{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp11_mechanism_context_external20_20260804","cell_id":"layer_decay_and_expiration_management__systems_cybernetics","judge_id":"J3","item_assessments":[{"opaque_id":"layer_decay_and_expiration_management__systems_cybernetics__A","supported_problem":3,"external_distinctiveness":3,"testability":5,"researchability":5,"evidence_quality":4,"fatal_issue":null},{"opaque_id":"layer_decay_and_expiration_management__systems_cybernetics__C","supported_problem":4,"external_distinctiveness":4,"testability":4,"researchability":4,"evidence_quality":5,"fatal_issue":null},{"opaque_id":"layer_decay_and_expiration_management__systems_cybernetics__B","supported_problem":3,"external_distinctiveness":2,"testability":4,"researchability":4,"evidence_quality":4,"fatal_issue":null}],"pairwise_comparisons":[{"pair_id":"A_vs_C","left_id":"layer_decay_and_expiration_management__systems_cybernetics__A","right_id":"layer_decay_and_expiration_management__systems_cybernetics__C","preference":"RIGHT","confidence":"MODERATE","rationale":"C has substantially stronger external evidence that uncontrolled temporary control changes persist and matter, and its three-arm replay tests behavioral consequences rather than only registry eligibility. Its attribution and local-mode requirements add feasibility risk, while management-of-change and Simplex are close precedents, but the bounded comparison still leaves a more consequential empirical question than A's largely governance-oriented lifecycle gate."},{"pair_id":"A_vs_B","left_id":"layer_decay_and_expiration_management__systems_cybernetics__A","right_id":"layer_decay_and_expiration_management__systems_cybernetics__B","preference":"LEFT","confidence":"MODERATE","rationale":"A offers a cleaner complete-family audit, explicit stale-eligibility denominator, identical-event replay, and decisive zero-baseline falsifier. B asks an important rare-regime preservation question, but longstanding multiple-model supervisory control already addresses regime mismatch and eligibility directly, leaving the lifecycle packaging as a narrower increment with less direct evidence that superseded artifacts remain exposed."},{"pair_id":"C_vs_B","left_id":"layer_decay_and_expiration_management__systems_cybernetics__C","right_id":"layer_decay_and_expiration_management__systems_cybernetics__B","preference":"LEFT","confidence":"HIGH","rationale":"C is supported by cross-jurisdictional official evidence about persistent temporary control changes and tests an attributable actuator-path consequence against both current practice and age-only expiry. B has weaker evidence for its specific failure mode and collides more directly with established multiple-model monitoring, switching, and regime-selection research."}],"overall_top_choice":"layer_decay_and_expiration_management__systems_cybernetics__C","overall_rationale":"C is the strongest research candidate because it joins the best-supported practical problem with a reversible, bounded, contrastive causal test of whether withdrawing one stale overlay's authority improves behavior without harming safety or recovery. Its main weakness is demanding identifiability and valid local diagnostics; those are genuine falsifiers and may terminate the study, but they do not create a fatal safety issue because the first step is offline and tightly halted. A is the best fallback due to its exceptionally clean audit design, while B's core selection problem is closest to mature supervisory-control prior art.","blinding_limitations":"The judgment relies on supplied summaries and source characterizations rather than independent inspection of the cited materials. Differences in proposal specificity and source selection may affect apparent quality, and public documentation may omit routine proprietary control practices."}