{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","cell_id":"predictive_residual_processing__criminology_forensic","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"prp_crime_program_efference_residual_monitoring_v0","proposal_index":4,"version":0,"title":"Efference-Residual Monitoring for Crime-Prevention Programs","problem":"When an agency changes patrol, outreach, reporting access, or another place-based crime-prevention program, the intervention can change both underlying events and how events become recorded. Increased officer presence may predictably change officer-initiated contacts, detected offenses, calls, complaints, or data completeness even if external conditions are otherwise unchanged. Routine monitoring that presents every aggregate in full can consume evaluation attention while unexpected spatial displacement, temporal displacement, reporting divergence, service withdrawal, or rights-related harm remains difficult to distinguish from the program's own measurement footprint.","actors":["Independent program evaluator","Agency data steward","Crime analyst assigned to evaluation rather than deployment","Program manager","Community oversight body","Service or outreach partner","Law-enforcement leadership operating under existing authority","Residents and people exposed to the program","Civil-rights, defense, victim-service, and community representatives reviewing aggregate effects"],"observable_state":"Within predeclared aggregate place-time cells, reviewers can observe the frozen intervention schedule, predicted and recorded officer-initiated activity, calls for service, reported incidents, service contacts, complaints, uses of force, injuries, data missingness, source reliability, population exposure, model version, structured residuals, reconstruction error, alternative-model disagreement, protected-signal bypasses, audit findings, fallback frequency, and the rationale for any subsequent human inquiry. Individual-level scores are neither produced nor routed.","consequence":"Program leaders may attribute predictable intervention-generated changes in recorded activity to external crime change, overlook unexpected displacement or harm, or continue a program on the basis of a self-confirming measurement stream. Conversely, an unexplained aggregate residual may be overinterpreted as causal proof of program success, failure, misconduct, or community behavior.","affected_objective":"Support timely, contestable monitoring of whether a place-based crime-prevention program and its measurement system are behaving outside their predeclared expectations, without producing individual risk scores or allowing residuals to determine enforcement, causal attribution, or program continuation automatically.","intervention":"Before each aggregate monitoring window, copy the authorized program schedule and operational intensity into a frozen, versioned forward model. This efference copy predicts the program's direct measurement footprint under its declared scope: expected officer-initiated contacts, administrative records, detection opportunities, service contacts, reporting-channel use, and associated uncertainty. It does not predict individual offending or define the true crime level. The system compares observed aggregate streams with the predicted footprint and emits structured residuals such as excess or missing activity, spatial or temporal displacement, cross-source disagreement, unexpected complaint or injury change, and unexplained missingness. A precision-weighted gate prioritizes residuals by source reliability, uncertainty, exposure, rights consequence, persistence, and evaluator capacity. Each routed residual includes the expected baseline, observed aggregate, model version, uncertainty, intervention history, and access to reconstruct the complete monitoring window. Complaints, force, injury, deaths, reporting outages, protected-class disparity checks, data-integrity failures, whistleblower submissions, and oversight requests bypass suppression and appear in full. Random place-time windows and risk-stratified windows receive independent full-data review using at least one information source not generated by the intervention operator where available. Residuals open an evaluation inquiry rather than an operational response. Human reviewers classify each as possible program effect, measurement effect, external shock, model failure, source failure, or unresolved discrepancy. Model revisions occur only offline after review; shocks, drift, version mismatch, audit disagreement, inadequate cell size, or reconstruction failure return the affected scope to complete multi-source reporting.","structural_mapping":[{"archetype_element":"Prediction target and boundary","domain_realization":"The aggregate measurement footprint expected from one declared place-based program within specified geographic cells, time windows, data sources, intervention intensities, and minimum privacy-preserving cell sizes."},{"archetype_element":"Predictive feedforward model","domain_realization":"A frozen model receives a copy of the authorized intervention schedule before outcomes arrive and predicts the direct records, contacts, detection opportunities, and reporting changes the program itself may generate."},{"archetype_element":"Expected and actual behavior","domain_realization":"The expected aggregate vector is versioned before the observation window; actual aggregates are captured separately with source, timestamp, completeness, transformation, and exposure metadata."},{"archetype_element":"Prediction comparator","domain_realization":"The comparator preserves signed and categorical discrepancies across amount, location, time, source agreement, reporting completeness, complaints, force, injury, and service access."},{"archetype_element":"Precision weighting","domain_realization":"Residual priority combines uncertainty, source independence and reliability, affected population exposure, persistence, rights consequence, rarity, and finite evaluator attention."},{"archetype_element":"Residual propagation and reconstruction","domain_realization":"Selected discrepancies reach evaluators with the intervention-generated expectation and sufficient aggregate context to reconstruct the full monitoring window rather than appearing as context-free anomaly alerts."},{"archetype_element":"Confidence and uncertainty state","domain_realization":"The system separately represents uncertainty in the intervention schedule, observation sources, forward model, residual, and any reconstruction, and widens or abandons predictive mode when uncertainty is excessive."},{"archetype_element":"Bounded update rule","domain_realization":"Residuals may revise a temporary evaluation hypothesis immediately, but structural model changes require offline review that separates intervention actions, observations, external shocks, and delayed outcomes."},{"archetype_element":"Synchronization and validity","domain_realization":"The evaluator and reporting system verify compatible program definitions, geographic boundaries, source transformations, model checksums, and validity windows before interpreting residuals."},{"archetype_element":"Independent raw-state audit","domain_realization":"Random and risk-stratified place-time windows receive complete multi-source review, including protected signals and an independent or differently generated source where available."},{"archetype_element":"Safety and rights bypass","domain_realization":"Force, injury, death, complaints, disparity checks, whistleblower information, missing reporting channels, integrity failures, and oversight requests always travel in full and cannot be suppressed as expected consequences."},{"archetype_element":"Drift-triggered fallback","domain_realization":"External shocks, intervention changes, reporting-system changes, model disagreement, small cells, cumulative reconstruction error, or audit misses suspend residual presentation and restore complete aggregate reporting."}],"mechanism_mapping":[{"mechanism_slug":"efference_copy_cancellation","role":"Uses a copy of the agency's own program schedule to predict the records and measurement opportunities likely to be generated by that action, allowing evaluators to foreground the unexplained remainder.","counterfactual_removal":"Without the action copy, the model cannot distinguish a predictable change in recorded activity caused by altered observation effort from an unexpected change requiring evaluation."},{"mechanism_slug":"forecast_backtesting","role":"Defines the geographic, temporal, intervention-intensity, and reporting regimes in which the footprint model may be used, with walk-forward evaluation and untouched windows.","counterfactual_removal":"Without scoped backtesting, residual suppression could be applied during program intensities, shocks, neighborhoods, or reporting regimes the predictor cannot reconstruct defensibly."},{"mechanism_slug":"event_triggered_residual_reporting","role":"Routes persistent or consequential aggregate discrepancies while heartbeats and completeness checks make silence distinguishable from failed reporting.","counterfactual_removal":"Without event-triggered reporting, the evaluation channel either repeats all routine aggregates or interprets absent reports as confirmation."},{"mechanism_slug":"precision_weighted_error_gate","role":"Allocates evaluator attention using uncertainty, source reliability, exposure, persistence, rights consequence, and capacity while retaining suppressed residuals for audit.","counterfactual_removal":"Without consequence-aware weighting, large noisy administrative changes could crowd out smaller, reliable complaint, injury, access, or cross-source discrepancies."},{"mechanism_slug":"model_version_checksum_handshake","role":"Verifies that program definitions, geographic boundaries, source transformations, and predictor parameters match between residual generation and review.","counterfactual_removal":"Without compatibility checks, an aggregate residual could be reconstructed against different intervention boundaries or data definitions and produce a plausible but invalid comparison."},{"mechanism_slug":"residual_comparison_test","role":"Tests whether residuals retain spatial, temporal, source-specific, or exposure-linked structure and compares them with a rival model and independent full-window samples.","counterfactual_removal":"Without residual comparison, structured displacement or measurement failure could be dismissed as random noise, while an overfit model graded itself."},{"mechanism_slug":"shadow_raw_channel_sampling","role":"Routes random and risk-stratified complete monitoring windows through an independent review path to reveal information suppressed or never represented by the primary model.","counterfactual_removal":"Without full-window sampling, the system could appear calibrated because it observes only discrepancies defined by its own intervention-footprint model."},{"mechanism_slug":"model_drift_monitoring","role":"Tracks source completeness, residual distributions, program implementation, reporting practices, external conditions, realized review findings, and model staleness.","counterfactual_removal":"Without drift monitoring, an unofficial program change or altered reporting relationship could be absorbed into the expected footprint and become self-confirming."},{"mechanism_slug":"raw_signal_fallback_switch","role":"Restores complete multi-source reporting when protected signals, shocks, incompatibility, small cells, source failure, reconstruction error, or audit disagreement invalidate residual processing.","counterfactual_removal":"Without a tested fallback, the predictor could continue filtering aggregate evidence during precisely the regime changes it cannot represent."},{"mechanism_slug":"prediction_error_review","role":"Requires an independent, provenance-tagged review to distinguish possible program effects, measurement effects, external shocks, source failures, model failures, and unresolved discrepancies before action or model revision.","counterfactual_removal":"Without structured review, residuals could be treated as causal conclusions or used to rationalize the program that generated the observations."},{"mechanism_slug":"surprise_to_action_bridge","role":"Routes a validated residual to a named evaluation owner and a bounded next step such as source verification, full-window review, community consultation, or a separately authorized program assessment.","counterfactual_removal":"Without a defined handoff, residuals would decorate a dashboard without ensuring investigation, acknowledgement, or accountable closure."}],"causal_chain":["The authorized program schedule is copied into a frozen forward model before the monitoring window begins.","The model predicts the aggregate records and observation opportunities the program itself is expected to generate, with explicit uncertainty and scope.","Independent data pipelines retain complete aggregate observations while a comparator computes signed, multi-source residuals against the expected footprint.","Reliability, exposure, persistence, and rights consequence determine which residuals reach the bounded evaluation channel, while protected signals bypass suppression.","Evaluators receive each discrepancy with its program history, expected baseline, uncertainty, provenance, and reconstructible full-window context.","Independent full-window audits test whether the predictor suppressed displacement, service withdrawal, harm, reporting divergence, or source failure.","Human review determines whether the discrepancy is consistent with a possible program effect, measurement effect, external shock, model defect, data failure, or unresolved condition; the residual alone establishes none of these.","Validated surprises can authorize only a bounded evaluation step under existing governance, not individual enforcement or automatic program change.","Residual and audit findings may support a separately approved offline model revision that keeps intervention actions distinct from observations and delayed outcomes.","Drift, shocks, source failure, incompatible versions, inadequate privacy-preserving aggregation, or excess reconstruction error return the affected scope to complete multi-source reporting."],"baseline":"Program monitoring uses full periodic dashboards, pre-program and post-program comparisons, maps, fixed performance indicators, narrative reports, and occasional formal evaluations. These approaches retain broad context but may conflate intervention-generated measurement changes with external change and do not necessarily make a frozen action-conditioned expectation, structured residual channel, raw-window audit, synchronized reconstruction, bounded learning loop, and decompression trigger part of continuing governance.","nearest_rivals":["Full multi-indicator dashboards: preserve context and may remain preferable when prediction is weak, but allocate attention across all measures and do not subtract the program's expected measurement footprint.","Pre-program versus post-program trend comparison: simple and interpretable, but vulnerable to concurrent changes and does not separate self-generated observation effects through a synchronized residual architecture.","Difference-based or comparison-area evaluation: can estimate contrasts under stated assumptions and is a strong rival for causal assessment, but it is an episodic estimator rather than a continuing residual communication, audit, and model-correction system.","Interrupted time-series or other formal program evaluation: can test program-associated changes more rigorously than a residual alert and should govern causal claims, but does not inherently route only action-conditioned innovations with full-state fallback.","Hot-spot or displacement mapping: visualizes spatial change but does not distinguish predictable observation changes, verify silence, synchronize model versions, or update a scoped generative footprint model.","Generic anomaly detection on aggregate crime data: ranks unusual cells but does not use an efference copy of the intervention, reconstruct expected context, preserve protected full-signal bypasses, or prevent anomalies from being treated as causal evidence."],"remaining_contrastive_claim":"The proposal's distinguishing testable feature is not crime forecasting or causal impact estimation. It copies the program's own authorized actions into a synchronized model of their expected measurement footprint, routes only precision- and consequence-weighted aggregate residuals for evaluation, and preserves reconstructible full context, independent multi-source audits, rights-critical bypasses, bounded offline updates, and automatic decompression when the action-conditioned representation is invalid.","authority_safety":{"decision_authority":"An independent evaluation lead or oversight body may authorize and terminate the retrospective test and approve model versions. Existing public authorities retain any lawful power to modify or pause a program, subject to their governing procedures. The residual system cannot allocate patrols, identify people, authorize enforcement, determine causation, or continue or terminate a program.","authorized_first_step":"Conduct a retrospective, read-only shadow evaluation of one completed place-based program using privacy-preserving aggregate data and frozen program records. Predeclare spatial and temporal units, sources, protected bypasses, predictor version, thresholds, rival analyses, audit windows, and stopping rules before scoring. Keep results separate from individual records and live operational systems.","excluded_actions":["Do not generate individual offending, victimization, dangerousness, or enforcement-risk scores.","Do not allocate patrols, initiate stops, searches, arrests, surveillance, investigations, sanctions, or service restrictions from a residual.","Do not treat a residual as causal proof of program success, failure, displacement, misconduct, or community behavior.","Do not suppress force, injury, death, complaint, disparity, whistleblower, missingness, or data-integrity information.","Do not use cells below the predeclared privacy and reliability threshold or attempt to reidentify people from residuals.","Do not change the active model online or optimize thresholds to produce a preferred program narrative or alert volume.","Do not replace a formal causal evaluation, legal review, public reporting duty, or community oversight process.","Do not use the bounded evidence step for live program or enforcement decisions."],"halt_rollback":"Halt residual processing for the affected scope after a protected-bypass failure, reporting outage, source-definition or model mismatch, inadequate cell size, audit-discovered material omission, unexplained reconstruction error, residual-channel saturation, intervention change, or external shock outside the declared model boundary. Restore complete multi-source reporting, preserve all versions and suppression logs, notify the independent evaluation owner, and require a new predeclared retrospective protocol before reuse."},"negative_tests":{"strongest_counterevidence":"A blinded retrospective comparison shows that full dashboards or a simple rival identify material monitoring changes with equal or lower effort; the action-conditioned predictor cannot separate its own measurement footprint from external change; residuals track omitted contextual variables; independent audits reveal suppressed rights or displacement signals; reviewers make stronger unsupported causal claims when shown residuals; or synchronization, audit, and maintenance costs consume the attention released.","problem_falsifier":"The inferred problem is absent if the program does not materially change how monitored events are observed or recorded, evaluation attention is not constrained by repetitive expected measures, or existing complete reporting already exposes unexpected cross-source, displacement, and rights-related changes within the required time.","intervention_falsifier":"The intervention is unsupported if a frozen, predeclared test cannot reconstruct full aggregate monitoring windows within tolerance, preserve complete presentation of protected signals, distinguish reporting failure from predicted silence, remain calibrated across its declared program intensities, or improve the joint criteria of evaluation effort, material-change coverage, detection timing, false escalation, and interpretive restraint against the strongest rival.","risks":["The model could normalize repeated over-enforcement, under-service, or rights harm as an expected program consequence.","Police-generated records could dominate independent sources and produce a self-confirming baseline.","Residuals could be misread as causal effects despite confounding, spillovers, anticipation, or concurrent events.","Aggregate analysis could invite ecological inferences about individuals or communities.","Intervention managers could alter schedules, reporting, boundaries, or thresholds to reduce visible residuals.","Underreporting or service-access changes could appear as reduced harm unless protected independent signals remain available.","Small reliable harm signals could be suppressed by large noisy administrative changes.","Risk-stratified audits could miss unanticipated communities, times, or forms of displacement.","Random audits may still be too sparse to detect rare consequential events.","External shocks or changing reporting practices could make the predictor confidently wrong.","Residual displays could expose sensitive geographic or community information even without individual records.","Saved evaluation attention could be redirected into broader surveillance or more intensive enforcement.","Online learning could absorb the program's own consequences and erase evidence that the model should be challenged.","An all-green residual dashboard could be interpreted as program success when it means only that no modeled discrepancy was observed.","Independent-source availability and quality may differ across areas, producing uneven uncertainty or attention." ]},"next_evidence_step":"Pre-register one offline, blinded comparison using privacy-preserving aggregate records from a completed program or a synthetic program replay. Freeze the intervention schedule, geographic boundaries, model, thresholds, minimum cell sizes, source transformations, protected bypasses, and holdout windows. Include known or simulated reporting outages, deployment-intensity changes, temporal and adjacent-area displacement, changes in officer-initiated activity, complaint and injury shifts, service withdrawal, cross-source disagreement, benign external shocks, and program-definition mismatch. Have evaluators review randomized monitoring periods through the existing full dashboard and the residual-first interface, while a separate team performs complete multi-source audits. Record reconstruction error, protected-signal presentation, time to acknowledge seeded or independently documented changes, false escalations, missed displacement or harm patterns, causal overinterpretations, version and heartbeat failures, fallback behavior, evaluator time, and maintenance cost. Stop on any protected-signal omission, privacy-threshold breach, unrecoverable context, or residual-driven individual or enforcement recommendation. The result may support only a further retrospective evaluation, not operational adoption.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Earlier proposal 1 predicts operating-system and application background events within seized-device timelines to allocate a forensic examiner's attention to content-level discrepancies. This proposal neither examines device content nor prioritizes evidentiary artifacts; it models an intervention's aggregate measurement footprint and routes program-monitoring residuals to independent evaluation. Earlier proposal 2 predicts authorized custodianship, location, seal, and workflow transitions for physical evidence. This proposal does not monitor evidence handling or reconcile custody state; its action input is a crime-prevention program schedule, its observations are aggregate community and agency measures, and its downstream response is program evaluation. Earlier proposal 3 predicts analytical responses within forensic toxicology instrument batches to assess measurement validity. This proposal does not assess specimens, instruments, controls, or batch release; it uses an efference copy of institutional action to distinguish expected self-generated observations from aggregate social and reporting deviations. It is independently adoptable as an evaluation-governance layer without digital-timeline triage, custody assurance, or toxicology residual assurance.","revision_record":{"parent_version":null,"progress_targets_addressed":["Created a fourth independently adoptable candidate for the same archetype-domain cell.","Selected a materially different crime-program evaluation problem and action-conditioned causal path.","Used the efference-copy mechanism as the core predictor rather than extending the prior content, custody, or instrument proposals.","Specified complete actors, observables, intervention, mappings, rivals, authority, safeguards, falsifiers, risks, and bounded evidence.","Explicitly differentiated the candidate from sealed proposals 1, 2, and 3."],"conceptual_changes":["Initial version; no parent proposal.","Applied predictive residual processing to an institutional intervention's aggregate measurement footprint rather than to forensic evidence examination, custody state, or laboratory measurement.","Separated residual monitoring from individual prediction, causal attribution, and enforcement authority."],"operational_changes":["Specified privacy-preserving aggregate cells, a frozen intervention efference copy, multi-source structured residuals, protected rights bypasses, independent audits, offline updates, and complete-report fallback.","Bound the first evidence step to retrospective shadow evaluation with no live program or enforcement decisions."],"evidence_changes":["Prior art remains unsearched.","Defined a pre-registered comparison against full dashboards and a simple rival analysis without asserting an effect size."],"claim_changes":["No novelty, prevalence, demand, causal effect, or operational effectiveness claim is made.","The remaining claim is limited to a falsifiable action-conditioned residual architecture and conditional evaluation value under predeclared reconstruction, privacy, workload, and rights safeguards."]}}