{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp09_archetype_breadth150_20260804","cell_id":"load_shedding__chemistry_materials","arm":"BREADTH_PROBE_ONE_SHOT","candidate_id":"load_shedding__chemistry_materials__P1","proposal_index":1,"version":0,"title":"Protected-capacity experiment shedding for an overloaded autonomous materials laboratory","problem":"An autonomous materials-synthesis laboratory can admit more proposed recipes than its ventilated reactors, inert-atmosphere transfer stations, and time-sensitive characterization instruments can safely service within required windows. Because the scheduler retains every requested experiment, exploratory grid points and optional repeats continue reserving downstream capacity while control runs, already-created unstable samples, and mandatory hazard checks compete for the same constrained resources.","actors":["Scientists submitting synthesis campaigns","Laboratory automation operator","Shared-instrument custodians","Facility scientific director","Environment, health, and safety reviewer","Robotic synthesis and characterization scheduler"],"observable_state":"The overload state is observable when the scheduler's forecast shows admitted but unstarted recipes plus active batches requiring transfer or characterization exceeding a locally validated capacity red-line for one or more protected resources. Supporting signals include growing reservation age, predicted misses of sample-stability windows, unavailable monitoring slots, and an increasing fraction of capacity committed to exploratory or optional work.","consequence":"Continuing to retain all work can make critical controls or hazard checks late, allow time-sensitive samples to become uninterpretable, occupy safe holding positions, and eliminate the recovery headroom needed to complete already-started experiments with intact provenance.","affected_objective":"Preserve safe completion and interpretability of active batches, required controls, and time-sensitive characterization while keeping the laboratory's admitted workload within validated handling capacity.","intervention":"Before reagents are dispensed, classify recipes using pre-registered, auditable tiers. Protect active-batch follow-ups, mandatory controls and calibrations, hazard-verification work, and samples with irreversible characterization windows. When a resource forecast crosses its overload red-line, apply selective rejection to unstarted, lower-tier work: cancel excess replicates beyond the precommitted minimum and reject exploratory grid points that have not begun. Return each rejected recipe with a reason code and permit later resubmission instead of placing it in an invisible retry queue. Reserve capacity for protected work, distribute discretionary acceptance across campaigns under a declared fairness rule, and resume ordinary admission only after the forecast remains below a lower recovery threshold for a specified number of planning cycles.","structural_mapping":[{"archetype_element":"Capacity-constrained system under overload","domain_realization":"The coupled synthesis, contained-transfer, and characterization workflow has finite concurrent slots and time-windowed throughput."},{"archetype_element":"Distinguishable load classes","domain_realization":"Unstarted recipes can be separated into protected controls, safety and active-batch obligations, committed minimum replicates, optional repeats, and exploratory composition-grid points."},{"archetype_element":"Detectable overload threshold","domain_realization":"A rolling resource forecast compares admitted obligations with validated reactor, transfer, monitoring, and characterization capacity red-lines."},{"archetype_element":"Selective sacrifice of lower-priority load","domain_realization":"Eligible unstarted optional recipes are explicitly rejected or expired before material preparation rather than retained in the schedule."},{"archetype_element":"Protected critical capacity","domain_realization":"Slots are reserved for safe completion of active reactions, mandatory controls, hazard verification, and unstable samples with irreversible deadlines."},{"archetype_element":"Explicit handling of shed load","domain_realization":"Every rejected recipe receives a recorded reason, remains visibly unexecuted, and requires deliberate resubmission after recovery."},{"archetype_element":"Recovery with hysteresis","domain_realization":"Ordinary admission resumes only below a lower recovery threshold maintained across multiple planning cycles, limiting oscillatory rejection and readmission."},{"archetype_element":"Observability and auditability","domain_realization":"The facility records trigger state, protected reservations, rejected recipes, campaign-level burden, classifier overrides, and recovery decisions."}],"mechanism_mapping":[{"mechanism_slug":"admission_control","role":"Places the shedding boundary before reagent dispensing so rejected recipes do not consume material, reactor, transfer, or characterization commitments.","counterfactual_removal":"Without admission control, low-priority recipes would enter execution and their downstream obligations could no longer be cleanly removed."},{"mechanism_slug":"priority_based_dropping","role":"Removes only declared eligible classes while preserving controls, safety checks, active-batch obligations, and time-sensitive samples.","counterfactual_removal":"Without priority-based dropping, rejection would be arbitrary or could sacrifice work needed for safety and scientific interpretation."},{"mechanism_slug":"controlled_drop","role":"Produces an explicit terminal status and reason code for each shed recipe rather than silently deleting or indefinitely queuing it.","counterfactual_removal":"Without controlled drop handling, requests could disappear, accumulate in a hidden backlog, or be automatically retried into continuing overload."},{"mechanism_slug":"threshold","role":"Activates shedding from forecasted resource stress and uses a separate recovery boundary to prevent rapid on-off cycling.","counterfactual_removal":"Without thresholded activation and recovery, operators could normalize discretionary cancellation or repeatedly shed and readmit the same workload."}],"causal_chain":["Admitted synthesis obligations exceed the coupled safe capacity of reactors, transfers, monitoring, and time-sensitive characterization.","Retaining every recipe commits scarce downstream slots to optional work and removes headroom for already-started or protected work.","A rolling forecast makes impending overload visible before eligible recipes consume reagents or equipment.","Pre-registered classifications distinguish protected obligations from optional, still-unstarted recipes that can be sacrificed cleanly.","Selective rejection removes those optional obligations and prevents them from becoming material-bearing work or an accumulating retry queue.","Freed reservations remain available for active batches, mandatory controls, hazard checks, and unstable samples with expiring windows.","Explicit records, fairness checks, and hysteretic recovery bound the sacrifice and allow normal admission to return after stress subsides."],"baseline":"The laboratory uses a FIFO or manually reprioritized scheduler that retains all submitted recipes. Operators may move important experiments forward, but optional work remains admitted and continues occupying forecast capacity; overload is therefore reordered or delayed rather than explicitly removed.","nearest_rivals":["Priority scheduling, which changes execution order but leaves total admitted obligations intact.","Rate limiting of new campaign submissions, which slows intake without removing lower-priority work already admitted during an overload episode.","Buffering prepared samples in cold or inert storage, which shifts pressure into finite storage and does not eliminate aging or later characterization demand.","Graceful degradation of measurement protocols, which reduces characterization detail rather than reducing the number of experiments served.","Capacity expansion through additional instruments or shifts, which increases supply rather than converting current overload into bounded sacrifice."],"remaining_contrastive_claim":"The defining move is the explicit, auditable rejection or expiration of selected unstarted recipes after forecasted overload, thereby releasing coupled downstream reservations for protected experimental obligations; merely reordering, slowing, storing, or measuring every recipe less completely would not instantiate this intervention.","authority_safety":{"decision_authority":"The facility scientific director approves the tier definitions and resource thresholds with the automation operator, instrument custodians, and environment, health, and safety reviewer. The scheduler may execute only the approved policy, while designated operators retain a logged halt and override authority.","authorized_first_step":"Run the policy in offline replay and then shadow mode, recording hypothetical decisions without rejecting, modifying, starting, or disposing of any experiment.","excluded_actions":["Aborting or deprioritizing an active reaction","Shedding mandatory controls, calibrations, hazard checks, or regulatory work","Allowing rejected recipes to retry automatically","Changing recipe priority after overload begins without a logged authorized override","Discarding samples or reagents","Waiving containment, monitoring, provenance, or characterization safety requirements","Using submitter seniority or informal influence as a priority rule"],"halt_rollback":"Stop the evaluation if a protected recipe is marked eligible for shedding, a rejected item loses traceability, the policy predicts an unsafe active-batch state, or campaign-level burden violates the declared fairness rule. Disable the policy version, preserve the audit log, and return to the existing scheduler; because the first step is non-operative, rollback requires no restoration of experiments."},"negative_tests":{"strongest_counterevidence":"Rejecting pending recipes would not relieve the actual bottleneck if overload is dominated by already-active reactions, fixed maintenance downtime, or characterization demand that cannot be identified before synthesis. The approach is also contraindicated if the supposedly optional recipes are necessary to interpret the protected experiments or if no experiment class can be cleanly sacrificed.","problem_falsifier":"Time-stamped records from a bounded campaign show that admitted obligations never exceeded validated handling capacity, protected work did not face forecastable deadline or monitoring conflicts, or the observed failures arose from instrument correctness faults rather than excess load.","intervention_falsifier":"Against the same recorded workload, the shedding policy fails to restore a feasible protected schedule, marks protected or scientifically indispensable work for rejection, merely moves recipes into a growing resubmission backlog, or creates materially concentrated sacrifice under its declared fairness rule.","risks":["Misclassification could reject an experiment essential to interpreting a batch.","Researchers could inflate priority labels unless tiers are pre-registered and overrides audited.","Repeated shedding could starve exploratory campaigns or concentrate loss on less influential groups.","Rejected recipes could reappear as an untracked personal or administrative backlog.","Forecast error could trigger shedding too early or too late.","Hysteresis could retain emergency restrictions longer than justified.","Temporary shedding could normalize underinvestment in laboratory capacity.","Removing selected grid points could bias the composition space ultimately observed."]},"next_evidence_step":"Using one completed materials campaign with time-stamped submissions, resource reservations, starts, transfers, and characterization deadlines, conduct a non-operative replay over the campaign's recorded horizon. Before replay, have the authorized reviewers label protected and shed-eligible recipe classes without seeing the policy output. Compare the existing scheduler with the proposed threshold, reserve, rejection, fairness, and recovery rules on four checks: feasibility of protected obligations, number and identity of wrong-shed classifications, boundedness of rejected or resubmitted work, and distribution of hypothetical shedding across campaigns. Do not enable live rejection unless the replay is traceable and produces no protected-work shedding under the reviewed scenarios.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"No comparison with prior proposals was performed because runtime isolation forbids inspecting them; this candidate was derived only from the supplied load-shedding archetype and chemistry-and-materials domain card.","revision_record":{"parent_version":null,"progress_targets_addressed":["Initial one-shot construction of a complete reverse-innovation candidate"],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}