{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"invariant_mode_decomposition_design__political_science","arm":"RETRIEVAL_FIRST","candidate_id":"H4-CANDIDATE-0","hypothesis_id":"H4","version":0,"title":"Responsibility-Transfer Mode Audit Prioritization","problem":"Oversight bodies often prioritize agencies with many visible complaints, even when longitudinal referral patterns show that responsibility is repeatedly transferred through mutually reinforcing groups of offices. Coordinate-level rankings can therefore miss offices whose position sustains the dominant responsibility-diffusion pattern.","actors":["Legislative oversight offices","Independent inspectors general and audit offices","Audit-planning analysts","Agency offices that receive, transfer, investigate, or close referrals","Complaint and referral data stewards","Complainants and other people affected by delayed resolution"],"observable_state":"For each agency-office–reporting-period, authorized administrative records show complaint counts, directed transfers between offices, repeated transfers, elapsed status, closure, and existing audit-risk scores. These records define a longitudinal directed weighted network in which an edge from office A to office B represents responsibility for a case being referred from A to B. The observable diagnostic state is a leading network mode concentrated on offices and transfer paths that repeatedly retain or recirculate unresolved cases, together with its spectral separation, temporal stability, and reconstruction residual.","consequence":"Audit capacity can be concentrated on busy offices while structurally important responsibility-transfer paths remain unexamined, allowing cases to stay unresolved or undergo repeated transfer and weakening answerability for case disposition.","affected_objective":"Improve the prospective identification of agency offices where an audit or reporting-rule review could reduce unresolved and repeatedly transferred cases, without treating structural position as evidence of misconduct.","intervention":"Create a decision-support shortlist by estimating the leading mode of the authorized referral operator, then simulate bounded, policy-plausible perturbations at each node—such as improved closure, deadline enforcement, or reduced onward-transfer rates—and rank offices by the resulting marginal reduction in that mode. Compare the ranking prospectively with complaint volume, degree, generic centrality, and the oversight office's existing risk score. The result may inform human audit planning but cannot automatically initiate an audit or reporting-rule change.","structural_mapping":[{"archetype_element":"Transformation Scope","domain_realization":"A reporting-period-specific directed weighted operator maps responsibility transfers among agency offices; its boundary, edge definition, weighting, and treatment of closures are fixed before estimation."},{"archetype_element":"State-Vector Definition","domain_realization":"Each coordinate represents an agency office in a reporting period, with authorized case-level records aggregated into transfer, repeat-transfer, unresolved-status, and closure measures."},{"archetype_element":"Invariant Mode Basis","domain_realization":"The referral operator's leading approximately invariant direction represents the combination of offices along which responsibility transfers are most structurally reinforced within the declared observation window."},{"archetype_element":"Modal Gain Spectrum","domain_realization":"Estimated eigenvalues summarize the relative amplification or persistence of referral-network modes; they are descriptive properties of the specified operator, not proof of causal responsibility."},{"archetype_element":"Dominant Mode Selection Rule","domain_realization":"Use the leading mode only when convergence, spectral separation, temporal stability, and held-out reconstruction meet preregistered thresholds; otherwise retain multiple modes or reject modal prioritization."},{"archetype_element":"Modal Intervention Map","domain_realization":"Each feasible audit or reporting-rule candidate is translated into a bounded change to the relevant node or incident transfer weights, and offices are ranked by the simulated reduction in the selected responsibility-transfer mode."},{"archetype_element":"Reconstruction Residual Check","domain_realization":"Reconstruct held-out transfer patterns from retained modes and inspect both residual magnitude and structured residuals so low-amplitude but consequential pathways are not silently discarded."},{"archetype_element":"Mode Drift and Interpretation Scope","domain_realization":"Track rotation, reordering, and spectral-gap changes across reporting periods; suspend use after material drift, regime change, altered recording rules, or repeated residual failure."}],"mechanism_mapping":[{"mechanism_slug":"network_spectral_centrality_analysis","role":"Treat the interoffice referral network as the transformation and measure each office's participation in its leading responsibility-transfer mode.","counterfactual_removal":"Without this mechanism, the proposal loses its coupled network representation and reduces to office-level complaint, risk, or local connectivity measures."},{"mechanism_slug":"power_iteration_probe","role":"Estimate the leading mode and its gain from a large or sparse referral operator while using convergence behavior and the spectral gap as trust checks.","counterfactual_removal":"Without this mechanism, there is no scalable, explicit estimate of the proposed dominant mode or warning that a small gap makes its identity unstable."},{"mechanism_slug":"modal_sensitivity_sweep","role":"Apply equal, bounded, policy-plausible perturbations to candidate nodes and rank them by change in the leading-mode gain and in predicted unresolved or repeated transfers, while recording cross-mode effects.","counterfactual_removal":"Without this mechanism, modal participation remains descriptive centrality and cannot support the hypothesis's outcome-specific intervention ranking."}],"causal_chain":["Repeated interoffice transfers create a directed responsibility-transfer operator rather than independent office-level workloads.","Mutually reinforcing transfer paths produce a leading mode whose office loadings need not follow complaint volume or degree.","A bounded perturbation to an office's closure or onward-transfer behavior changes both its incident edges and the operator's leading response.","The modal sensitivity sweep identifies offices where the same feasible perturbation most reduces the leading responsibility-transfer mode.","If that mode captures a consequential pathway, higher marginal-reduction ranks predict more future unresolved or repeatedly transferred cases than the named baselines.","Human audit planners can use the validated incremental signal as one input when selecting audits or reporting-rule reviews."],"baseline":"Freeze all feature definitions and model choices before a temporal holdout. Compare the modal-perturbation ranking against complaint volume, directed and weighted degree, generic eigenvector centrality without outcome-specific perturbation, and the oversight office's existing conventional risk score. Evaluate all rankings on the same future reporting periods and the same outcomes: unresolved status and repeated transfer.","nearest_rivals":["Complaint-volume ranking of agency offices","Directed or weighted degree centrality","Generic eigenvector-centrality ranking without an outcome-specific sensitivity sweep","Existing risk-based audit-planning scores","The Key Player Problem's outcome-specific node-removal ranking","Public-sector network-intervention frameworks applied to governance networks"],"remaining_contrastive_claim":"In longitudinal agency referral networks, ranking audit targets by each node's marginal reduction of the leading responsibility-transfer mode predicts future unresolved or repeatedly transferred cases better than complaint volume, degree centrality, and existing risk scores.","authority_safety":{"decision_authority":"The legally authorized legislative or independent oversight office retains sole authority over audit selection, investigative action, and reporting-rule changes; the model and its analysts have advisory authority only.","authorized_first_step":"Authorized analysts may conduct a non-deployment feasibility and temporal-holdout study on access-controlled referral records, producing aggregate performance, stability, residual, and sensitivity results for internal review.","excluded_actions":["Automatically opening, prioritizing, or closing an audit or investigation","Treating a high modal score as evidence of wrongdoing, intent, or legal responsibility","Publishing identifiable office or complainant rankings before authorized review","Changing live referral routing, deadlines, staffing, or reporting rules during the first evidence step","Using complaint narratives, protected attributes, or records beyond the approved data scope","Simulating elimination of an office as though it were a feasible policy intervention"],"halt_rollback":"Do not issue or withdraw any operational shortlist if the spectral gap is below its preregistered threshold, mode identity drifts materially, residuals remain structured, referral coding changes, access controls fail, or performance does not exceed the named baselines. Roll back by discarding the shortlist, retaining existing audit-planning procedures, and documenting the failed validity condition."},"negative_tests":{"strongest_counterevidence":"The proposal substantially collides with established complaint-network analysis, responsibility-shifting networks, public-sector network interventions, key-player perturbation methods, coordinated risk-based audit planning, and operational referral tracking. Its defensible distinction is only the narrow integration of a longitudinal interoffice referral operator, leading-mode perturbation ranking, and prospective validation against unresolved or repeatedly transferred cases.","problem_falsifier":"The nominated problem is falsified if authorized longitudinal records show that repeated responsibility transfers are rare or nonpersistent, that no reproducible coupled transfer structure exists across reporting periods, or that unresolved cases are fully accounted for by complaint volume and existing risk factors without network dependence.","intervention_falsifier":"The intervention claim is falsified by a preregistered temporal holdout across multiple oversight systems showing no reliable incremental predictive value over complaint volume, degree centrality, and existing risk scores; it also fails operationally if the leading mode or perturbation ranking is too unstable to reproduce under reasonable edge definitions and reporting periods.","risks":["Incomplete or inconsistent referral records can manufacture apparent bottlenecks.","A small spectral gap can make the leading mode an unstable mixture of several modes.","High participation may reflect lawful specialization or mandatory routing rather than accountability diffusion.","Outcome-specific perturbations may encode unrealistic changes or confuse prediction with causal effect.","Rankings may redirect scrutiny toward already burdened offices or produce strategic changes in recordkeeping.","Aggregating case records can obscure consequential low-volume pathways or expose sensitive information.","A strong result in one oversight system may not transport to another institutional or legal setting."]},"next_evidence_step":"In one consenting oversight system, conduct a bounded feasibility study using only approved structured fields from a fixed sequence of reporting periods. Preregister the node and edge definitions, feasible perturbation size, gap and drift thresholds, residual tolerance, baselines, outcomes, and temporal split; estimate rankings on earlier periods and score unresolved and repeatedly transferred cases in the next untouched period. Report incremental performance, ranking stability under reasonable network specifications, residual structure, and data-quality failures. Do not use the result for live audit selection; progression to a multi-system prospective test requires both baseline improvement and passage of all stability and safety checks.","prior_art_status":"SEARCHED_BOUNDED","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}