive conclusions, publication claims, data-release decisions, and any change in permitted use remain with the authorized human governance process.","authorized_first_step":"Represent one synthetic trajectory fixture set in two deliberately different stores—such as charge-row relational tables and nested episode documents—and run a predeclared contract suite through opaque adapters without accessing identifiable or live administrative records.","excluded_actions":["No ingestion or linkage of identifiable person records","No use of the component for policing, sentencing, supervision, release, eligibility, or individual risk decisions","No inference that an administrative assertion proves conduct or guilt","No conversion of unknown observation into a negative event finding","No alteration of an approved study outcome definition after observing implementation results","No release of source identifiers, warehouse keys, or pseudonym crosswalks","No production migration based only on synthetic conformance"],"halt_rollback":"Stop if the contract permits reidentification, exposes source keys, collapses unknown observation into absence, erases corrected assertions, changes an approved policy without versioning, or cannot represent a source-specific distinction required by the study. Preserve the existing research environment, discard only synthetic sandbox state, and refer the disputed abstraction to the data steward and study governance process."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence would show that the study-relevant meaning of events and observation gaps is inseparable from source-specific legal definitions, collection practices, or native record context that the abstract trajectory omits. It would also count against the proposal if independently valid representations cannot share event-equivalence or adequate-observation rules without concealing material uncertainty.","problem_falsifier":"The problem is falsified for the bounded study if dependency inspection shows that all clients already use an explicit behavioral trajectory contract, no analysis reads undocumented representation details, and independently implemented adapters return identical abstract histories, classifications, uncertainty states, and errors across the fixture set.","intervention_falsifier":"The intervention is falsified for the tested scope if two implementations pass every predeclared conformance and leakage check yet produce materially different approved-study eligibility, event, timing, or observation classifications from information the contract was supposed to preserve.","risks":["The abstract event categories may erase legal or administrative distinctions relevant to interpretation.","An equivalence rule may merge separate events or count one event multiple times.","Adequate-observation criteria may be framed too strongly and turn incomplete coverage into apparent absence.","Synthetic fixtures may not exercise irregular corrections, delayed entry, or overlapping source coverage that matter to the study.","A passing suite may be mistaken for evidence that source assertions are accurate.","Opaque access may impede legitimate source-context review unless sanctioned provenance operations are sufficient.","Versioned policy changes can still undermine comparability if analysts combine outputs from different versions.","Pseudonymous trajectories may remain sensitive and vulnerable to reidentification through detailed event combinations."]},"next_evidence_step":"Construct 15–25 synthetic subject trajectories covering duplicate charge rows, consolidated cases, split episodes, late-recorded dispositions, corrected dates, overlapping and gapped observation intervals, identical occurrence times with different recorded-at times, and policy-version changes. Encode each trajectory independently in a relational row model and a nested document model. Before running adapters, specify expected abstract events, as-of views, three-way window results, error categories, and metamorphic relations. Run the identical black-box suite against a simple transparent model and both adapters, then classify each divergence as an implementation defect, contract omission, source-context distinction, or over-specified test. This step evaluates only whether the bounded abstraction is coherent and discriminating; it does not use real records or estimate research effects.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 governed physical-evidence continuity across custody, seals, sampling, consumption, and corrections. Proposal 2 governed artifact retrieval, provenance, qualifications, and queries over a fixed digital-forensic source image. Proposal 3 governed the hidden classification, role visibility, and lifecycle of a blind proficiency exercise. Proposal 4 instead addresses longitudinal criminology research, where warehouse rows, joins, corrections, and observation gaps become implicit definitions of a pseudonymous justice-event trajectory. Its causal path is analytic classification drift caused by representational assumptions, and its intervention is an observation-bounded trajectory type with as-of semantics and a mandatory distinction between recorded absence and unknown observation. It can be adopted as a research data-access component without changing evidence custody, digital-artifact extraction, or proficiency administration, and each earlier proposal remains independently adoptable without it.","revision_record":{"parent_version":null,"progress_targets_addressed":["Fourth independently adoptable proposal","Explicit differentiation from proposals 1, 2, and 3","Representation-independent criminology research component","Authority, privacy, and non-operational-use safeguards","Bounded falsifiable evidence plan"],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}