{"schema_version":1,"assessment_id":"eoa_inverse_innovation_exp03_opportunity320_20260801","source_experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"deadweight_loss_reduction__data_science","archetype_slug":"deadweight_loss_reduction","domain_slug":"data_science","title":"Guardrailed borrowing of stranded team compute capacity","opportunity_summary":"Evaluate whether revocable borrowing of compatible, unprotected team-quota slack can reduce queue delay and increase completed priority-weighted analytical work while preserving minimum access, security, residency, reliability, reservation, and recall protections. The opportunity is conditional on first demonstrating meaningful overlap between idle eligible capacity and runnable workloads blocked specifically by quota ownership.","adopter_authorizer":"The institutional compute-governance body is the authorizer; platform or cluster operators are the implementing adopters, while security, privacy, data, and infrastructure-risk owners retain mandate-specific vetoes.","scores":{"meaningful_impact":{"score":4,"rationale":"If the proposed overlap exists, using otherwise idle compatible accelerators for queued priority work could improve completed analytical work and delay without acquiring hardware. The magnitude remains uncertain because the packet provides no evidence about frequency, duration, or organizational scale of the overlap."},"stakeholder_pull":{"score":2,"rationale":"The packet identifies affected teams, operators, governance bodies, and downstream users, but supplies no interviews, requests, incident history, adoption commitments, or other evidence that these stakeholders currently prioritize the problem."},"incremental_advantage":{"score":4,"rationale":"Relative to fixed non-borrowable quotas, the mechanism directly releases eligible slack; relative to the stated work-conserving rival, it adds minimum guarantees, eligibility controls, recall rights, incidence review, halt rules, and sunset. Whether these protections deliver a material net advantage must still be tested."},"distinctiveness_plausibility":{"score":2,"rationale":"The composition is internally coherent and testable, but prior art is explicitly unsearched. The packet therefore cannot establish whether guardrailed quota borrowing is distinctive from existing scheduling practices."},"technical_implementability":{"score":4,"rationale":"A one-pool intervention limited to compatible, checkpointable, non-production jobs, with a capacity cap and disable switch, is technically bounded. Implementability still depends on local scheduler hooks, workload metadata, checkpoint behavior, observability, and enforceable isolation rules that are not demonstrated."},"adoption_authority_feasibility":{"score":4,"rationale":"The proposal names compute governance as authorizer, operators as enforcers, and security and data owners as veto holders, creating a plausible decision path. Actual approval, operator capacity, and agreement on thresholds remain unverified."},"evidence_readiness":{"score":4,"rationale":"The candidate supplies separate problem and intervention falsifiers, observable workload states, comparison periods, unaffected pools, outcome classes, and rollback conditions. Readiness is reduced by undefined matching variables, estimand, contamination controls, and quantitative success and harm thresholds."},"safety_net_benefit":{"score":5,"rationale":"The design preserves minimum guarantees and protected reservations, excludes production and emergency capacity, respects security and residency controls, caps borrowing, provides recall and immediate rollback, and automatically sunsets. These safeguards directly address the proposal's main operational and distributional risks."},"scalability":{"score":3,"rationale":"Scheduler-based borrowing could in principle be repeated across pools, but hardware heterogeneity, workload checkpointability, data locality, security boundaries, institutional priorities, and pool-specific service guarantees may require substantial local adaptation."}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"10K_TO_50K","scope":"A bounded retrospective workload-log analysis and shadow scheduler replay for one pool, comparing observed fixed-quota outcomes with eligible guardrailed borrowing and reporting overlap, queue delay, completed-work proxies, interruptions, and access concentration without changing live allocation.","confidence":"MODERATE","assumptions":["Existing logs identify quota ownership, job readiness, resource compatibility, reservations, priority, and completion outcomes.","A small platform and data-analysis team can obtain approved access and complete the analysis in several weeks.","No new production scheduler integration or live workload exposure is included."]},"initial_deployment_startup":{"band_2026_usd":"50K_TO_250K","scope":"Engineer, security-review, instrument, and evaluate a reversible four-week pilot in one compute pool, including eligibility enforcement, borrowing caps, minimum guarantees, recall, halt automation, dashboards, and preregistered analysis.","confidence":"LOW","assumptions":["The local scheduler exposes suitable extension or configuration points.","Checkpointable non-production jobs can be identified reliably.","Existing monitoring and access-control systems can be extended rather than replaced.","The band excludes hardware purchases and institution-wide rollout."]},"operational_launch":{"band_2026_usd":"250K_TO_1M","scope":"Productionize the mechanism across multiple approved pools within one institution, including reliability engineering, policy configuration, security and compliance validation, operator training, governance review, documentation, and causal evaluation.","confidence":"LOW","assumptions":["The pilot supports continuation and does not reveal a need for major scheduler replacement.","Pools share enough scheduler and telemetry infrastructure for partial reuse.","Launch retains exclusions for incompatible, protected, production, and emergency-reserved capacity."]},"annual_recurring":{"band_2026_usd":"50K_TO_250K","scope":"Ongoing monitoring, operator support, incident response, policy and eligibility reviews, fairness and concentration audits, telemetry maintenance, and periodic outcome evaluation for an institutional deployment.","confidence":"LOW","assumptions":["No dedicated new hardware is required.","The mechanism becomes part of an existing platform team's responsibilities.","Governance reviews occur periodically rather than requiring continuous manual approval of each borrowing event."]}},"research_burden":"HIGH","earliest_credible_horizon":"3_TO_12_MONTHS","pipeline_gates":{"recognizable_externally_supportable_problem":{"status":"YES","reason":"The packet defines an observable and falsifiable problem: compatible, unprotected capacity is idle during the same windows that runnable jobs wait specifically because quotas are non-borrowable. External measurement is still needed to establish prevalence and magnitude."},"identifiable_adopter_or_authorizer":{"status":"YES","reason":"The institutional compute-governance body is identified as experiment authorizer, platform operators as implementers, and security and data owners as mandate-specific veto holders."},"distinct_testable_incremental_claim":{"status":"YES","reason":"The proposal claims that capped, revocable borrowing with guarantees and recall will outperform fixed quotas on queue delay or completed priority-weighted work without exceeding declared safety and distributional bounds; this is separable from the stated unguarded work-conserving rival."},"bounded_next_evidence_step":{"status":"YES","reason":"A one-pool retrospective analysis and shadow replay can compare observed fixed-quota allocation with simulated eligible borrowing and can falsify the opportunity if meaningful compatible overlap or outcome improvement is absent."},"no_unresolved_safety_or_authority_stop":{"status":"UNCERTAIN","reason":"The packet provides exclusions, vetoes, halt rules, rollback, and sunset, but actual governance authorization, security validation, service thresholds, interruption limits, and concentration bounds have not been obtained or operationally defined."},"implementation_cost_scope_and_range":{"status":"YES","reason":"The candidate bounds initial implementation to one pool, four weeks, capped borrowing, eligible non-production jobs, instrumentation, and rollback, permitting broad resource bands. Local scheduler architecture and compliance effort could still move realized costs across bands."}},"blocking_evidence":["Workload-level evidence that meaningful simultaneous overlap exists between idle compatible unprotected capacity and runnable jobs blocked specifically by quota ownership.","Evidence that observed waiting is not predominantly caused by true scarcity, hardware incompatibility, maintenance, data movement, locality, dependencies, or protected reservations.","Shadow-replay evidence that the eligibility, cap, guarantee, and recall rules improve queue delay or completed-work proxies relative to fixed quotas without unacceptable interruption or access concentration.","Locally authorized quantitative definitions for priority-weighted work, meaningful improvement, minimum access, reliability, interruption, security, and concentration limits.","Confirmation that scheduler controls, workload metadata, checkpointing behavior, monitoring, and security boundaries can enforce the proposed safeguards.","Governance and mandate-owner approval for any later live pilot."],"next_evidence_step":"Using four to eight weeks of historical data from one candidate pool, preregister eligibility and protected-capacity rules, then run a workload-level diagnostic and shadow scheduler replay comparing observed fixed quotas against capped guardrailed borrowing and, where feasible, the stated conventional work-conserving rival. Stop before live deployment. Reject or redesign the opportunity if there is no material compatible quota-blocked overlap, no improvement in queue delay or completed-work proxies, or simulated interruption, guarantee violations, or access concentration exceeds predeclared bounds.","research_questions":["How much compatible, unprotected compute-time is idle while runnable jobs are blocked specifically by quota ownership?","Which workload, hardware, reservation, locality, security, and checkpointability attributes are necessary to classify borrowing eligibility reliably?","What estimand, matching variables, outcome window, spillover treatment, and contamination controls would support causal attribution in a later pilot?","How does guardrailed borrowing compare with both fixed quotas and the stated conventional work-conserving scheduler on completed priority-weighted work, wait time, interruption, and distribution of access?","What minimum-access, reliability, security, interruption, recall-latency, and concentration thresholds will the authorized governance body accept?","Would strategic request inflation, hoarding, rebound demand, or unequal capture of slack erase or redistribute the expected benefit?","What scheduler engineering, monitoring, compliance, and operator effort is required locally?","Does external prior-art review show that the proposed composition differs materially from existing fair-share, quota-borrowing, backfill, or work-conserving scheduling approaches?"],"recommendation":"VALIDATE_PROBLEM_FIRST","uncertainty_constraints":["No external evidence establishes the prevalence, duration, or value of simultaneous compatible idle capacity and quota-blocked demand.","The packet provides no stakeholder-demand or adoption-commitment evidence.","Priority-weighted work and acceptable harm thresholds contain governance judgments that are not yet operationalized.","Local scheduler capabilities, telemetry quality, checkpoint reliability, security boundaries, and integration effort are unknown.","The causal comparison lacks preregistered matching variables, estimand, spillover handling, and contamination controls.","Pilot workloads may not represent production workloads or other compute pools.","Costs are broad resource-equivalent planning bands, not observed quotations or point estimates.","Novelty, duplication, market size, and realized impact cannot be inferred from the sealed packet."],"closed_book_prior_art_boundary":"Prior art was not searched and remains unverified; this closed-book assessment makes no claim that the mechanism or its composition is novel, uncommon, or absent from existing fair-share, quota-borrowing, backfill, or work-conserving scheduling practices."}