{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp04_retrieval_first_paired20_20260802","cell_id":"computability_boundary_mapping__engineering_design","arm":"RETRIEVAL_FIRST","candidate_id":"C-H5-ORACLE-ROUTER","hypothesis_id":"H5","version":0,"title":"Oracle-aware routing for multidisciplinary optimization","problem":"A multidisciplinary optimizer treats vendor simulation, laboratory measurement, or expert judgment as an always-available deterministic subroutine, without declaring the external capability's input promise, availability, latency, consistency, refusal, or failure semantics.","actors":["Multidisciplinary optimization engineer","Systems engineer","External simulation owner or vendor","Laboratory evaluator","Human domain expert","Program design-review board","Downstream consumer of the design-gate verdict"],"observable_state":"For one candidate-design evaluation, the optimizer requests evidence from an external simulator, laboratory, vendor, or human evaluator; the request may violate a declared precondition or return success, delay, refusal, inconsistent judgments, error, or no result. The workflow then emits either a qualified result, a labeled unresolved or escalated state, or an unqualified feasible, infeasible, or optimal verdict.","consequence":"When missing or invalid external evidence is translated into an objective penalty, constraint violation, or ordinary value, the optimizer can launder unavailable information into a definitive design-gate verdict. Systems engineers and review boards then receive a conclusion whose apparent guarantee exceeds the computation model and evidence actually available.","affected_objective":"Preserve the validity, traceability, and scope of multidisciplinary feasibility and optimality verdicts at engineering design gates without claiming that unavailable external capabilities have supplied evidence.","intervention":"Introduce a mechanically enforced oracle contract for each external evaluator. The contract declares its evaluator class, input promise, success evidence, latency and availability limits, and typed delay, refusal, inconsistency, error, and no-result states. Before every call, a promise checker either admits the request or labels it OUT_OF_PROMISE. After the call, a fallback-mode router may accept valid evidence, perform only preauthorized retry or controlled recovery, or emit UNRESOLVED and route the case to the design-review board. A versioned computability-boundary decision record preserves the route and guarantee through the final gate and prohibits an unqualified feasible, infeasible, or optimal verdict while required evidence remains unresolved.","structural_mapping":[{"archetype_element":"Problem-Class Specification","domain_realization":"The class is candidate-design evaluations whose gate verdict depends on at least one external simulator, laboratory, vendor, or human evaluator."},{"archetype_element":"Instance Representation Contract","domain_realization":"Each evaluation request records the candidate identifier, evaluator class, required inputs, declared promise, expected evidence type, and applicable deadline."},{"archetype_element":"Computation Model Contract","domain_realization":"The optimizer is classified as relative to named external capabilities rather than as a self-contained deterministic procedure."},{"archetype_element":"Solvability Guarantee Profile","domain_realization":"A definitive gate verdict is permitted only when every required external evaluation returns valid evidence within its declared contract; otherwise the guarantee is limited to a labeled unresolved or escalated result."},{"archetype_element":"Scope Boundary","domain_realization":"Requests failing a mechanically checked evaluator promise are OUT_OF_PROMISE and cannot inherit the in-promise optimization guarantee."},{"archetype_element":"Unknown and Nontermination Policy","domain_realization":"Delay, refusal, inconsistency, error, and absence of evidence remain distinct typed non-results and are never collapsed into infeasible, feasible, or optimal."},{"archetype_element":"Fallback Solution Contract","domain_realization":"The router selects only contract-authorized retry, controlled recovery, or board escalation and labels the resulting guarantee."},{"archetype_element":"Decision Record and Recheck Trigger","domain_realization":"A versioned record links each gate verdict to evaluator contracts, routes, evidence, unresolved residue, and triggers reclassification when an evaluator, promise, failure policy, or required evidence changes."}],"mechanism_mapping":[{"mechanism_slug":"promise_problem_restriction","role":"Defines the inputs on which each external evaluator is accountable and requires mechanical precondition checking before the optimizer relies on its output.","counterfactual_removal":"Without this mechanism, the workflow cannot distinguish an evaluator failure from a request the evaluator never promised to handle, so out-of-domain behavior may appear authoritative."},{"mechanism_slug":"fallback_mode_router","role":"Routes successful, delayed, refused, inconsistent, failed, and out-of-promise evaluations to the strongest permitted mode while preserving qualified labels through the gate.","counterfactual_removal":"Without the router, typed non-results can still be converted ad hoc into dummy objectives, constraint penalties, or ordinary values, recreating the false definitive verdict."},{"mechanism_slug":"computability_boundary_decision_record","role":"Records the external-capability assumptions, shipped guarantee, routing decision, evidence status, and triggers that require the boundary to be reconsidered.","counterfactual_removal":"Without the record, the live router may handle an individual failure correctly while design documents or later gate decisions silently restore an unrestricted guarantee."}],"causal_chain":["An optimization evaluation is specified as relative to a named external evaluator and a mechanically checkable input promise.","The promise checker rejects or quarantines requests outside that promise before their outputs can influence the optimization state.","For admitted requests, the evaluator adapter converts success, delay, refusal, inconsistency, error, and absence into typed states rather than fabricated objective or constraint values.","The fallback-mode router accepts valid evidence or chooses only an authorized retry, controlled recovery, or escalation route.","Any unresolved required evaluation propagates as UNRESOLVED or ESCALATED through candidate aggregation and the design-gate interface.","The boundary decision record binds the final label to the applicable contracts, evidence, assumptions, and recheck triggers.","Therefore missing external information cannot, under the contract, become an unqualified feasible, infeasible, or optimal verdict."],"baseline":"Existing practice already supplies substantial constituent prior art: Dakota declares abort, retry, recovery, and continuation for simulation failures and can substitute dummy function values; FMI provides typed statuses and undefined-output semantics; OpenMDAO and MathWorks propagate analysis errors or NaN values to solvers; hidden-constraint research models unevaluable points; human-in-the-loop research documents inconsistent judgments; and NASA guidance calls for assumptions, uncertainty, methods, and rationale in decision records. The baseline lacks evidence, in the bounded prior-art record, of one mechanically enforced contract carrying a typed unresolved state across all four evaluator classes through the final multidisciplinary design gate.","nearest_rivals":["Dakota Simulation Failure Capturing","Functional Mock-up Interface typed status contract","OpenMDAO AnalysisError handling","MathWorks NaN handling for failed simulation evaluations","Surrogate optimization for hidden constraints","Human-in-the-loop Bayesian optimization failure analysis","NASA Systems Engineering Handbook decision-analysis record"],"remaining_contrastive_claim":"No examined source implements one mechanically enforced contract spanning simulator, laboratory, vendor, and human evaluations that checks each input promise and preserves a labeled unresolved or escalated state through the final multidisciplinary design-gate verdict.","authority_safety":{"decision_authority":"The program design-review board retains authority over final gate disposition, evaluator-contract approval, exceptions, and any decision to proceed with unresolved evidence; the optimizer and router may only enforce labels and authorized routes.","authorized_first_step":"Run a non-decisional shadow evaluation on one existing multidisciplinary workflow using recorded or scripted candidate requests; compare the proposed promise checks and typed routes with the workflow's actual outputs without changing optimization results or gate decisions.","excluded_actions":["Automatically declare a candidate feasible, infeasible, or optimal after any required evaluation is unresolved","Treat timeout, refusal, inconsistency, error, missing output, or promise violation as an objective or constraint value","Override laboratory, vendor, expert, or design-review-board authority","Change production optimizer behavior or formal design-gate decisions during the first evidence step","Generalize a contract validated for one evaluator or promise to another without review","Represent board escalation as if it were an oracle answer","Claim worldwide novelty, patentability, or absence of additional prior art"],"halt_rollback":"Halt the shadow exercise if it affects a live gate, exposes restricted evaluator information, cannot distinguish shadow from production records, or produces an unqualified definitive verdict from a typed non-result. Quarantine the affected trace, disable the shadow instrumentation, preserve the original workflow, and return the contract and route table to board review."},"negative_tests":{"strongest_counterevidence":"The combined prior art is close: Dakota already performs explicit external-simulation failure routing; FMI standardizes typed failed-call states and undefined outputs; OpenMDAO and MathWorks propagate evaluation failures; hidden-constraint methods address missing outputs; NASA requires decision provenance. A deployed extension or workflow joining these established elements across simulator, laboratory, vendor, and human evaluators could eliminate the remaining integration claim.","problem_falsifier":"Across a bounded representative trace set, if external delay, refusal, inconsistency, failure, or promise violation is already always preserved as a typed non-result through the final design-gate verdict, then the nominated laundering problem is not present in the observed workflow.","intervention_falsifier":"The intervention is falsified if any shadow case with a required oracle failure or promise violation yields an unqualified feasible, infeasible, or optimal verdict, or if valid in-promise evidence is incorrectly blocked often enough that the declared contract cannot support the workflow's required candidate evaluations.","risks":["A mechanically checked promise may encode the wrong engineering boundary and reject valid evaluations.","A typed status adapter may misclassify a valid result or conceal distinctions among delay, refusal, inconsistency, and failure.","Excessive escalation may make the workflow formally honest but operationally unusable.","A human or vendor evaluator may not expose enough information to check its promise mechanically.","A board may override the unresolved label without recording that the original optimization guarantee no longer applies.","The router may merely relabel dummy-value recovery while downstream systems continue reading it as definitive.","The decision record may become stale when evaluators, interfaces, evidence requirements, or gate rules change."]},"next_evidence_step":"Within a fixed shadow set of at most 20 candidate evaluations, include at least one recorded or scripted case for each evaluator class—simulator, laboratory, vendor, and human—and exercise success plus available instances of promise violation, delay, refusal, inconsistency, error, or no result. Record the promise-check outcome, typed evaluator state, selected route, downstream optimization label, and shadow gate label. Pass only if every required non-result remains OUT_OF_PROMISE, UNRESOLVED, or ESCALATED and no case produces an unqualified feasible, infeasible, or optimal verdict; report coverage gaps rather than treating unobserved failure modes as passed.","prior_art_status":"SEARCHED_BOUNDED","revision_record":{"parent_version":null,"progress_targets_addressed":["Converted selected hypothesis H5 into a complete operational candidate","Preserved the narrow residual distinction after substantial prior-art collision","Specified actors, state, consequences, authority, exclusions, falsifiers, and a bounded evidence step"],"conceptual_changes":["Defined optimization solvability as relative to explicit heterogeneous external capabilities","Limited the residual contribution to end-to-end contract integration across four evaluator classes and the final gate"],"operational_changes":["Added mechanically checked evaluator promises","Added typed non-result routing and propagation to the design-gate label","Added a versioned boundary record and recheck triggers","Restricted the first test to non-decisional shadow operation"],"evidence_changes":["Incorporated the seven-source independent prior-art record as substantial counterevidence","Named the closest product, standard, research, and guidance rivals","Made unobserved shadow failure modes coverage gaps rather than evidence of success"],"claim_changes":["Retained the independent prior-art record's exact remaining contrastive claim","Excluded worldwide novelty and patentability claims","Made any unqualified definitive verdict following oracle failure or promise violation an explicit intervention falsifier"]}}