{"schema_version":1,"assessment_id":"eoa_inverse_innovation_exp03_opportunity320_20260801","source_experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"invariant_mode_decomposition_design__library_information_science","archetype_slug":"invariant_mode_decomposition_design","domain_slug":"library_information_science","title":"Offline modal-control testing for coupled discovery-exposure imbalance","opportunity_summary":"Test whether cross-cycle combinations of subject, format, branch, or creator-group exposure deviations persist or grow in digital-library discovery systems, and whether mode-targeted ranking controls outperform ordinary monitoring and direct per-stratum constraints without breaching utility, privacy, representation, or governance limits.","adopter_authorizer":"A library's accountable discovery-policy owner, supported by ranking engineers and discovery librarians and subject to privacy, accessibility, and collection-governance review.","scores":{"meaningful_impact":{"score":4,"rationale":"If the proposed coupled underexposure exists, it could systematically narrow discovery for multiple resource strata while aggregate engagement and individual-stratum alerts remain acceptable. The magnitude and prevalence of this harm are not established in the sealed packet."},"stakeholder_pull":{"score":2,"rationale":"The packet identifies affected patrons, represented creators, librarians, and ranking-service operators, but supplies no evidence that any library has recognized this cross-cycle problem, requested modal analysis, or committed data and staff."},"incremental_advantage":{"score":3,"rationale":"Modeling coupled cycle-to-cycle dynamics could reveal persistent combinations missed by aggregate or one-stratum monitoring, and the proposed replay directly compares this leverage with ordinary practice and direct constraints. Whether it improves equity and utility out of sample is wholly untested."},"distinctiveness_plausibility":{"score":2,"rationale":"The composition of fitted update operators, modal decomposition, sensitivity replay, and governance constraints is internally differentiated from the stated rivals, but prior art is explicitly unsearched and world novelty is unmeasured."},"technical_implementability":{"score":3,"rationale":"The state vector, fitted Jacobian, six-cycle fit, two-cycle validation, controls, residual checks, and rollback logic form a technically testable workflow. Reliability may fail because eight cycles can be insufficient for the state dimension or because the operator is nonlinear, nonstationary, or ill-conditioned."},"adoption_authority_feasibility":{"score":4,"rationale":"The accountable discovery-policy owner and required privacy, accessibility, and collection-governance reviews are explicitly identified, while analysts are barred from deployment. Actual access to logs, engineering capacity, and institutional approval remain unconfirmed."},"evidence_readiness":{"score":4,"rationale":"A bounded, preregistered offline comparison is specified with held-out validation and separate problem and intervention falsifiers. Readiness is reduced because residual, utility, conditioning, spectral-gap, representation, and mode-alignment thresholds have not been operationally fixed."},"safety_net_benefit":{"score":4,"rationale":"The proposal keeps the existing ranking configuration, prohibits patron-facing changes in the first step, excludes protected-trait inference and individual mode scores, and specifies halt conditions. Residual privacy and representational risks depend on stratum granularity and review quality."},"scalability":{"score":3,"rationale":"The same offline fitting and monitoring architecture could in principle be repeated across discovery systems, but each library may require local strata, policy-approved controls, data pipelines, thresholds, and governance review; operator stability across settings is unknown."}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"50K_TO_250K","scope":"Preregister, prepare and privacy-review eight cycles of deidentified logs, fit and validate the modal model, replay permissible controls, compare against the ordinary baseline and direct-constraint rival, and document equity, utility, residual, conditioning, and representation results.","confidence":"LOW","assumptions":["Existing cycle-level logs contain usable exposure, engagement, catalog-availability, and query-mix fields.","No new patron data collection or production-system change is required.","The state dimension is small enough for analysis by a compact library, engineering, privacy, and evaluation team.","The band includes labor, data preparation, governance coordination, computing, and evaluation."]},"initial_deployment_startup":{"band_2026_usd":"250K_TO_1M","scope":"If offline evidence passes, integrate bounded controls and monitoring into a staging or production-adjacent ranking workflow, establish approvals and audit records, test rollback, and validate accessibility, privacy, utility, and collection-policy constraints before any separately authorized pilot.","confidence":"LOW","assumptions":["The library controls or can modify its ranking service.","Existing deployment, logging, and rollback infrastructure can be extended rather than replaced.","No procurement of an entirely new discovery platform is required.","A later pilot receives separate governance authorization."]},"operational_launch":{"band_2026_usd":"250K_TO_1M","scope":"Conduct a separately authorized limited launch with monitoring for residual error, subgroup error, spectral gap, conditioning, drift, utility, representation, and protected harm, including evaluation and rollback readiness.","confidence":"LOW","assumptions":["A limited launch is permitted only after successful offline validation.","Launch requires ranking-engineering, librarian, privacy, accessibility, governance, and evaluation participation.","The control remains bounded and does not introduce individual patron mode scoring or unapproved personalization.","Existing service capacity can support the added computation and monitoring."]},"annual_recurring":{"band_2026_usd":"50K_TO_250K","scope":"Maintain data pipelines, periodically refit and validate the operator, monitor drift and safety thresholds, conduct governance reviews, investigate alerts, and preserve rollback capability.","confidence":"LOW","assumptions":["Monitoring is integrated into an existing discovery service.","Retraining and review occur periodically rather than continuously.","Recurring cost is dominated by specialist labor and governance rather than major new hardware.","Material catalog, policy, or ranking changes trigger additional validation."]}},"research_burden":"HIGH","earliest_credible_horizon":"3_TO_12_MONTHS","pipeline_gates":{"recognizable_externally_supportable_problem":{"status":"UNCERTAIN","reason":"The candidate defines observable cross-cycle exposure propagation and a clear falsifier, but provides no external or empirical evidence that consequential coupled modes occur in an actual library system."},"identifiable_adopter_or_authorizer":{"status":"YES","reason":"The library's accountable discovery-policy owner is explicitly named as decision authority, with privacy, accessibility, and collection-governance review."},"distinct_testable_incremental_claim":{"status":"YES","reason":"The proposal claims that modal modeling can detect and damp coupled cross-cycle exposure deviations better than ordinary monitoring or direct per-stratum constraints, and specifies held-out comparisons that can reject that claim."},"bounded_next_evidence_step":{"status":"YES","reason":"The sealed candidate authorizes a preregistered, deidentified, eight-cycle offline test using six cycles for fitting and two for validation, with no patron-facing changes."},"no_unresolved_safety_or_authority_stop":{"status":"YES","reason":"The first step is offline, preserves the existing ranking, reserves deployment authority to accountable governance, excludes protected-trait inference and individual scoring, and contains explicit halt conditions. Thresholds must be fixed before execution but do not require live exposure."},"implementation_cost_scope_and_range":{"status":"UNCERTAIN","reason":"The candidate identifies the principal analytical, integration, monitoring, governance, and rollback activities, permitting broad resource bands, but gives no institution-specific data condition, staffing, platform architecture, procurement, or coordination evidence."}},"blocking_evidence":["Whether exposure deviations propagate across cycles after controlling for catalog availability and contemporaneous query mix.","Whether any consequential modes reproduce in the two held-out cycles with acceptable conditioning, spectral separation, alignment, and residual error.","Whether mode-targeted controls outperform both ordinary practice and direct per-stratum constraints on held-out equity and retrieval utility.","Whether available logs support policy-approved strata without privacy leakage, sparse-group instability, or masking catalog omissions.","Whether a prospective adopter recognizes the problem and can supply the required data, engineering capacity, and governance approvals.","Whether the composition is materially distinct from existing research or practice; prior art is unsearched."],"next_evidence_step":"Before execution, preregister numerical decision rules, then run the authorized offline test on eight deidentified update cycles: fit on six and validate on two while controlling for catalog availability and query mix; compare the modal model and control replay with one-step stratum monitoring and direct per-stratum constraints. Reject further intervention work if propagation is not reproducible, modal prediction does not improve held-out explanation, controls do not improve equity without violating the utility floor, or residual, conditioning, spectral-gap, representation, privacy, or subgroup-error limits fail.","research_questions":["Do exposure deviations propagate across update cycles after catalog availability and query mix are controlled?","Are fitted consequential modes stable, well-conditioned, spectrally separated, and reproducible out of sample?","What residual budgets, utility floors, subgroup-error limits, conditioning bounds, spectral-gap rules, and mode-alignment tolerances are decision-relevant and governable?","Does modal control replay improve held-out equity and utility relative to ordinary monitoring and direct per-stratum constraints?","Are apparent ranking effects instead attributable to catalog omissions, policy interventions, or changing patron intent?","Can approved strata be measured with sufficient sample support while preserving privacy and avoiding trait reification?","Does an accountable library owner recognize the problem and have authority, data access, engineering capacity, and review pathways for further work?","What prior research and operational practice already cover dynamic exposure feedback, spectral auditing, recommender stability, and library discovery governance?"],"recommendation":"VALIDATE_PROBLEM_FIRST","uncertainty_constraints":["Closed-book assessment: no external evidence was used.","Problem prevalence, stakeholder demand, realized impact, market size, and world novelty are unmeasured.","All technical and cost judgments are broad resource-equivalent estimates rather than observed institution-specific results.","The eight-cycle design may be underpowered relative to the number and sparsity of approved strata.","The local linear operator may be invalidated by nonlinearity, catalog changes, policy changes, human intervention, or feedback from later deployment.","Modes must not be interpreted as intrinsic patron, creator, or community traits.","A successful offline result would not authorize live deployment or personalization."],"closed_book_prior_art_boundary":"Prior-art status is UNSSEARCHED in substance: the packet labels it UNSEARCHED and supplies no literature, product, patent, or practice evidence. The assessment therefore treats the proposal's structural differentiation from its stated baselines as testable internal plausibility only, not as novelty or freedom-to-operate evidence."}