{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","source_assessment_id":"predictive_residual_processing__criminology_forensic:P4:v0","cell_id":"predictive_residual_processing__criminology_forensic","proposal_index":4,"qualification":"EMPIRICAL_PARTNER_CANDIDATE","criteria":{"specific_differentiated_claim":{"status":"YES","reason":"The remaining claim is specific and falsifiable: an action-conditioned residual-first interface can reduce total evaluation effort by at least 20% while keeping material-change recall within five percentage points of the best comparator, preserving protected signals and reconstruction, and not increasing unsupported causal interpretations."},"credible_problem_signal":{"status":"YES","reason":"Government guidance and empirical research establish divergent crime measures, spatially variable reporting error, displacement concerns, and the need to distinguish program implementation and measurement effects from impact."},"identifiable_partner_or_adopter":{"status":"YES","reason":"A law-enforcement agency running place-based programs, paired with an independent evaluator and community or civil-rights oversight body, is a concrete adopter-authorizer class; BJA-supported agency-research partnerships provide a credible institutional pathway."},"partner_access_is_necessary":{"status":"YES","reason":"The decisive uncertainties require nonpublic frozen program schedules, linked historical administrative and independent-source aggregates, audit labels, local governance approval, and blinded practitioner reviewers; public research or a synthetic replay cannot establish real reconstruction, workflow fit, miss patterns, or total maintenance burden."},"safe_authorized_first_step":{"status":"YES","reason":"The proposed first step is a retrospective, read-only, privacy-preserving aggregate shadow study isolated from live enforcement and individual records, with independent authorization, protected-signal bypasses, stopping rules, and fallback to complete reporting."},"bounded_decisive_empirical_design":{"status":"YES","reason":"The 12- to 16-week, one-program preregistered study fixes 150–250 windows, comparator interfaces, blinded crossover review, audit samples, thresholds, noninferiority and time-saving targets, and explicit safety and performance falsifiers."},"no_material_negative_gate":{"status":"YES","reason":"All substantive verified gates are YES; cost scope is uncertain but not a safety, authority, problem, adopter, distinctiveness, or test-design stop, and agency-specific costing can be collected within the partnered study."},"not_merely_more_research":{"status":"YES","reason":"The next action is operationally specified: secure one agency-evaluator-oversight partnership and execute a fixed comparator study with named access requirements, metrics, thresholds, audits, and termination conditions."}},"uncertainty_types":["PROBLEM_PREVALENCE","ADOPTER_PULL","INCREMENTAL_EFFECT","WORKFLOW_FIT","DATA_ACCESS","COST_SCOPE"],"partner_profile":"One law-enforcement agency with a completed place-based prevention program, an independent program-evaluation team, an authorized data steward, and a community, civil-rights, or public oversight body able to approve aggregate retrospective access and independent auditing.","required_access":"The completed program's frozen schedule and geographic definitions; privacy-reviewed aggregate officer activity, incidents, calls, complaints, force, injury, service, missingness, and independent-source data; existing comparator dashboards; provenance and version records; documented change or audit labels; eight or more blinded evaluators; and authority for independent full-window audits and reviewer-behavior measurement.","bounded_empirical_test":"Run a preregistered 12- to 16-week offline comparison over 150–250 historical place-time windows. Randomize at least eight blinded evaluators, using crossover and balanced ordering, among the existing full dashboard, a conventional process-plus-impact dashboard, and the residual-first interface. Audit every protected-signal window and a random sample of other windows; test seeded or documented outages, displacement, activity changes, complaint or injury shifts, service withdrawal, shocks, source disagreement, and version mismatch.","success_condition":"Zero protected-signal omissions and privacy breaches; material-change recall no more than five percentage points below the best comparator; at least 20% lower total reviewer-plus-maintenance effort; reconstruction failures below 1% of windows; reliable heartbeat, mismatch, and fallback behavior; no increase in unsupported causal interpretations; and no systematic miss pattern across geography, exposure, source availability, or disparity checks.","falsification_condition":"Any protected-signal omission or privacy breach; recall outside the five-point noninferiority margin; failure to reduce total effort by 20%; systematic independent-audit misses; reconstruction failure above 1%; failed heartbeat, version, or fallback handling; increased unsupported causal attribution; or subgroup/place miss patterns falsifies the incremental claim.","rationale":"This is a narrow partnered-research candidate because the problem and institutional pathway are externally credible and the integrated residual architecture retains a testable contrast from established evaluation practice. Its claimed advantage and safety cannot be established through further web research: they depend on lawful agency data access, actual evaluator behavior, independent audits, and agency-specific integration and governance work. The retrospective comparator study is bounded enough to reject the claim without exposing individuals or influencing live enforcement."}