DispatchAtlasAPI Reference
The DispatchAtlas public API by distribution package — core contracts, benchmarks, solvers, the campaign engine, and the analytica statistics and export surface.
The public API is grouped by distribution. Import from the narrowest namespace
that owns the behavior. The lists below mirror the current public namespace
exports so readers can audit the supported import surface without reading the
package __init__.py files.
📦 Root Distribution
Use dispatchatlas.version for the lightweight root package version surface:
__version__
🧱 Core
dispatchatlas.core is the inner package boundary. It exports domain objects,
validation, provenance, serialization, seed, filesystem, logging, and protocol
contracts.
Configuration contracts:
SeedPolicyValidationMode
Error contracts:
DataQualityErrorDispatchAtlasErrorDomainValidationErrorInfeasibleScheduleErrorMissingOptionalDependencyErrorProvenanceErrorScheduleValidationErrorSeedDerivationErrorSerializationContractErrorUnsupportedCapabilityErrorrequire_finite— return the value coerced to a finite float or raise the given error contract when it is NaN or infinite.require_non_negative_int— return the integer or raise the given error contract when it is negative.require_positive_int— return the integer or raise the given error contract when it is not positive.require_text— return the stripped text or raise the given error contract when it is blank.
Filesystem helpers:
read_text_resilientsafe_child_directorysafe_directory_namesafe_namespaced_idwrite_json_artifactwrite_json_artifact_atomicwrite_text_artifactwrite_text_artifact_atomic
Observability (centralized logging):
get_loggermake_formatterconsole_handlerfile_handlerscoped_loggingconfigure_root_loggingMESSAGE_FORMATCONSOLE_FORMATSTRUCTURED_FORMATDEFAULT_DATEFMTROOT_LOGGER_NAME
Campaign manifest contracts:
CampaignCheckpointManifestCampaignFailureRecordCampaignRunRecordcampaign_checkpoint_from_jsonablecampaign_checkpoint_to_jsonablecampaign_failure_record_from_jsonablecampaign_failure_record_to_jsonablecampaign_run_record_from_jsonablecampaign_run_record_to_jsonable
Scheduling model contracts:
AssignmentConstraintConstraintKindDependencyDependencyKindDurationMetadataObjectiveObjectiveSenseObjectiveValueParityBasisProblemIdProblemSpecResourceAmountResourceIdResourceRequirementResourceSpecScheduleScheduleResultScheduleStatusTaskIdTaskModeTaskSpecTimePointmetadata_items
Integration protocols:
AnalysisExporterBenchmarkDescriptorBenchmarkProviderDisclosureLabelDisclosurePolicyExperimentPlanExperimentResultExperimentRunnerResultRepositorySolverSolverDescriptorSolverRunSupportsDiagnostics— optional capability a solver advertises so a campaign can enable per-iteration population diagnostics without the runner reaching into solver internals.
Provenance contracts:
ArtifactHashCitationStatusEnvironmentStampEvidenceClassPackageRecordProvenanceQualityGateStatusSeedLineageSourceReferencehash_bytesutc_now
Seed contracts:
MAX_SEEDSEED_DOMAIN_SEPARATORSeedSeedStreamderive_seed
Serialization contracts:
ArtifactEnvelopecanonical_json_bytesdumps_envelopeenvelope_artifactenvelope_from_jsonableenvelope_to_jsonableenvironment_from_jsonableenvironment_to_jsonableloads_envelopeproblem_from_jsonableproblem_to_jsonableprovenance_from_jsonableprovenance_to_jsonableschedule_from_jsonableschedule_to_jsonable
Validation contracts:
ValidatedProblemValidationIssueValidationReportValidationSeverityValidationStampcollect_problem_issuescollect_schedule_issuesvalidate_problemvalidate_schedule
Core sizing constants:
DEFAULT_CPU_RESERVE— CPUs held back from the worker pool by default.DEFAULT_MEMORY_RESERVE_FRACTION— the fraction of system memory kept in reserve by default.OBJECTIVE_FAMILY— the registry mapping each objective kind to its canonical definition.
Host concurrency:
available_cpu_count/available_system_memory_bytes— the CPUs available to this process and the best-effort available system memory.maximal_safe_worker_count— the maximal safe parallel worker count for the host.
Continuum model (edge–fog–cloud):
ComputeTier— a continuum compute layer (edge, fog, cloud) with its aggregate capacity.NetworkTopology/NetworkLink— the communication fabric and its directed inter-tier links.Bandwidth/DataVolume— positive link throughput in bytes per second and non-negative transferable data in bytes.PlacementAffinity— a task's preference for a placement target on the continuum.
Cost model (R||Cmax, communication, load-dependent execution, sequence-dependent setup):
CostModel/cost_model_from_jsonable/cost_model_to_jsonable— the cost-model bundle linked to a problem by identifier, and its serialization.ExecutionTimeMatrix/ExecutionTimeEntry/execution_matrix_from_iterable— the unrelated-machine execution-time matrix (p_ij).CompatibilityMask— the explicit task-to-machine eligibility set.CommunicationModel/CommunicationLink— cross-resource communication penalties on a sparse graph.LoadModel/LoadModelKind/LoadBreakpoint— the load-dependent execution curve and its interpolation between breakpoints.SetupMatrix/SetupEntry/SetupReferenceKind— sequence-dependent setup times keyed by transition and machine.
Objective family:
ObjectiveKind/ObjectiveDefinition/objective_definition— named scheduling objectives and their canonical metadata.ObjectiveReduction/VectorObjectiveValue— the scalarization rule and a measured multi-objective vector with disclosure metadata.vector_objective_from_jsonable/vector_objective_to_jsonable— vector-objective serialization.
Objective evaluation:
evaluate_objective/evaluate_objectives/ObjectiveEvaluation— evaluate one or each named objective over a schedule.MultiObjectiveOutcome/computed_vector/multi_objective_outcome_from_jsonable— scalar and vector objectives plus feasibility for one schedule.EvaluationStatus/objective_sense— whether a value was computed or deferred, and an objective's optimization sense.
Feasibility and constraints:
FeasibilityReport/build_feasibility_report— hard-violation rows plus soft-constraint penalty accounting.InfeasibleRow/SoftConstraintPenalty— one reviewer-visible hard violation and one weighted soft-constraint penalty.ServiceLevelConstraint/ServiceLevelOutcome/evaluate_service_level/evaluate_service_level_over_schedule/resolve_metric— named service-level constraints on a schedule metric and their evaluation.ConstraintViolationSummary/summarize_constraints— aggregated hard, soft, and service-level constraint accounting.
Pareto front and quality indicators:
dominates/to_minimization— Pareto dominance under minimization and mixed-sense conversion.fast_non_dominated_sort/non_dominated_front/non_dominated_indices— NSGA-II non-dominated sorting and the non-dominated set.FrontierPoint/LabelledPoint/frontier_data— labelled objective vectors and frontier-ready records.hypervolume/igd_plus/additive_epsilon_indicator/spread— the four quality indicators: hypervolume, IGD+, additive epsilon, and nearest-neighbor spread.QualityIndicatorReport/quality_indicator_report— all four indicators for one front against a reference set.
Robustness:
PerturbationScenario/PerturbationSet— named perturbation scenarios for one objective.DEFAULT_CVAR_ALPHA/RobustnessAggregation/RobustnessValue/evaluate_robustness— how a perturbation set aggregates into a robustness number with its stated basis, and the default CVaR tail probability.
Run manifests:
RunTiming/ResourceUsage/SolverCounters— wall-clock and anytime timing, peak resource use, and solver effort counters for one run.ConvergenceSample/TimedObjectiveSample— objective values observed at an iteration count and at a wall-clock offset.
Termination:
TerminationPolicy— a composable, multi-criterion solver stopping policy.
Filesystem safety:
safe_path_under_root— a path returned only when it resolves below an authorized root.
Continuum serialization:
bandwidth_from_jsonable/bandwidth_to_jsonablecompute_tier_from_jsonable/compute_tier_to_jsonabledata_volume_from_jsonable/data_volume_to_jsonablenetwork_link_from_jsonable/network_link_to_jsonablenetwork_topology_from_jsonable/network_topology_to_jsonableplacement_affinity_from_jsonable/placement_affinity_to_jsonable
🧪 Benchmarks
dispatchatlas.bench exports benchmark catalogs, generic and continuum family
generators, generator metadata, materialization, citation validation,
characterization, hardness and distribution-distance metrics, selection, download, and
taxonomy contracts.
Catalog contracts:
StaticBenchmarkProviderbenchmark_provider_for_selectorsbuild_feasibility_diagnostic_catalogbuild_full_catalogbuild_smoke_catalogfull_benchmark_providerfull_catalog_configssmoke_benchmark_providersmoke_catalog_configs
Characterization contracts:
CharacterizationMetricsCharacterizationReportcharacterize_problem
Citation contracts:
CitationClaimCitationMatrixSourceKinddefault_citation_matrixvalidate_citation_matrix
Generator contracts:
BenchmarkGeneratorDagWorkflowGeneratorGeneratorRegistryIndependentTaskGeneratorcloud_edge_profiledefault_registryworkflow_profile
Extensibility:
BENCHMARK_GENERATOR_ENTRY_POINT_GROUPdiscover_plugin_generators— opt-in registration of third-party benchmark generators advertised under a packaging entry-point group, preserving the citation-backed invariant.
Materialization contracts:
BenchmarkInstanceBenchmarkInstanceSetCatalogManifestCatalogManifestEntryconfig_to_jsonablemanifest_to_jsonablematerialize_benchmarkmaterialize_with_generatorwrite_instance_set
Metadata contracts:
BenchmarkConfigDomainProfileGeneratorMetadataScaleSpec
Generic family contracts:
CostModelGeneratorGenericFamilyGeneratorflexible_job_shop_generatorflow_shop_generatorgeneric_family_full_configsgeneric_family_generatorsgeneric_family_smoke_configsjob_shop_generatormachine_scheduling_generatorrcpsp_generatorsetup_flow_shop_generator
Continuum family contracts:
ContinuumFamilyGeneratorContinuumFamilySpecContinuumScenarioContinuumStratumDistinctivenessRecordbuild_continuum_catalogbuild_continuum_full_catalogcontinuum_family_distinctivenesscontinuum_family_full_configscontinuum_family_generatorscontinuum_family_smoke_configscontinuum_family_specs
Hardness contracts:
HardnessAxisHardnessSpechardness_axishardness_axis_registrymeasured_communication_compute_ratiomeasured_deadline_slack_ratiomeasured_dependency_densitymeasured_duration_cvmeasured_fabric_churnmeasured_offload_deceptionmeasured_release_stagger
Distribution-distance contracts:
MAX_DISTRIBUTION_DISTANCEDistributionDistanceBridgeDistributionDistanceScoreDistributionDistanceStatusdefault_distribution_distance_bridgesmetrics_from_instance_setper_feature_wassersteindistribution_distance_scorevalidate_distribution_distance_bridge
Selection contracts:
DifficultyStratumSubsetCriteriadifficulty_scoredifficulty_stratuminstance_familyinstance_profile_classselect_benchmark_subsetstratify_instances
Catalog-metadata and download contracts:
CATALOG_SCALAR_COLUMNSCatalogInstanceMetadataDownloadFormatcatalog_instance_metadatacatalog_instance_to_jsonablecatalog_instance_to_rowcatalog_metadata_for_instancescatalog_metadata_for_setscatalog_subset_from_bundlecatalog_subset_from_csvcatalog_subset_from_jsoncatalog_subset_to_bundlecatalog_subset_to_csvcatalog_subset_to_jsonselect_catalog_metadataserialize_catalog_subset
Taxonomy contracts:
DEFAULT_TAXONOMYBenchmarkTaxonomyBenchmarkTierConstraintFeatureDynamismFeatureEnvironmentDomainInfrastructureRealismObjectiveFeatureProfileClassSchedulingFamilySchedulingStructureUncertaintyFeature
Reference-suite registry contracts:
InstanceCountStatusReferenceSuiteSuiteFormatdefault_reference_suitesreference_suitevalidate_reference_suites
Best-known-solution contracts:
BestKnownIngestBestKnownKindBestKnownRegistryBestKnownValueload_best_knownload_best_known_registry
Canonical-instance ingestion contracts:
SUPPORTED_FORMATSParseContextParsedInstanceReferenceSuiteGeneratorload_reference_suiteparse_fjsparse_standard_jspparse_taillard_pfspparse_wfformat_jsonreference_suite_configreference_suite_distribution_distance_bridge
🧮 Solvers
dispatchatlas.solve exports solver metadata, registries, constructive
baselines, metaheuristics, NDSO variants, operators, optional adapters, and
profiling helpers.
Baseline and execution contracts:
DispatchRuleDispatchingSolverEvaluationMeter— the per-solve objective-evaluation tally and optional budget. Every candidate the search scores passes through it, including the evaluationslocal_search_ordermakes inside the operators, so equal-evaluation parity binds to measured consumption rather than an iteration count multiplied by an assumed per-iteration cost. A budget ofNonemeters without bounding; exhaustion is a predicate polled at the loop head, never an interrupt.OrderRunRequestTerminationTracker— the stateful, first-to-fire evaluator that folds a search loop's progress into aTerminationPolicyand reports which criterion stopped the run.baseline_solversdeadline_exceeded— whether a wall-clock deadline has passed for a run.neh_order— the NEH insertion-construction order (Nawaz–Enscore–Ham).
Solver metadata contracts:
DependencyRequirementSolverCapabilitySolverDescriptorSpecSolverFamilySolverMetadataStochasticity
Extensibility:
SOLVER_ENTRY_POINT_GROUPdiscover_plugin_solvers— opt-in registration of third-party solvers advertised under a packaging entry-point group, preserving the citation-backed invariant.
Metaheuristic contracts:
MetaheuristicKindPermutationMetaheuristicSolverPermutationSearchConfigmetaheuristic_solvers
Rank-based list-scheduling contracts:
ListSchedulingRankListSchedulingSolverlist_scheduling_solversheft_profile— upward-rank list scheduling (HEFT).cpop_profile— combined upward+downward-rank list scheduling (CPOP).peft_profile— optimistic-cost-table list scheduling (PEFT).
Ready-set mapping-rule contracts:
MappingRuleMappingRuleSolvermapping_rule_profilemapping_rule_solvers— the min-min, max-min, and sufferage batch rules.
Pareto multi-objective competitor contracts:
NSGA2ConfigNSGA2SolverMOEADConfigMOEADSolvernsga2_profilemoead_profilemulti_objective_solverscrowding_distances— the NSGA-II crowding-distance density estimator.nsga2_environmental_selection— rank-then-crowding elitist survival.tchebycheff— the MOEA/D Tchebycheff scalarizing function.
Iterated-greedy contracts:
ITERATED_GREEDY_RS_CITATIONIteratedGreedyRSConfigIteratedGreedyRSSolver— the canonical destruction–construction configuration for permutation flow shops.iterated_greedy_rs_profileiterated_greedy_rs_solvers
Critical-path tabu contracts:
CriticalPathTabuConfigCriticalPathTabuSolvercritical_path_tabu_profilecritical_path_tabu_solverscritical_pairs— the critical-path adjacent-pair neighborhood generator.
Mode-assignment tabu contracts:
TabuMrcpspModeSearchConfigTabuMrcpspModeSearchSolvertabu_mrcpsp_mode_search_profiletabu_mrcpsp_mode_search_solversmoldable_activities— the multi-mode activities the mode-assignment search ranges over.
Project-scheduling (serial SGS) contracts:
SerialSGSJustificationSolver— serial schedule generation with double justification for resource-constrained project scheduling.serial_sgs_justification_profileserial_sgs_justification_solversdouble_justified_order— the backward–forward justification pass.latest_finish_order— the latest-finish-time priority order.
NDSO contracts:
NDSOAblationNDSOMechanismsNDSOSolverNDSOVariantndso_ablation_catalogndso_solvers
Scheduling operator contracts:
BatchScoreResultScheduleMetricsadjacent_swap_neighborbatch_score_ordersconstruct_scheduleearliest_finish_task_orderlocal_search_ordernormalize_task_orderobjective_values_forrandom_topological_task_orderrepair_scheduleschedule_metricsscore_orderscore_scheduletopological_task_order
Optional adapter contracts:
OptionalExactSolveroptional_solversortools_dependency
Performance contracts:
BatchScoringProfileProfiledResultprofile_call
Registry contracts:
ProfiledSolverSolverRegistrydefault_solver_registry
Solver tuning constants:
NATIVE_EXACT_TASK_CAP— the task-count ceiling above which the native exact solver fails closed.RUN_COUNT_FLOOR— the minimum repeated-run count a fair comparison requires.VELOCITY_WINDOW— the improvement-velocity window the anytime controller watches.IMPROVEMENT_FLOOR— the minimum relative improvement that counts as progress.FEATURE_ORIGIN— the benchmark-platform metric each selection feature derives from.
Baseline dispatching families:
DispatchingFamilySpec— a named dispatching family bound to a deterministic priority rule.RetiredFamily/RetirementReason/retired_dispatching_families— named families deliberately set aside, each with its rationale.
Diversified competitors (the named metaheuristic field):
CompetitorKind/CompetitorSolver/CompetitorConfig— named competitors with shared bounded search parameters.competitor_solvers/diversified_competitors— the diversified, disclosure-aware competitor set.post_2020_currency_anchor— the post-2020 currency-anchor references for the field.recent_peer_strength_bar— the recent peers retained under the venue-strength bar.
Continuous optimizers (decoded onto the discrete order):
clpso_optimize/CLPSOConfig— comprehensive-learning particle swarm optimization.lshade_optimize/LSHADEConfig— success-history adaptive differential evolution with linear population reduction.EncodedSolverKind/EncodedAdapterSolver/encoded_adapter_solvers— the continuous family bound to a discrete encoding adapter.ContinuousResult/SearchBudget— a continuous run's outcome and its iteration, randomness, and wall-clock budget.
Continuous-to-discrete encoding:
EncodingAdapter/encoding_adapter/encoding_adapter_catalog— named continuous-to-discrete encodings with explicit repair.DecodeRule/DecodeOutcome— the rule that turns a transformed key vector into a task order, and its repaired result.RepairPolicy— the policy restoring precedence feasibility after decoding.TransferFunction— the squashing function mapping a continuous component to a selection signal.
Exact solvers:
ExactMethod/ExactSolver— named exact methods and an adapter that fails closed on a missing backend or oversized input.ExactAdapterSpec/exact_adapter_specs/exact_solvers— the bundled exact-solver adapter contracts.
Many-objective competitor (NSGA-III):
NSGA3Solver/NSGA3Config/nsga3_solvers/nsga3_profile— the NSGA-III competitor over precedence-safe orders.nsga3_reference_points— Das & Dennis structured reference points on the unit simplex.nsga3_environmental_selection— NSGA-III reference-point niching selection.
Scholarly scheduling variants:
SchedulingVariantKind/PermutationSchedulingVariantSolver/SchedulingVariantConfig— named scholarly hybrid variants over the discrete order encoding.scheduling_variant_solvers— the first-class scheduling-variant baselines.
Algorithm selection:
RuleBasedSelector— a benchmark-only baseline that ranks candidates from metadata alone.SupervisedSelector— a deterministic distance-weighted nearest-neighbor learned selector.SelectionFeature/SelectionFeatures/feature_origin— the named characterization features the selector consumes and their origin.SelectionLabel/SolverRecommendation/RecommendationSource— labeled training examples and ranked recommendations with confidence and limitations.
Selector cross-validation (leakage-free generalization):
cross_validate/GeneralizationReport— leave-one-family-out held-out generalization over a corpus.leave_one_family_out/partition_by_families/TrainTestSplit— family-disjoint train/test splits.held_out_generalization— train on the split's train half and report top-1 accuracy on its test half.leakage_report/LeakageReport— the instance, family, and characterization-record overlaps a leakage-free split must keep empty.validated_supervised_selector— a selector whose confidence reflects its held-out generalization.
Learning and hybrid interfaces (optional, fail-closed):
LearningInterfaceKind/LearningInterface/learning_interface/learning_interface_catalog— named learning and hybrid interfaces with status, policy, and fallback.InterfaceStatus/LearningEvidencePolicy— the lifecycle status and evidence contract every such surface declares.ConfidenceLabel/confidence_rank— the confidence a recommendation may carry and its ordinal strength.learning_backend_available/learning_dependency/require_learning_backend— the optional learning backend (an extra, never a default) and its typed missing-dependency error.LearnedPriorityPolicySolver/learned_solvers— the runnable learning-family solver: a policy-gradient (REINFORCE) learned priority dispatching rule (Zhang et al. 2020) that trains a linear task-feature scoring policy and decodes it greedily, seed-deterministic and dependency-free.RLDispatchingSolver/rl_dispatching_solvers/rl_dispatching_profile— the value-based learning-family solver: a tabular Q-learning agent (Aydin & Öztemel 2000) that selects among dispatching rules per coarse scheduling state and decodes the greedy argmax, seed-deterministic and dependency-free.
Solver metadata and citations:
Citation/CitationStatus— a canonical seminal reference (or an explicit not-applicable rationale) and whether a family carries one. Thisdispatchatlas.solve.CitationStatus(CANONICAL/NOT_APPLICABLE) is a distinct enum from the evidence-supportdispatchatlas.core.CitationStatus(CITATION_BACKED/EXPLORATORY/NOT_APPLICABLE) listed under Core › Provenance contracts; import from the package you mean.SolverEncoding— the solution encoding a solver searches or constructs over.SUMMIT_EVIDENCE_LABEL— the disclosure context label for metadata, evidence, and exports.
NDSO family surface:
NDSOFamilySolver— the common surface of every NDSO-family solver, single swarm or council.fast_mechanisms— the fast composition's fixed mechanism set.
NDSO mechanisms (validity-by-design construction):
PrecedenceIndex/precedence_index— predecessor sets and a stable task ordering every constructor uses.GuidanceSource/GuidanceContext/select_guidance— the three Triple-Guidance sources and a candidate's shared construction inputs.guided_construction— build a feasible candidate from a guidance order.confidence_weighted_elite— forge the Elite by Confidence-Weighted Voting.ConfidenceMatrix/consolidate_confidence/latent_archive— learned per-cell trust, its cross-swarm consolidation, and confident cells the Elite has abandoned.quality_guidance/quantity_update/BetaSchedule— the quality and quantity rules and the unified adaptive coefficient.rank_reward— a rank-derived reward in [0, 1].
NDSO inter-group council:
NDSOSummitSolver/summit_solver/summit_profile/SummitConfig— the inter-group council, the quality composition of the family.CouncilTopology— the inter-swarm exchange topology connecting the council's swarms.SummitDiagnostics— the immutable council-coordination summary attached to a run.
NDSO mechanism ablation:
AblationEntry/ablation_map— one isolating ablation per canonical mechanism.build_ablation_solver/run_ablation_suite/AblationRunResult— building and running each feasibility-bounded isolating ablation.ABLATION_STATISTICAL_PLAN/StatisticalTestPlan— the named analysis-layer statistical tests an ablation comparison uses.
NDSO anytime control:
AnytimeController/AnytimeReport— the improvement-velocity early-termination controller and its activity record.
NDSO diagnostics:
DiagnosticsCollector/DiagnosticsSnapshot— an optional, zero-overhead-when-disabled convergence recorder and its summary.IterationTrace— one iteration's convergence record.population_diversity— mean pairwise position-disagreement in [0, 1].population_positions— per-candidate task-slot vectors over the shared sorted-task axis, the population snapshot the search-space scatter projects.
NDSO disclosure (fail-closed mechanism gating):
MechanismDescriptor/mechanism_descriptor/mechanism_catalog— named mechanisms and the disclosure floor each belongs to.DisclosureFilter/DisclosureViolationError— the fail-closed gate on which mechanisms a report scope may expose.ReportScope/EvidenceProgram— staged reporting scopes and the four staged evidence programs mapped to a disclosure floor.base_disclosure_filter/program_disclosure_filter— the base-scope and per-program fail-closed filters.blocks_summit_evidence— whether a program's filter blocks the council-only quality floor.
🚀 Campaigns
dispatchatlas.lab exports reproducible campaign configuration, budget,
environment, repository, failure, and execution contracts.
Budget contracts:
CampaignBudgetReportdry_run_campaign
Configuration contracts:
FULL_CAMPAIGN_BLOCKERBudgetScalingCampaignConfigCampaignPlanCampaignRunSpecCampaignStageExecutionModeFailureCategoryOutputPolicyResourceBudgetRetryPolicycampaign_config_hashconfig_to_jsonableplan_to_jsonable
Environment contracts:
capture_campaign_environment
Error contracts:
BudgetEstimationErrorCampaignConfigErrorCampaignExecutionErrorCheckpointRecoveryErrorEnvironmentCaptureError
Repository contracts:
CampaignResultIndexCheckpointManifestFailureRecordFileResultRepositoryRunRecord
Runner contracts:
CampaignRunnerclassify_failuredefault_campaign_runnerresume_campaignrun_campaignvalidate_campaign
Campaign configuration:
CampaignKind— the experiment-design class a campaign realizes.StoppingProtocol/StoppingProtocolKind— a named, independently reportable stopping protocol for a stochastic-solver report.RunCountPolicy/STATISTICAL_POWER_FLOOR— the run-count policy and the minimum replicate count below which a solver's runs are flagged under-power.TuningBudget— the equal per-algorithm tuning budget that keeps a comparison fair.campaign_config_from_jsonable— rehydrate a validated campaign config from JSON-compatible data.
Run-count resolution:
ReplicateResolution/resolve_replicates— the resolved replicate count for a solver, with any under-power flag, under the run-count policy.
Fair-comparison invariant:
assert_fair_comparison— validate the structural preconditions of the fair-comparison invariant.assert_planned_budget_parity— check the runs the runner will actually execute, not the declaration they came from, and fail closed when a declared protocol planned no runs or when the solvers with runs under one protocol are not exactly the enrolled roster.campaign_parities— every parity the declared protocol set realizes, deduplicated across protocols.fair_comparison_manifest— the manifest rows recording the fair-comparison guarantee.protocol_parities— the budget parities a single stopping protocol realizes, read off its stop block so a protocol reports what it enforces rather than what its name suggests; a protocol bounding nothing returns an empty tuple.
Engine settings (four-layer provenance-stamped resolution):
EngineSettings/resolve_engine_settings— the resolved engine knobs across the four configuration layers, each field stamped with provenance.ResolvedSetting/SettingSource— one setting's value paired with the layer that supplied it.apply_environment_overrides— overlay the environment layer onto a file or call-site campaign config.
Worker topology and aggregation:
WorkerTopology/topology_for_plan/probe_capacity— the budget-respecting worker topology selected for a campaign after probing host capacity.AggregateManifest/merge_only_aggregation— deterministic merge-only aggregation of completed run records into one bit-stable campaign hash.
Experiment atlas:
ExperimentAtlas/default_experiment_atlas— a disclosure-aware atlas of owned evidence tracks.EvidenceTrack— one owned evidence slice in the experiment atlas.
Benchmark-family by solver applicability matrix:
ApplicabilityMatrix/default_applicability_matrix— the full benchmark-family by solver applicability matrix, derived from the live registries then annotated by the override table.applicability_matrix_markdown— render the applicability matrix as a compact Markdown status grid (solvers × families, one glyph per cell) with a legend and per-status counts.ApplicabilityCell— one(benchmark family, solver)applicability statement carrying its status, rationale, and optional evidence pointer.ApplicabilityStatus— the per-cell verified / approximate / not-applicable status enum.ApplicabilityEvidence/ApplicabilityEvidenceKind— a portal-safe evidence pointer (campaign, citation, or test) and its provenance class.ApplicabilityOverride/default_applicability_overrides— one reviewable manual upgrade of a derived cell, and the in-code override table where verified cells live.SchedulingFamilyCell— one scheduling-family by solver rollup that aggregates the constituent benchmark-family cells.derive_applicability_cell— derive one approximate-or-not-applicable cell from declared metadata.constraint_capability_bridge— the typed constraint-feature to capability-tag bridge (single source of truth).ApplicabilityScreen/screen_campaign_applicability— screen a planned campaign cross-product against the matrix, reporting blocked, approximate, and verified pairs.
📊 Analysis
dispatchatlas.analytica exports disclosure filtering, ingestion, statistical
summaries, deterministic Markdown table writers, thirty-one chart families
(twenty-eight rendered in dual format — themeable SVG and editable PGFPlots
.tex — and three SVG-only), evidence
bundles, and portal datasets.
Disclosure contracts:
DisclosureExclusionDisclosureFilterResultDisclosurePolicyEvidenceTierdefault_disclosure_policyensure_dataset_allowedfilter_dataset
Export contracts:
ExportBundleevidence_bundle_directorywrite_evidence_bundlewrite_portal_dataset
Ingestion contracts:
PlanRunFactsload_campaign_datasetload_result_dataset
Model contracts:
AnalysisConfigAnalysisDatasetAnalysisSummaryCampaignManifestCorrectionMethodExportSurfaceFailureObservationObjectiveObservationPairwiseComparisonRankEntryRunObservationSolverSummary
Statistical contracts:
bootstrap_mean_intervalcliffs_deltacohens_dpaired_cohens_dzpaired_observation_countssummarize_datasettwo_sided_sign_test
Table contracts:
ablation_table_markdownarpd_markdownbenchmark_table_markdownconstraints_table_markdownensure_effect_sizes_beside_p_valuesinfeasible_rows_markdownmultiobjective_indicators_markdownoptimality_gap_table_markdownpairwise_comparison_markdownportal_csvranking_markdownsensitivity_table_markdownsolver_summary_markdown
Visualization contracts (each chart family renders as themeable SVG; most also render as editable PGFPlots .tex):
ablation_svg·ablation_texcharacterization_svg·characterization_texconvergence_band_svg·convergence_band_texconvergence_svg·convergence_textimed_convergence_svg·timed_convergence_tex(best objective against elapsed wall-clock seconds, the axis an equal-time comparison is read on)critical_difference_svg·critical_difference_texdata_profile_svg·data_profile_texdistribution_svg·distribution_texecdf_svg·ecdf_texexploration_exploitation_svg·exploration_exploitation_texdiversity_fitness_portrait_svg·diversity_fitness_portrait_texfrontier_svg·frontier_texfrontier_3d_svg(oblique three-objective Pareto front for the many-objective solvers)gantt_svg·gantt_tex(single-schedule Gantt chart of task bars over resource lanes)radar_svg·radar_tex(multi-metric radar of each solver's normalized profile)parallel_coordinates_svg(each solver as a polyline across one parallel axis per metric)gbest_lbest_svg·gbest_lbest_texadaptability_svg·adaptability_texmultiobjective_indicators_svg·multiobjective_indicators_texperformance_profile_svg·performance_profile_texranking_svg·ranking_texreliability_svg·reliability_texrobustness_svg·robustness_texrun_order_trend_svg·run_order_trend_texruntime_quality_svg·runtime_quality_texscalability_svg·scalability_texsearch_space_animation_svg(animated SVG; SMIL flipbook)search_space_3d_scatter_svg·search_space_3d_scatter_texsearch_space_scatter_svg·search_space_scatter_texsearch_space_trajectory_svg·search_space_trajectory_texsensitivity_svg·sensitivity_texstability_svg·stability_tex
Comparison contracts:
ComplexityRowEvidenceCaveatEvidenceCaveatScopeHeadToHeadComparisonHeadToHeadRowcomplexity_table_markdownevidence_caveats_markdownhead_to_head_comparison
Inference and effect sizes (deterministic, dependency-free significance testing):
friedman_test/FriedmanResult— Friedman omnibus rank test over multiple algorithms.iman_davenport_test/ImanDavenportResult— Iman-Davenport F-correction of the (conservative) Friedman omnibus statistic.FriedmanOutcome— Friedman omnibus rank test over the full solver field.nemenyi_critical_difference/CriticalDifference— Nemenyi all-pairs critical difference for average ranks.wilcoxon_signed_rank/WilcoxonResult— two-sided Wilcoxon signed-rank test for paired differences.mann_whitney_u/MannWhitneyResult— two-sided Mann-Whitney U (Wilcoxon rank-sum) test for two independent, possibly unequal-length samples, with an enumerated exact null for small tie-free samples.bayesian_sign_test/BayesianSignTestResult— Bayesian sign test with a region of practical equivalence over paired differences.bayesian_signed_rank_test/BayesianSignedRankResult— seeded Dirichlet-process Bayesian signed-rank test with a region of practical equivalence over paired differences.vargha_delaney_a12/interpret_a12— Vargha-Delaney A12 effect size over two independent samples, and its magnitude label. The companion tomann_whitney_u; the paired summary pipeline reports the matched-pair form instead.cohens_d/interpret_cohens_d— Cohen's d_s, the independent-samples standardized mean difference over a pooled variance, and its Cohen (1988) magnitude label. The parametric companion tomann_whitney_u, not to a paired test.paired_cohens_dz— Cohen's d_z, the standardized mean of the paired differences: the parametric companion towilcoxon_signed_rank, and what the pairwise comparison table reports under "Cohen dz". Absent (None) when fewer than two pairs, or differences with no spread, leave it undefined. Not interchangeable withcohens_d: the pooled variance of d_s carries the between-instance spread that pairing removes.paired_observation_counts— the paired-observation count of every comparable solver pair: the problems each pair is matched on, and the sample size the Wilcoxon test's power rests on.bca_bootstrap_interval/BootstrapInterval— bias-corrected and accelerated bootstrap confidence interval.standard_normal_cdf/standard_normal_ppf— standard-normal cumulative distribution and quantile.chi_square_sf— chi-square upper-tail (survival) probability.f_distribution_sf— F-distribution upper-tail (survival) probability via a regularized incomplete beta.detectable_effect_size/required_paired_count/PowerAssessment— the smallest detectable standardized paired effect at a given count of paired observations, and its inverse. The count is paired observations (benchmark problems, one signed difference each), not the per-cell independent-run floor;PowerAssessment.paired_countnames the unit.ArpdRow/arpd_rows— average relative percentage deviation per solver against caller-supplied best-known reference values with explicit provenance.ensure_equal_budgets— fail-closed verification of the size-scaled equal-budget protocol.
Analysis-method registry (fail-closed method selection):
AnalysisMethod/method_registry/method_by_id— named methods with their assumptions and disclosure floors.AnalysisFamily/methods_in_family— method families and their membership.MethodSupport/evaluate_support— the fail-closed support verdict for a method on a given sample size.MethodLimitation— an unsupported named method routed to the limitations surface.STATISTICAL_POWER_FLOOR— the minimum sample size below which underpowered methods fail closed.
Analysis frames (auxiliary row and series frames a figure or table consumer renders):
AnalysisFrames— the bundle of auxiliary frames for one dataset.BenchmarkRow/benchmark_rows— per-benchmark-instance coverage across the solver field.CharacterizationRow/characterization_rows— per-problem difficulty descriptors from observed spread.ConstraintRow— one constraint-accounting row over the dataset.ConvergenceBand/convergence_bands— per-solver per-iteration best-objective minimum/median/maximum across runs for the convergence-band figure.ConvergenceSeries/convergence_series— per-solver best-objective traces by iteration.TimedConvergenceSeries/timed_convergence_series— per-solver best-objective traces by elapsed wall-clock seconds, aggregated over the union of the solver's sample times; runs whose samples carry noelapsed_secondsare omitted rather than plotted against a fabricated axis.DataProfileSeries/data_profile_series— per-solver share of cells solved to a target accuracy within an evaluation-group budget for the data-profile figure.DiversitySeries/diversity_series— per-solver population-diversity traces by iteration for the exploration-exploitation balance figure.DiversityFitnessPortrait/diversity_fitness_portrait— per-solver joint diversity-versus-incumbent-objective path for the diversity-objective phase-portrait figure.GbestLbestSeries/gbest_lbest_series— per-solver global-best and population-best objective by iteration for the gbest/lbest figure.InfeasibleRunRow/infeasible_rows— infeasible completed runs and recorded failures routed to supplements.AdaptabilityPoint/adaptability_points— per-solver cross-instance consistency (coefficient of variation of the per-problem performance ratio) for the adaptability figure.OptimalityGapRow/optimality_gap_rows— per-(problem, solver) optimality gaps against reference bounds.ReliabilityPoint/reliability_points— per-solver fraction of completed runs that returned a feasible schedule for the reliability figure.RobustnessPoint/robustness_points— per-solver CVaR + worst-case tail risk over per-problem performance ratios (each run's objective over the problem's best across the field) for the robustness figure.RuntimeQualityPoint/runtime_quality_points— runtime-quality observations for the runtime-quality figure.ScalabilitySeries/scalability_series— per-solver mean-runtime-by-instance-scale traces for the scalability figure.SearchSpaceAnimation/search_space_animation— per-solver per-iteration population clouds for the animated search-space flipbook (SVG-native).SearchSpace3DScatter/search_space_3d_scatter— per-solver final-population positions projected to three representative dimensions for the 3D search-space scatter figure.SearchSpaceScatter/search_space_scatter— per-solver final-population positions projected to two representative dimensions for the search-space scatter figure.SearchSpaceTrajectory/search_space_trajectory— per-solver population-centroid path through two representative dimensions for the search-space trajectory figure.SensitivityPoint— one parameter-level mean for a sensitivity sweep.StabilityPoint/stability_points— per-solver seed-stability (mean coefficient of variation) for the stability figure.AblationRow— one mechanism-isolation outcome versus the full configuration.
Multi-objective indicators:
MultiObjectiveIndicatorRow/multiobjective_indicators— the four named quality indicators per solver's approximation front.MultiObjectiveIndicatorReport/multiobjective_indicator_report— indicator rows beside the archived shared reference point that produced them.ReferencePointRule/ReferencePointRecord— the reference-point derivation rules and the archived record with its fail-closed provenance vocabulary.ensure_shared_reference_points— fail-closed check that compared artifacts share one archived reference point.DistributionBox/distribution_boxes— per-solver five-number summaries of the feasible objective set for the distribution figure.quality_indicator_method_ids— the registry method ids this surface computes end to end.
Platform comparison (capability and distribution-distance evidence):
PlatformComparison/platform_comparison_markdown— the four platform-comparison row sets and their Markdown.RivalFrameworkRow/rival_framework_markdown— capability comparison against a rival framework.FeatureRichnessRow/feature_richness_markdown— the feature-coverage matrix.DistributionDistanceRow/distribution_distance_markdown— the distribution-distance bridge assessment per benchmark family.ArtifactInspectionRow/artifact_inspection_markdown— the portal artifact-inspection evidence.EvidenceSource— how a platform-comparison row's claim is supported.
Report scaffolding, claim gates, and redaction:
ReportScaffold/ReportScaffoldSpec/ReportProfile/ReportSection/CANONICAL_SECTION_ORDER/write_report_scaffold— deterministic, source-linked, disclosure-filtered report scaffolds.ClaimGate/ClaimGateKind/ClaimGateState/ClaimGateOutcome/evaluate_claim_gate— evidence-gated report claims and their promotion verdicts.ContributionMatrix/ContributionRow/ContributionOverlap/contribution_partition— a zero-overlap contribution partition across evidence tiers.RedactionFinding/redaction_report/assert_redaction_clean— scanning a candidate public surface for forbidden terms and cross-tier leaks.ReproducibilityBadge— a reproducibility-badge target and how the artifact index satisfies it.MaintenanceLedgerEntry/maintenance_ledger_markdown— future-extension surfaces that must not weaken the disclosure boundary.
Reproducibility manifest:
ReproducibilityManifest/build_reproducibility_manifest/reproducibility_manifest_markdown— the per-run reproducibility facts and their Markdown document.FIGURE_GENERATORS— the registry of figure generators the manifest records.
Solver glossary (metaphor-free naming surface):
SOLVER_OPERATIONAL_DESCRIPTIONS/solver_operational_description— the metaphor-free operational description per solver.SOLVER_EQUIVALENCE_NOTES/solver_equivalence_note— the known-algorithm equivalence note per solver.INSPIRATION_DOMAIN_TERMS/metaphor_terms_in— the inspiration-domain term inventory and its detection.
Display-name registries:
BENCHMARK_DISPLAY_NAMES/benchmark_display_name— registered display names for benchmark identifiers.SOLVER_DISPLAY_NAMES/solver_display_name— registered display names for solver identifiers.ABLATION_DISPLAY_NAMES/ablation_display_name— display names for mechanism-isolation ablation variants.benchmark_instance_label— a display label that keeps instance identity visible.UnknownDisplayNameError— raised when an identifier has no registered display name.
Evidence tiers:
Tier/tier/TIER_REGISTRY— the public evidence tiers and where each tier's evidence may appear.UnknownTierError— raised when a tier identifier has no registry entry.
Figure theme:
FigureTheme/DEFAULT_THEME— externally controllable colours, typography, and canvas width shared across every chart family.