{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"invariant_mode_decomposition_design__library_information_science","considered_ids":["H1","H4","H5"],"selected_id":"H4","dispositions":[{"hypothesis_id":"H1","lane":"RESEARCH","reason":"The rare-class residual gate is falsifiable and adjacent to established stratified migration validation, but problem support is only moderate and the added SVD mechanism is less directly tied to a demonstrated library need."},{"hypothesis_id":"H4","lane":"FINALIST","reason":"Despite substantial component collision, the critic identifies a material, testable residual: rolling spectral lender dependency confirmed by fulfillment-specific node-loss replay and used to trigger diversification under a median-fulfillment-time constraint. Strong problem support and deployed non-spectral comparators make the distinction recognizable and falsifiable."},{"hypothesis_id":"H5","lane":"COLLISION","reason":"Generic eigenspace monitoring, thresholded alarms, temporal graph change detection, and longitudinal vocabulary maintenance are already well represented. The residual is primarily an application-and-validation claim with only moderate problem support."}],"selection_reason":"H4 offers the strongest supported problem and the clearest bounded experiment against direct operational comparators: test whether spectral dependency plus node-loss replay improves lender resilience without violating a fulfillment-time constraint. Its residual claim is material and separable from the collided load-balancing and spectral-analysis components."}