{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"negative_space_design__tech_ethics_ai_governance","hypothesis_id":"H5","search_queries":["AI governance portfolio \"reserved capacity\" red teaming evaluation","AI portfolio allocate percentage capacity responsible AI testing non-deployment","innovation portfolio ring fence capacity discovery experiments percentage","regulatory sandbox reserve resources AI testing before deployment","\"capacity allocation\" percentage \"innovation\" portfolio management","\"reserve\" \"15%\" capacity innovation team experiments","site:scaledagileframework.com capacity allocation percentage enablers innovation","AI governance budget percentage red team evaluation capacity allocation"],"sources":[{"source_id":"C1","title":"Govern — NIST AI RMF Playbook","publisher":"National Institute of Standards and Technology","url":"https://airc.nist.gov/airmf-resources/playbook/govern/","source_class":"OFFICIAL_GUIDANCE","claims_supported":["AI risk-management resources are finite and are ordinarily allocated according to assessed system risk.","The playbook addresses executive responsibility for AI portfolios and recommends independent red-teaming, effective challenge, and organizational policies supporting testing and course correction."]},{"source_id":"C2","title":"People process technology and operations framework for establishing AI governance in healthcare organizations","publisher":"npj Digital Medicine","url":"https://www.nature.com/articles/s41746-026-02419-6","source_class":"PRIMARY_RESEARCH","claims_supported":["The framework recommends an initial AI-governance budget covering a predefined period or number of projects and states that sustaining governance should account for 10–15% of the total governance budget.","It includes dedicated quantitative assessment, secure development and validation environments, lifecycle decision points, and patient-community input."]},{"source_id":"C3","title":"Transitioning to Value Stream Funding Competency","publisher":"Scaled Agile, Inc.","url":"https://framework.scaledagile.com/lean-portfolio-management-discipline/transitioning-to-value-stream-funding-competency/","source_class":"COMMERCIAL_FIRST_PARTY","claims_supported":["Lean portfolio management funds fixed organizational capacity rather than temporary projects and governs it through portfolio-level financial guardrails.","The practice is explicitly aimed at portfolio leadership and balancing responsiveness with rigorous financial stewardship."]},{"source_id":"C4","title":"Horizontal and vertical capacity allocation in SAFe","publisher":"KEGON AG","url":"https://www.kegonacademy.com/en/knowledge/blog/detail/horizontale-und-vertikale-capacity-allocation-in-safe","source_class":"AUTHORITATIVE_SECONDARY","claims_supported":["Capacity-allocation practice divides team or portfolio capacity into quotas before planning, including exploration and compliance categories.","Its stated purpose is to prevent urgent delivery work from consuming all capacity; quotas are governed across organizational levels and periodically reviewed."]},{"source_id":"C5","title":"Regulation (EU) 2024/1689 — Artificial Intelligence Act","publisher":"European Union","url":"https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng","source_class":"GOVERNMENT_OR_REGULATOR","claims_supported":["Article 57 establishes sufficiently resourced, controlled, limited-time environments for development, testing, and validation before market placement or service.","Sandbox activities operate under an agreed plan, include risk-identification and mitigation support, and produce exit reports describing activities, results, and learning outcomes."]},{"source_id":"C6","title":"NayaOne’s AI Sandbox","publisher":"UK Department for Science, Innovation and Technology","url":"https://www.gov.uk/ai-assurance-techniques/nayaones-ai-sandbox","source_class":"OFFICIAL_GUIDANCE","claims_supported":["The documented product provides a secure environment disconnected from production and organizational networks for evaluating and procuring AI models.","It supports stress, bias, performance, explainability, and hallucination testing and presents sandboxing as a consistent governance and risk-evaluation process."]},{"source_id":"C7","title":"OpenAI's Approach to External Red Teaming for AI Models and Systems","publisher":"OpenAI","url":"https://arxiv.org/abs/2503.16431","source_class":"PRIMARY_RESEARCH","claims_supported":["External red-teaming is an established predeployment assurance activity used to discover novel risks, stress-test mitigations, and create new safety measurements."]}],"proximity":"SUBSTANTIAL_COLLISION","closest_analogues":[{"name":"PPTO framework’s 10–15% AI-governance sustainment budget","similarity":"It already combines organizational AI governance, a bounded percentage allocation, a predefined planning horizon or project count, dedicated evaluation capability, lifecycle gates, and community input.","remaining_difference":"Its percentage applies to sustaining and operating the governance system rather than exclusively reserving AI pilot capacity for nondeployment inquiry, and it supplies no release rule for converting unused reserve into deployment capacity.","source_ids":["C2"]},{"name":"Lean-Agile portfolio capacity allocation","similarity":"It uses advance portfolio quotas and cross-level governance to protect exploration and compliance work from being displaced by urgent delivery—the same capacity-ring-fencing mechanism.","remaining_difference":"It is a general portfolio-management practice rather than an AI-harm inquiry rule, and the reviewed sources do not prescribe a 15% nondeployment category or harm-detection outcome metric.","source_ids":["C3","C4"]},{"name":"AI regulatory and technical sandboxes","similarity":"They protect bounded, resourced environments for testing, validation, risk discovery, mitigation, and documented learning before deployment or market entry.","remaining_difference":"They allocate an environment per program or participant, not a fixed share of an organization's annual pilot portfolio, and their exit criteria concern sandbox participation rather than release of reserved portfolio capacity.","source_ids":["C5","C6"]},{"name":"Risk-tiered AI governance and external red-teaming","similarity":"They establish the proposed nondeployment activities, dedicated challenge functions, finite-resource allocation, and predeployment discovery of novel risks.","remaining_difference":"Resources are assigned by system risk or engagement rather than through a portfolio-wide minimum reserve protected before individual deployment approvals.","source_ids":["C1","C7"]}],"overlapping_components":["Advance allocation of bounded organizational capacity","Portfolio-level guardrails and quota governance","Protected exploration, compliance, or evaluation work","Dedicated AI-governance budget","Predeployment sandbox or shadow environment","Independent red-teaming and effective challenge","Risk, bias, performance, and stress evaluation","Stakeholder or affected-community input","Documented entry, exit, and review processes","Periodic quota or resource review"],"remaining_contrastive_claim":"Unlike published governance budgets, capacity-allocation practices, and sandboxes, the treatment reserves at least 15% of an organization's annual AI pilot capacity exclusively for nondeployment harm inquiry before individual approvals and releases that capacity only through explicit portfolio-level criteria.","claim_falsifier":"A pre-existing policy, standard, patent, product manual, or organizational case study would falsify this contrast if it specified a minimum percentage of an annual AI pilot portfolio reserved against live deployment for red-teaming, shadow evaluation, or affected-community inquiry and included explicit release or reallocation criteria.","problem_support":"MODERATE","recommendation":"REJECT","world_novelty_boundary":"Across eight ordinary-web queries and seven opened sources, no direct implementation of the full pilot-capacity denominator plus exclusive nondeployment restriction plus release criteria was found; however, the treatment is substantially assembled from an existing 10–15% AI-governance budget recommendation, established portfolio quota protection, and established predeployment sandbox and red-team practices. No patent surfaced in this bounded search, which is not a patent-clearance conclusion."}