{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"deadweight_loss_reduction__data_science","trajectory_id":"R","attempt_index":0,"archetype_slug":"deadweight_loss_reduction","domain_slug":"data_science","decision":"CANDIDATE","problem_id":"rigid_team_compute_quotas_strand_capacity","causal_lever_id":"guardrailed_work_conserving_compute_reallocation","proposal":{"problem":"Rigid team-level accelerator or cluster quotas can leave compatible capacity idle in one allocation while valuable, ready workloads assigned elsewhere remain queued. The independently recognizable problem is quota-induced compute underutilization and delay, not merely insufficient hardware.","actors_substrate":["data scientists and research teams submitting workloads","platform and cluster operators","research or product governance setting priorities","security, privacy, and infrastructure-risk owners","downstream users awaiting analytical outputs"],"observable_state":"HYPOTHESIS: During the same measurement windows, quota owners have unused compatible compute while runnable jobs outside their quotas wait, downscale, or move to inferior substitutes; manual exceptions are slow or inconsistently granted.","consequence":"INFERENCE: If compatibility and reservation needs do not explain the overlap, a stale allocation rule destroys usable compute-time and delays analytical learning without producing an offsetting protection.","affected_objective":"Increase completed priority-weighted analytical work and reduce compatible-capacity idleness and queue delay while preserving minimum access, security, data-residency, reliability, and legitimate priority commitments.","structural_mapping":[{"archetype_element":"value-blocking wedge","domain_realization":"Non-borrowable team quotas prevent a scheduler from assigning temporarily idle compatible capacity to queued workloads.","claim_kind":"HYPOTHESIS"},{"archetype_element":"blocked beneficial activity","domain_realization":"Ready experiments, training runs, and analytical pipelines cannot use otherwise available compute.","claim_kind":"HYPOTHESIS"},{"archetype_element":"protected purpose","domain_realization":"Quotas may guarantee fair access, reserved headroom, budget control, isolation, priority work, and operational safety.","claim_kind":"INFERENCE"},{"archetype_element":"less-distortive redesign","domain_realization":"A work-conserving shared pool lends unused quota subject to compatibility, minimum guarantees, priority rules, preemption notice, and automatic return.","claim_kind":"HYPOTHESIS"},{"archetype_element":"incidence and rebound","domain_realization":"Borrowers gain throughput, while quota owners may face reduced burst readiness and participants may inflate requests or fragment jobs.","claim_kind":"HYPOTHESIS"}],"component_map":[{"component":"Distortion Map","status":"direct","domain_realization":"Trace fixed quota, idle compatible capacity, blocked runnable jobs, and manual-exception paths."},{"component":"Protected Constraint Safeguard","status":"direct","domain_realization":"Retain security, residency, isolation, reliability headroom, priority reservations, and minimum team guarantees."},{"component":"Surplus Estimate","status":"adapted","domain_realization":"Estimate recoverable compatible GPU-hours, avoided queue-hours, and priority-weighted jobs completed, with uncertainty."},{"component":"Affected-Party Incidence Map","status":"direct","domain_realization":"Record effects on lending teams, borrowing teams, operators, risk owners, and downstream users."},{"component":"Redesign Lever","status":"direct","domain_realization":"Create revocable borrowing from unused quotas through a guardrailed work-conserving scheduler."},{"component":"Distributional Review","status":"direct","domain_realization":"Test whether small or low-status teams lose predictable access or whether favored teams capture slack."},{"component":"Behavioral Response Model","status":"adapted","domain_realization":"Model request inflation, strategic reservation, job splitting, demand rebound, and preemption avoidance."},{"component":"Implementation Boundary","status":"direct","domain_realization":"Limit the pilot to compatible non-production workloads on one pool, with capped borrowed capacity."},{"component":"Monitoring and Rebound Check","status":"direct","domain_realization":"Monitor utilization, waits, completions, preemptions, failures, access concentration, and request inflation."},{"component":"Rollback or Adjustment Rule","status":"direct","domain_realization":"Disable borrowing and restore original quotas when safety, reliability, or access thresholds fail."},{"component":"Cost–Benefit Assessment Frame","status":"adapted","domain_realization":"Compare recovered work and reduced delay against engineering, interruption, reliability, and distributional costs."},{"component":"Price-Wedge Diagnostic","status":"incompatible","domain_realization":"No monetary or administered-price wedge is posited; allocation eligibility is the binding candidate cause."},{"component":"Friction Source Breakdown","status":"adapted","domain_realization":"Separate quota blocking from hardware incompatibility, data locality, reservations, maintenance, and true scarcity."},{"component":"Compensating Adjustment Plan","status":"adapted","domain_realization":"Give lenders minimum guarantees, recall rights, preemption notice, and service credits rather than monetary compensation."},{"component":"Legitimacy and Authority Review","status":"direct","domain_realization":"Confirm who may lend, recall, prioritize, and reassign institutional compute."},{"component":"Sensitivity Analysis","status":"direct","domain_realization":"Vary workload value weights, compatibility assumptions, reservation levels, and interruption costs."},{"component":"Pilot or Sunset Path","status":"direct","domain_realization":"Run a time-limited pilot that automatically returns to fixed quotas unless evidence supports renewal."}],"mechanism_dispositions":[{"slug":"congestion_or_capacity_pricing_adjustment","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Monetary willingness may track budget rather than analytical value and would alter the problem into price-based rationing.","counterfactual_removal":"No change; pricing is absent from the causal chain."},{"slug":"cost_benefit_assessment_protocol","disposition":"selected_supporting","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Compare recovered compute and delay with reliability, implementation, and distributional costs under sensitivity ranges.","counterfactual_removal":"The pilot could run, but its net-value decision would be poorly grounded."},{"slug":"distortion_reduction_review","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Establish that quota rules, rather than incompatibility or protected reservations, cause simultaneous idleness and waiting.","counterfactual_removal":"Removing it permits misdiagnosing genuine scarcity or safety headroom as waste."},{"slug":"impact_assessment_table","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Track gains, losses, protections, and halt signals by team and affected role.","counterfactual_removal":"Aggregate utilization could conceal concentrated access loss."},{"slug":"matching_improvement_program","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Adapt its matching logic so the scheduler pairs eligible queued jobs with compatible lent capacity; market-thickening claims are omitted.","counterfactual_removal":"Quota borrowing would lack an operational method for assigning heterogeneous capacity."},{"slug":"permit_or_approval_streamlining","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Manual exceptions may be symptomatic, but the candidate binding wedge is non-borrowable allocation, not approval duplication.","counterfactual_removal":"No material change to the proposed quota-to-pool causal chain."},{"slug":"price_control_redesign","disposition":"incompatible","contribution_type":"NONE","adaptation_or_rejection":"There is no price cap, floor, or subsidy to reset or compensate.","counterfactual_removal":"No change."},{"slug":"quota_or_allocation_rule_review","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Keep the total capacity constraint while replacing non-borrowable assignments with revocable, rule-governed lending.","counterfactual_removal":"The specific wedge remains intact, so compatible idle capacity cannot serve blocked jobs."},{"slug":"regulatory_simplification_pilot","disposition":"selected_load_bearing","contribution_type":"TEST_DESIGN","adaptation_or_rejection":"Translate regulatory cases into a bounded workload pool with monitoring, capped exposure, and automatic expiry.","counterfactual_removal":"Testing would expose the whole compute estate or provide no credible causal evidence; bounded viability is lost."},{"slug":"sunset_clause_review","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Apply expiry to the pilot and require evidence for continuation, while leaving standing security controls untouched.","counterfactual_removal":"The experimental allocation could persist through inertia, though the immediate reallocation mechanism still functions."},{"slug":"tariff_fee_or_toll_redesign","disposition":"incompatible","contribution_type":"NONE","adaptation_or_rejection":"No authority-imposed charge is identified as the wedge.","counterfactual_removal":"No change."}],"causal_chain":["Fixed, non-borrowable team quotas partition short-run compute capacity.","HYPOTHESIS: Demand and usage vary enough that compatible capacity is sometimes idle while outside jobs wait.","Protected reservations and incompatibilities are separated from avoidable allocation blocking.","A scheduler temporarily lends eligible slack while preserving guarantees and recall rights.","Queued compatible jobs use capacity that otherwise would remain idle.","INFERENCE: If completed priority-weighted work rises without unacceptable access or reliability harm, avoidable allocation loss falls."],"baseline":"Fixed team quotas with unused allocations retained by the owner and ad hoc manual exceptions for blocked jobs.","nearest_rival":"A conventional work-conserving priority scheduler that ignores team quotas. It may improve utilization but lacks minimum guarantees, incidence review, recall rights, and a bounded test of the quota’s protective purpose.","authority_safety":{"affected_parties":["quota-owning teams","teams seeking borrowed capacity","cluster operators","security and privacy owners","research-priority governance","downstream users"],"decision_authority":"The institutional compute-governance body may authorize the experiment; platform operators enforce eligibility and halt rules, and data/security owners retain vetoes within their mandates.","authorized_first_step":"For four weeks, enable revocable borrowing for compatible, checkpointable, non-production jobs in one pool, cap borrowed capacity, preserve minimum guarantees, and compare with matched pre-pilot periods and contemporaneous unaffected pools.","excluded_actions":["selling or auctioning compute access","overriding data-residency, security, or isolation controls","borrowing production or emergency-reserved capacity","changing institutional research priorities without governance approval","permanent or estate-wide rollout from pilot evidence alone"],"halt_rollback":"Immediately disable borrowing and restore fixed quotas if security boundaries fail, protected workloads miss predeclared service thresholds, failure or interruption exceeds limits, or access concentration crosses the approved bound; otherwise sunset automatically after four weeks pending review."}},"negative_tests":{"strongest_counterevidence":"Measured 'idle' capacity may be incompatible, under maintenance, deliberately reserved for burst or incident response, or unusable by queued jobs; waiting jobs may also be low-priority, non-checkpointable, or bottlenecked elsewhere.","analogy_break":"Compute allocation is not ordinary exchange: hardware is heterogeneous, workloads have uncertain value, preemption can destroy work, and governance priorities cannot safely be reduced to willingness to pay or aggregate utilization.","failure_condition":"The intervention fails if borrowing increases nominal utilization but not completed priority-weighted work, or if gains depend on violating reservations, fairness guarantees, or operational constraints.","problem_falsifier":"After workload-level measurement, no meaningful overlap exists between idle compatible unprotected capacity and runnable jobs blocked specifically by quota ownership; observed delay is instead explained by true scarcity, incompatibility, data movement, or dependencies.","intervention_falsifier":"Such overlap exists, but guardrailed borrowing produces no material improvement in wait time or completed work, or causes predeclared reliability, security, interruption, or distributional harm beyond the accepted bound.","risks":["request inflation or strategic hoarding","advantaged teams capturing most slack","preemption wasting experiments","hidden data-residency or isolation violations","rebound demand recreating congestion","priority-weight choices embedding contested values","pilot workloads being unrepresentative"]},"null_rationale":null,"classification":{"candidate_kind":"MECHANISM_COMPOSITION","prior_art_status":"UNSEARCHED","evidence_maturity":"HYPOTHESIS"},"revision_change_log":{"revision_kind":"ORIGINAL","prior_problem_id":null,"prior_causal_lever_id":null,"problem_changed":false,"causal_lever_changed":false,"conceptual_changes":[],"operational_changes":[],"repairs_addressed":[]},"confidence":0.84,"generator_notes":"Closed-book structural inference from the supplied packet. The candidate is conditional on demonstrating simultaneous compatible idle capacity and quota-blocked runnable demand; no novelty or empirical prevalence claim is made."}