{"schema_version":1,"research_id":"eoa_inverse_innovation_exp04_external_evaluation_20260802","source_assessment_id":"invariant_mode_decomposition_design__library_information_science:SENTINEL:v0","cell_id":"invariant_mode_decomposition_design__library_information_science","search_queries":["interlibrary loan lender load concentration turnaround time queue depth routing OCLC","interlibrary loan network social network analysis centrality lender concentration study","interlibrary loan routing load balancing ReShare supplier rota ratio","interlibrary loan consortium data privacy transaction records ALA","site:projectreshare.org consortium resource sharing governance steering committee load balancing libraries","DOCLINE routing random lender selection NLM routing cells official","interlibrary loan network queueing model delay load library JASIS 1976","spectral critical nodes network robustness node removal degree arxiv 1101.5019","site:bls.gov/ooh data scientists median pay 2025","site:bls.gov/ooh software developers median pay 2025","site:bls.gov/ooh librarians library media specialists median pay 2025","library privacy interlibrary loan records official ALA code 2023","\"Application of social network analysis in interlibrary loan services\" filetype:pdf","\"Limits of Concentration and Dispersion\" interlibrary loan pdf","interlibrary loan network analysis transaction logs centrality primary research pdf","interlibrary loan lending concentration libraries study resource sharing network","\"interlibrary loan\" eigenvector routing lender","\"interlibrary loan\" spectral centrality routing","patent interlibrary loan spectral routing lender centrality","library resource sharing lender node removal robustness routing"],"sources":[{"source_id":"S1","title":"Limits of Concentration and Dispersion of Lending Service Based on a Model of Interlibrary Loan Network","publisher":"Japan Society of Library and Information Science / J-STAGE","url":"https://www.jstage.jst.go.jp/article/ajsls/41/3-4/41_130/_pdf/-char/ja","source_class":"PRIMARY_RESEARCH","publication_date":"1995-12","accessed_at":"2026-08-02","claims_supported":["ILL demand assignment can be modeled subject to which lenders hold requested resources.","Achievable lending-service concentration and dispersion are constrained by lender capacity, capacity concentration, and network size.","Collection and capacity heterogeneity can impose a floor below which lender load cannot be evenly distributed."]},{"source_id":"S2","title":"Smart lender strings through the Automated Request Manager","publisher":"OCLC Support","url":"https://help.oclc.org/Resource_Sharing/WorldShare_Interlibrary_Loan/Smart_fulfillment/040Smart_lender_strings_through_the_Automated_Request_Manager","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"n.d.","accessed_at":"2026-08-02","claims_supported":["WorldShare automatically constructs and orders eligible lender strings.","Ordering uses historical turnaround time and hourly queue depth specifically to avoid overwhelming fast lenders.","Holdings, licensing, deflection, cost, and group preferences constrain lender selection before ordering."]},{"source_id":"S3","title":"Load balancing algorithm","publisher":"Open Library Foundation ReShare","url":"https://openlibraryfoundation.atlassian.net/wiki/spaces/PR/pages/2551775240/Load%2Bbalancing%2Balgorithm","source_class":"OFFICIAL_PRODUCT_DOCUMENTATION","publication_date":"2021-12-20","accessed_at":"2026-08-02","claims_supported":["ReShare dynamically orders suppliers using a ratio-based load-balancing score.","Each library may define a target lending-to-borrowing ratio.","Scores change as active requests open and close, demonstrating operational demand for load-aware lender ordering."]},{"source_id":"S4","title":"How does DOCLINE routing work?","publisher":"U.S. National Library of Medicine","url":"https://www.nlm.nih.gov/docline/context_help/context_help_routing_choices.html","source_class":"OFFICIAL_GUIDANCE","publication_date":"2018-05-08","accessed_at":"2026-08-02","claims_supported":["DOCLINE filters lenders using holdings, participation, delivery, service, payment, embargo, and exclusion rules.","Among eligible libraries in routing cells and broader groups, lender selection is randomized.","Requests are not routed repeatedly to an institution already tried."]},{"source_id":"S5","title":"Distributed Algorithm to Locate Critical Nodes to Network Robustness based on Spectral Analysis","publisher":"arXiv","url":"https://arxiv.org/abs/1101.5019","source_class":"PRIMARY_RESEARCH","publication_date":"2011-01-26","accessed_at":"2026-08-02","claims_supported":["Spectral analysis can rank nodes associated with network robustness.","Node-removal experiments can validate whether spectrally identified nodes are consequential.","The most damaging node is not always the highest-degree node, establishing the possible incremental value of topology-aware measures over local degree."]},{"source_id":"S6","title":"Community Charter (Legacy)","publisher":"Project ReShare","url":"https://projectreshare.org/about/community-charter/","source_class":"OFFICIAL_ORGANIZATION_DATA","publication_date":"n.d.","accessed_at":"2026-08-02","claims_supported":["Libraries and consortia describe resource-sharing networks as critical services and seek more responsive, interoperable technology.","Project ReShare expressly supports experimentation and innovation in resource sharing.","Its Steering Committee oversees strategy, resource allocation, risk, and funding, identifying a credible authorizer for a platform-level study."]},{"source_id":"S7","title":"State Privacy Laws Regarding Library Records","publisher":"American Library Association","url":"https://www.ala.org/advocacy/privacy/statelaws","source_class":"OFFICIAL_GUIDANCE","publication_date":"n.d.","accessed_at":"2026-08-02","claims_supported":["Forty-eight states and the District of Columbia protect library-record confidentiality, with protections also recognized in the remaining states through attorney-general opinions.","Interlibrary-loan records may contain protected personally identifiable use data.","Any study requires jurisdiction-specific policy review, data minimization, and appropriate authorization rather than assuming transaction logs are unrestricted operational data."]},{"source_id":"S8","title":"WorldShare Interlibrary Loan: Resource sharing that delivers resources fast","publisher":"OCLC","url":"https://www.oclc.org/en/worldshare-ill.html","source_class":"COMMERCIAL_FIRST_PARTY","publication_date":"n.d.; page contained 2026 content when accessed","accessed_at":"2026-08-02","claims_supported":["WorldShare ILL automates resource sharing across more than 10,000 libraries and reports more than three million annual e-resource deliveries.","Existing reports include borrowing and lending counts, fill rates, and turnaround times.","The platform exposes resource-sharing APIs and uses standards including ISO 18626, making integration or offline extraction technically plausible, although the page does not establish availability of complete historical candidate sets."]}],"problem_evidence":{"support":"MODERATE","rationale":"The general problem—lender-load concentration and the need to balance eligible assignments—is visible in an ILL concentration model and in two deployed systems that explicitly prevent overload or balance lending ratios. ILL also operates at consequential scale. However, no opened source measures the candidate's narrower condition: a current consortium whose fulfillment depends on a spectrally pivotal lender not already identified by volume, degree, queue depth, or turnaround. Prevalence and realized disruption therefore remain unverified.","source_ids":["S1","S2","S3","S8"]},"stakeholder_evidence":{"support":"MODERATE","rationale":"Project ReShare identifies libraries, consortia, and its Steering Committee as stakeholders with roadmap, funding, risk, and resource-allocation authority, while ReShare and OCLC already invest in load-aware routing. This establishes a credible adopter/authorizer class and expressed need for better resource-sharing technology, but no named consortium has requested spectral routing or committed data, staff, or funding to this evaluation.","source_ids":["S2","S3","S6","S8"]},"prior_art":{"proximity":"SUBSTANTIAL_COLLISION","closest_analogues":[{"name":"OCLC WorldShare smart lender strings","similarity":"A deployed ILL system filters eligible lenders and dynamically orders them using historical turnaround time and current queue depth to improve fulfillment without overwhelming lenders.","remaining_difference":"The opened documentation does not disclose a rolling request-flow spectral score, lender-loss outcome replay, a sensitivity threshold, or a predeclared fulfillment-time noninferiority rule.","source_ids":["S2"]},{"name":"ReShare ratio-based load balancing","similarity":"A deployed resource-sharing platform dynamically reorders eligible suppliers to distribute lending according to consortium-defined target ratios.","remaining_difference":"The documented score is a local active-request ratio rather than a topology-aware dependency estimate, and it does not validate priority changes through lender-removal fulfillment replay.","source_ids":["S3"]},{"name":"DOCLINE randomized eligible-lender routing","similarity":"An official operational router already diversifies assignments within policy-defined eligibility and preference cells.","remaining_difference":"Its documented randomization does not use spectral dependency, critical-node analysis, or outcome-sensitive node removal.","source_ids":["S4"]},{"name":"Spectral critical-node localization with removal validation","similarity":"Research already combines spectral node ranking with removal experiments and shows that degree can miss damaging nodes.","remaining_difference":"It concerns generic network connectivity, not directed weighted ILL flows, eligible-lender ordering, fulfillment outcomes, lender loads, or latency constraints.","source_ids":["S5"]},{"name":"ILL capacity-constrained concentration model","similarity":"Prior ILL research models how requests are assigned to holding libraries and analyzes bounds on lending concentration and dispersion.","remaining_difference":"It does not estimate rolling spectral dependency or implement an outcome-constrained routing policy.","source_ids":["S1"]}],"distinctive_claim_remaining":"For a consortium with reconstructable eligibility and outcome histories, a routing-risk policy that uses a stable rolling request-flow spectral score plus fulfillment-specific lender-removal replay to trigger diversification among already eligible lenders will reduce both maximum lender-load share and held-out lender-loss disruption beyond the existing router, volume, degree, random, queue-aware, and ratio-based comparators, without worsening median fulfillment time beyond a predeclared noninferiority margin.","confidence":"HIGH"},"implementation_evidence":{"support":"MODERATE","rationale":"The calculation and offline replay are technically ordinary at consortium scale, and existing platforms already compute dynamic ordering, retain outcome metrics, expose APIs, and apply eligibility constraints. The proposed first step can be non-operational and reversible. Feasibility is nevertheless contingent on proprietary data not established by public sources: complete historical eligible-but-unselected sets, reroutes, declines, expirations, and timestamps. Privacy and authority require a consortium data agreement, jurisdiction-specific review, minimization or deletion of patron and request-content fields, pseudonymous institution identifiers, and an explicit prohibition on institutional performance labeling. Spectral-gap failure, mode instability, scarce-holdings constraints, and cross-load transfer are analytical stop conditions, not merely tuning issues.","source_ids":["S2","S3","S4","S6","S7","S8"]},"scores":{"meaningful_impact":{"score":3,"rationale":"On a 0–4 scale, the affected service is large and operationally important, and avoiding concentrated lender overload could protect fulfillment and staff workload. The exact incremental harm attributable to hidden spectral dependency is not yet measured.","source_ids":["S1","S2","S8"]},"stakeholder_pull":{"score":3,"rationale":"Operational platforms already implement load balancing, and ReShare's community governance explicitly seeks innovation and can allocate resources. Pull for this specific spectral method is not documented.","source_ids":["S2","S3","S6"]},"incremental_advantage":{"score":2,"rationale":"Topology-aware scoring may identify consequential nodes missed by degree, but the candidate must beat strong deployed queue, turnaround, random, and ratio baselines on fulfillment outcomes rather than merely produce a different ranking.","source_ids":["S2","S3","S4","S5"]},"distinctiveness_plausibility":{"score":2,"rationale":"The complete ILL-specific combination was not found in the bounded search, but nearly every component exists separately and terminology-mismatched publications, patents, proprietary routing logic, or local implementations remain unchecked.","source_ids":["S1","S2","S3","S4","S5"]},"technical_implementability":{"score":3,"rationale":"Network construction, eigensolvers, held-out replay, and constrained reordering are feasible with standard software, existing platform data, and APIs. Faithful counterfactual reconstruction and modal stability are the main unresolved technical risks.","source_ids":["S2","S3","S8"]},"adoption_authority_feasibility":{"score":3,"rationale":"A resource-sharing consortium or the ReShare Steering Committee is identifiable and has relevant governance authority, but platform vendors, participating libraries, and local data controllers may each need separate approval.","source_ids":["S6","S7"]},"evidence_readiness":{"score":2,"rationale":"Relevant metrics and transaction infrastructure exist, but public evidence does not show that a consenting consortium retains every candidate, eligibility decision, decline, reroute, and outcome needed for unbiased replay.","source_ids":["S2","S4","S7","S8"]},"safety_net_benefit":{"score":4,"rationale":"The proposed first test is offline, pseudonymizable, threshold-gated, and makes no live routing change; failure can end in a no-go report. Privacy review remains mandatory before data access.","source_ids":["S6","S7"]},"scalability":{"score":2,"rationale":"Computation should scale readily, but operator definitions, eligibility rules, privacy constraints, network topology, and integration differ across consortia, limiting transfer without repeated local validation.","source_ids":["S2","S3","S4","S7","S8"]}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"10K_TO_50K","scope":"An 8–12 week offline study for one consortium: data audit, privacy review, one historical extract, preregistration, baseline and spectral replays, stability checks, and a go/no-go report.","confidence":"MODERATE","assumptions":["Existing compute and analysis software are used.","The consortium can produce one bounded extract without vendor engineering.","Approximately 0.15–0.35 resource-equivalent FTE-years are divided among an analyst, ILL subject-matter expert, data engineer, and privacy reviewer.","No live integration, patron contact, or routing change occurs."],"source_ids":["S6","S7","S8"]},"initial_deployment_startup":{"band_2026_usd":"50K_TO_250K","scope":"Production-quality data pipeline, repeatable spectral and replay service, access controls, audit logs, monitoring dashboard, governance documentation, and shadow-mode platform integration for one consortium.","confidence":"LOW","assumptions":["Existing platform APIs or database exports are sufficient.","Vendor licensing and major data-retention redesign are excluded.","Roughly 0.5–1.5 resource-equivalent FTE-years plus security and integration review are required.","This phase remains shadow-mode and does not alter live lender order."],"source_ids":["S3","S6","S7","S8"]},"operational_launch":{"band_2026_usd":"50K_TO_250K","scope":"Separately approved limited live pilot, platform configuration, staff training, monitoring, incident response, independent analysis, and rollback readiness across a bounded request subset.","confidence":"LOW","assumptions":["The offline study passes all preregistered falsifiers.","The platform permits reversible priority changes without replacing eligibility logic.","One consortium and a limited traffic share are included.","No material vendor procurement or multi-year contract is required."],"source_ids":["S2","S3","S6","S8"]},"annual_recurring":{"band_2026_usd":"10K_TO_50K","scope":"One-consortium recurring data refresh, mode and gap monitoring, outcome audits, governance review, incident handling, and periodic threshold recalibration.","confidence":"LOW","assumptions":["Existing infrastructure is reused after launch.","Approximately 0.1–0.35 resource-equivalent FTE-years cover operations and governance.","Major platform upgrades, consortium expansion, and external audits are excluded."],"source_ids":["S3","S6","S8"]}},"verified_pipeline_gates":{"externally_supported_problem":{"status":"YES","reason":"ILL concentration is modeled in primary research, and deployed OCLC and ReShare routing explicitly address overload or lending imbalance. The narrower hidden-spectral-dependency prevalence remains an empirical gap, but the general operational problem is externally supported.","source_ids":["S1","S2","S3"]},"externally_credible_adopter_or_authorizer":{"status":"YES","reason":"Project ReShare's Steering Committee has documented authority over strategy, resources, risk, and funding for a platform already performing lender load balancing; a participating consortium remains necessary for data authorization.","source_ids":["S3","S6"]},"distinct_testable_incremental_claim":{"status":"YES","reason":"The candidate states outcome comparators, held-out metrics, a latency noninferiority constraint, modal validity checks, and conditions under which the spectral policy loses to existing methods.","source_ids":["S2","S3","S4","S5"]},"bounded_next_evidence_step":{"status":"YES","reason":"A single-consortium, at-most-12-month, offline replay can be time-boxed, preregistered, compared with six baselines, and terminated without operational change.","source_ids":["S2","S3","S4","S8"]},"no_unresolved_safety_or_authority_stop":{"status":"UNCERTAIN","reason":"An offline design avoids live-routing harm, but no named consortium authorization, data-processing agreement, retention audit, or jurisdiction-specific determination has established that the necessary ILL logs may be retained and analyzed.","source_ids":["S6","S7"]},"credible_cost_scope_and_range":{"status":"YES","reason":"All four estimates are broad resource-equivalent bands with bounded phase scopes and explicit staffing, integration, and procurement assumptions. Confidence is limited because no vendor quote or consortium-specific labor plan was available.","source_ids":["S3","S6","S8"]}},"next_evidence_step":"Secure written participation from one named consortium and complete a data-retention and privacy audit before extraction. For one fixed window of no more than 12 months, verify that logs reconstruct the complete eligible-lender set, ordering, declines, reroutes, expirations, fulfillment outcome, and timestamps without patron identity or request-content fields. Preregister the directed weighted operator, fit/hold-out split, gap and mode-alignment thresholds, sensitivity trigger, median-time noninferiority margin, and missing-data rules. On the held-out segment, replay the recorded router, volume, degree, random diversification, deployed queue/turnaround logic where reconstructable, ratio balancing, and the spectral-plus-removal policy using identical eligibility sets. Report maximum lender-load share, per-lender removal disruption, filled and unfilled share, median and tail fulfillment time, rerouting cross-loads, uncertainty, spectral gap, mode stability, and subgroup checks. Falsify the intervention if it does not reduce both load concentration and removal disruption, fails to outperform the strongest non-spectral comparator reliably, breaches the time margin, transfers excessive load to smaller lenders, or lacks a stable interpretable mode. End after 8–12 weeks with an offline go/no-go report; authorize no live routing change.","blocking_evidence":["No current consortium-level estimate shows how often structurally pivotal lenders exist after controlling for volume, degree, queue depth, turnaround, holdings scarcity, and existing routing policy.","Public sources do not establish availability or quality of complete eligible-but-unselected, decline, reroute, expiration, and outcome histories needed for faithful counterfactual replay.","No consortium-specific test demonstrates a stable spectral gap, mode alignment across windows, or robustness to alternative edge orientation, weighting, normalization, and window length.","No named consortium has supplied written authorization, a data-processing agreement, or jurisdiction-specific privacy approval for the required ILL transaction analysis.","No held-out comparison establishes improvement over the strongest deployed queue-aware or ratio-based router while satisfying fulfillment-time and cross-load constraints.","Operational feedback effects—changed lender acceptance, staff behavior, and formation of a new pivotal lender—cannot be resolved by ordinary web research."] ,"research_disposition":"PARTNERED_RESEARCH_PROGRAM","world_novelty_boundary":"This assessment is a bounded public-web screen conducted with the recorded queries and eight directly opened sources. It found substantial collision with deployed ILL load balancing, randomized diversification, capacity-concentration modeling, and generic spectral critical-node removal testing, but did not find the complete ILL-specific combination of rolling spectral dependency, fulfillment-specific lender-loss replay, threshold-triggered diversification, and a fulfillment-time constraint. This is not a measurement of world novelty, patentability, freedom to operate, market size, or realized impact. Proprietary OCLC and consortium logic, local implementations, non-indexed literature, source code, procurement documents, and a comprehensive patent search remain outside the evidence boundary.","arm":"SENTINEL","candidate_version":0,"controller_recommendation":{"action":"STOP_EMPIRICAL_RESEARCH_NEEDED","repairable":true,"material_progress_observed":true,"progress_targets":["Obtain a named consortium's written study authorization, data-processing terms, and privacy approval.","Demonstrate that a bounded historical extract reconstructs complete eligibility, ordering, decline, reroute, expiration, and fulfillment histories with patron-identifying and request-content fields excluded.","Preregister the operator, modal-validity thresholds, outcome sensitivity trigger, missing-data rules, strongest comparators, fulfillment-time margin, cross-load guardrail, and all falsifiers.","Complete the held-out offline replay and show reproducible improvement in both maximum lender-load share and lender-loss disruption over the strongest non-spectral comparator without breaching latency, tail-delay, unfilled-share, or cross-load limits.","Show that results persist across reasonable time windows and operator specifications and that a stable, interpretable mode exists; otherwise retire the spectral policy.","Only after those targets pass, seek separately authorized shadow-mode or limited live testing with immediate rollback."],"reason":"Bounded web research has established the general problem, credible authorizer class, substantial prior-art collision, and a falsifiable remaining claim. The decisive uncertainties require proprietary consortium logs, privacy authorization, counterfactual replay, and eventually live feedback testing; they cannot be resolved by further ordinary web search alone."}}