{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"invariant_mode_decomposition_design__library_information_science","arm":"RETRIEVAL_FIRST","round_index":0,"hypotheses":[{"hypothesis_id":"H1","title":"Residual-Gated Metadata Crosswalks","problem":"Field-by-field crosswalks can preserve common records while silently corrupting rare combinations of metadata.","affected_stakeholder":"Metadata librarians migrating heterogeneous collections.","workflow_boundary":"From source-record export through target-schema acceptance.","failure_mode":"Aggregate validation hides structured loss concentrated in low-volume record types.","unit_of_analysis":"A record-by-target-field reconstruction error.","causal_lever":"Retain crosswalk modes until every governed record class meets an out-of-sample residual tolerance.","archetype_mapping":"Treat the source-to-target crosswalk as a rectangular transformation, decompose its paired input-output directions, and inspect residual structure before discarding modes.","expected_value":"Fewer semantically damaged records without requiring exhaustive rule review.","falsifiable_claim":"Compared with field-level validation, residual-gated mode selection will reduce post-migration semantic-defect rates in held-out rare record classes without increasing manual review per record by more than 20%.","diversity_rationale":"Targets one-time schema transformation, record-level fidelity, and a stopping-rule intervention rather than dynamic stability or network ranking.","mechanism_slugs":["singular_value_decomposition","residual_reconstruction_test"],"search_questions":["Have metadata crosswalks been selected or validated using SVD and class-stratified reconstruction residuals?","Which record classes generate the most structured migration residuals?","What defect and review-cost benchmarks exist for metadata migrations?"]},{"hypothesis_id":"H2","title":"Unstable-Mode Checks for Discovery Feedback","problem":"Clicks, ranking, and future exposure can reinforce coupled popularity patterns that ordinary relevance metrics miss.","affected_stakeholder":"Patrons using library discovery systems, especially for less-exposed holdings.","workflow_boundary":"From result ranking through click-derived ranking updates.","failure_mode":"A weakly visible exposure mode grows across update cycles and suppresses otherwise relevant materials.","unit_of_analysis":"A query-cluster by item-group exposure state per ranking update.","causal_lever":"Dampen update components classified as unstable while preserving stable relevance components.","archetype_mapping":"Linearize the ranking-feedback update, decompose its modes, classify their repeated-update behavior, and test outcome sensitivity to mode-specific damping.","expected_value":"More stable access to relevant holdings with less self-reinforcing exposure concentration.","falsifiable_claim":"Mode-specific damping will reduce exposure-concentration growth over repeated offline updates while holding judged relevance within a prespecified noninferiority margin.","diversity_rationale":"Uses a repeated temporal transformation, group-level exposure state, and stability control rather than reconstruction fidelity.","mechanism_slugs":["eigendecomposition_workflow","modal_stability_analysis","modal_sensitivity_sweep"],"search_questions":["Do library discovery studies model click-ranking feedback as a dynamical operator?","Which exposure modes predict concentration beyond item-level popularity?","What relevance and exposure-concentration measures permit an offline controlled test?"]},{"hypothesis_id":"H3","title":"Modal Preservation Risk Sampling","problem":"Digital-preservation audits often sample by format or collection even when failures arise from coupled storage, software, provenance, and handling conditions.","affected_stakeholder":"Digital preservation managers allocating limited audit capacity.","workflow_boundary":"From preservation-risk feature collection through selection of objects for integrity review.","failure_mode":"Independent risk flags miss low-variance combinations associated with preservation failure.","unit_of_analysis":"A preserved object represented by its longitudinal risk-feature vector.","causal_lever":"Allocate audit samples to poorly reconstructed or outcome-sensitive modal regions rather than to isolated high-risk fields.","archetype_mapping":"Extract correlated risk modes, test retained-mode residuals, and perturb modal coordinates against observed preservation outcomes to rank audit leverage.","expected_value":"Higher failure yield per audit and earlier detection of cross-factor preservation risks.","falsifiable_claim":"At an equal audit budget, modal-region sampling will identify more previously unknown integrity failures than format-stratified random sampling in retrospective evaluation.","diversity_rationale":"Concerns scarce inspection allocation, object-level longitudinal risk, and detection yield rather than migration acceptance or feedback control.","mechanism_slugs":["principal_component_analysis","residual_reconstruction_test","modal_sensitivity_sweep"],"search_questions":["Has digital-preservation audit sampling used latent modes rather than format strata?","Are integrity failures associated with low-variance or residual risk patterns?","Which repositories retain sufficient longitudinal features and verified failures for retrospective testing?"]},{"hypothesis_id":"H4","title":"Spectral Routing for Interlibrary Loan","problem":"Interlibrary-loan policies based on local volume can overload structurally pivotal lenders and make fulfillment fragile.","affected_stakeholder":"Resource-sharing consortia, lending staff, and requesting patrons.","workflow_boundary":"From lender-candidate generation through routing of an eligible request.","failure_mode":"Degree or volume metrics miss libraries whose network position concentrates system-wide dependency.","unit_of_analysis":"A consortium library node within a rolling request-routing network.","causal_lever":"Diversify routing away from high-dependency dominant-mode nodes when simulated removal sensitivity exceeds a threshold.","archetype_mapping":"Treat request flows as a connectivity operator, rank nodes by the dominant network mode, and test fulfillment sensitivity to rerouting or node removal.","expected_value":"Lower bottleneck load and more resilient fulfillment without adding holdings.","falsifiable_claim":"Spectrally diversified routing will reduce the maximum lender-load share and simulated node-loss disruption without materially increasing median fulfillment time.","diversity_rationale":"Uses a network node as the unit, routing policy as the lever, and resilience as the outcome—distinct from record, object, and query dynamics.","mechanism_slugs":["network_spectral_centrality_analysis","power_iteration_probe","modal_sensitivity_sweep"],"search_questions":["Which centrality measures have been tested in interlibrary-loan routing?","Does eigenvector centrality predict disruption better than request volume or degree?","What routing logs permit counterfactual replay of lender diversification?"]},{"hypothesis_id":"H5","title":"Mode-Drift Alarms for Authority Maintenance","problem":"Authority files can change through many small edits while aggregate quality metrics remain stable, leaving an obsolete semantic structure in use.","affected_stakeholder":"Catalogers and communities represented by controlled vocabularies.","workflow_boundary":"From approved authority-record edits through scheduled vocabulary review.","failure_mode":"Coupled shifts among terms, references, and co-assignment contexts rotate the operative semantic basis before surface counts trigger review.","unit_of_analysis":"An authority term within successive term-reference and co-assignment graphs.","causal_lever":"Trigger scoped human review when dominant modes rotate or their spectral separation falls below a governed threshold.","archetype_mapping":"Re-estimate dominant vocabulary modes over time, monitor spectral gaps and direction drift, and issue an interpretation-bounded review report.","expected_value":"Earlier, more targeted review of structural vocabulary change with fewer blanket audits.","falsifiable_claim":"Mode-drift alerts will predict subsequent clusters of authority corrections better than edit volume and term-frequency baselines at the same review capacity.","diversity_rationale":"Centers longitudinal governance of a semantic graph, term-level prediction, and alarm timing rather than direct transformation, sampling, control, or routing.","mechanism_slugs":["spectral_gap_monitor","power_iteration_probe","spectral_decomposition_report"],"search_questions":["Have authority-control systems monitored eigenvector rotation or spectral-gap change over time?","Do later correction clusters align with prior modal drift after controlling for edit volume?","How should near-degenerate modes and community interpretation limits be represented in review alerts?"]}]}