Exhaustive Machine Learning

Tests of the Exhaustive machine-learning training pipeline: dataset assembly, variable preparation, training trials, promotion and model recall in the engine.

General

Accuracy

Jube.Test.Exhaustive.Accuracy — 4 tests.

Test	Kind	Verifies
AssertAccuracy_BelowThreshold_Fails	Fact	Assert accuracy below threshold fails.
AssertAccuracy_NoRows_Fails	Fact	Assert accuracy no rows fails.
AssertAccuracy_AboveThreshold_Passes	Fact	Assert accuracy above threshold passes.
BatchAccuracyAsync	Theory	Batch accuracy.

AnomalyRunnerTests

Jube.Test.Exhaustive.AnomalyRunnerTests — 12 tests.

Test	Kind	Verifies
TheOneClassSvmPathProducesTheSameLabelsAsTheOriginalCodePath	Fact	The one class svm path produces the same labels as the original code path.
TheOneClassSvmAdapterScoresTheNegatedProbabilityAndIsThreadSafe	Fact	The one class svm adapter scores the negated probability and is thread safe.
TheAdapterHonoursCancellationAndRefusesUnfittedUse	Fact	The adapter honours cancellation and refuses unfitted use.
ANativeDetectorFlagsTheContaminationFractionAndReportsTheReduction	Fact	A native detector flags the contamination fraction and reports the reduction.
RunsAreDeterministic	Fact	Runs are deterministic.
ReductionCanBeSwitchedOffForANativeDetector	Fact	Reduction can be switched off for a native detector.
TheContaminationParameterOverridesTheProbabilityAndIsValidated	Fact	The contamination parameter overrides the probability and is validated.
AutomaticRecordsEveryCandidateAndResolvesTheAlgorithm	Fact	Automatic records every candidate and resolves the algorithm.
AutomaticUsesTrustedPositivesWhenSupplied	Fact	Automatic uses trusted positives when supplied.
ACancelledRunThrows	Fact	A cancelled run throws.
TheAutomaticCandidateSetIsRestrictedByDimensionalityAndRows	Fact	The automatic candidate set is restricted by dimensionality and rows.
AnExplicitCandidateListOverridesTheDefaultsButNeverAdmitsTheSvm	Fact	An explicit candidate list overrides the defaults but never admits the svm.

CategoricalRecallHttpTests

Jube.Test.Exhaustive.CategoricalRecallHttpTests — 2 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
ACategoricalStringIsAcceptedAndScoredThroughTheEncoderAsync	Theory	A categorical string is accepted and scored through the encoder.
WrongTypesAreStillRejectedWith400Async	Theory	Wrong types are still rejected with400.

CategoricalRecallTests

Jube.Test.Exhaustive.CategoricalRecallTests — 14 tests. Category: Unit.

Test	Kind	Verifies
Sync_BuildsTheEncoderFromTheStoredVocabularyAndSharesItAcrossIndicators	Fact	Sync builds the encoder from the stored vocabulary and shares it across indicators.
Sync_AnInvalidVocabularyReportsALoadError	Fact	Sync an invalid vocabulary reports a load error.
Sync_ModelsWithoutCategoricalMetadataAreUnchanged	Fact	Sync models without categorical metadata are unchanged.
BuildScoreInputs_ComputesTheIndicatorsFromTheIncomingString	Theory	Build score inputs computes the indicators from the incoming string.
Score_AppliesTheNetworkToTheIndicators	Fact	Score applies the network to the indicators.
Score_ANormalisedIndicatorIsStandardisedLikeTraining	Fact	Score a normalised indicator is standardised like training.
EliminatedIndicatorsFoldTheLevelIntoTheReference	Fact	Eliminated indicators fold the level into the reference.
PipelineAndHttpRecallComputeIdenticalInputsForTheSameString	Theory	Pipeline and http recall compute identical inputs for the same string.
Pipeline_AbsentPayloadValueIsMissing	Fact	Pipeline absent payload value is missing.
Pipeline_NumericPayloadValueBecomesItsInvariantString	Fact	Pipeline numeric payload value becomes its invariant string.
TypeGate_AcceptsStringsForACategoricalSourceAndStillRejectsWrongTypesElsewhere	Fact	Type gate accepts strings for a categorical source and still rejects wrong types elsewhere.
TypeGate_TheIndicatorNameIsNotARequestKey	Fact	Type gate the indicator name is not a request key.
TypeGate_ModelsWithoutCategoricalMetadataRejectStringsAsBefore	Fact	Type gate models without categorical metadata reject strings as before.
HostileStringsNeverThrowOrProduceNonFiniteInputs	Fact	Hostile strings never throw or produce non finite inputs.

CategoricalTrainingTests

Jube.Test.Exhaustive.CategoricalTrainingTests — 5 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
Extraction_ExplodesAPayloadStringFieldIntoIndicatorColumnsFittedOnBothClassesAsync	Fact	Extraction explodes a payload string field into indicator columns fitted on both classes.
Extraction_WithoutTheClassRowsInTheFitAFraudOnlyLevelWouldBeLostAsync	Fact	Extraction without the class rows in the fit a fraud only level would be lost.
Extraction_WithoutTheOptionOnlyTheNumericFieldsAreReadAsBeforeAsync	Fact	Extraction without the option only the numeric fields are read as before.
Extraction_AnUndeclaredTypeIsInferredFromTheSampleAsync	Fact	Extraction an undeclared type is inferred from the sample.
Training_ReachesAPromotedTrialWithIndicatorVariablesAndPersistsTheMetadataAsync	Fact	Training reaches a promoted trial with indicator variables and persists the metadata.

ChallengerCategoricalRecallTests

Jube.Test.Exhaustive.ChallengerCategoricalRecallTests — 7 tests. Category: Unit.

Test	Kind	Verifies
Sync_AChallengerKindBuildsChallengerRecallNotATopologyNetwork	Fact	Sync a challenger kind builds challenger recall not a topology network.
PipelineAndExternalRecallComputeIdenticalScoresForAChallengerModel	Theory	Pipeline and external recall compute identical scores for a challenger model.
ChallengerCategoricalIndicatorIsNeverZScored	Fact	Challenger categorical indicator is never z scored.
ChallengerContinuousVariableIsNotStandardisedEither	Fact	Challenger continuous variable is not standardised either.
Pipeline_AbsentPayloadValueFallsBackToTheReferenceIndicatorAndTrainingMean	Fact	Pipeline absent payload value falls back to the reference indicator and training mean.
Pipeline_NumericPayloadValueBecomesItsInvariantStringForAChallengerModel	Fact	Pipeline numeric payload value becomes its invariant string for a challenger model.
HostileStringsNeverThrowOrProduceNonFiniteInputsForAChallengerModel	Fact	Hostile strings never throw or produce non finite inputs for a challenger model.

ChallengerModelTests

Jube.Test.Exhaustive.ChallengerModelTests — 7 tests.

Test	Kind	Verifies
ChallengerProbability_IsBaseRateWhenComplexityIsZero	Fact	Challenger probability is base rate when complexity is zero.
ChallengerProbability_RampsTowardsOneAsComplexityApproachesTheLimit	Fact	Challenger probability ramps towards one as complexity approaches the limit.
ChallengerProbability_IsUnaffectedByAZeroLimit	Fact	Challenger probability is unaffected by a zero limit.
SelectFamily_NeverChallengesWhenProbabilityIsZero	Fact	Select family never challenges when probability is zero.
SelectFamily_AlwaysReturnsARecognisedKindWhenProbabilityIsOne	Fact	Select family always returns a recognised kind when probability is one.
Train_ProducesARecognisableModelWithSaneRecall	Theory	Train produces a recognisable model with sane recall.
Train_UnrecognisedKindThrows	Fact	Train unrecognised kind throws.

DemoDataSourceTests

Jube.Test.Exhaustive.DemoDataSourceTests — 3 tests. Collection: Database.

Test	Kind	Verifies
DemoDataset_LoadsTheBuiltInRowsOncePerModelIntoMockArchiveAsync	Fact	Demo dataset loads the built in rows once per model into mock archive.
ArchiveDataSource_NeverLoadsTheDemoRowsAsync	Fact	Archive data source never loads the demo rows.
TheRetiredEnvironmentVariable_HasNoEffectOnEitherDataSourceAsync	Fact	The retired environment variable has no effect on either data source.

ExhaustiveBaseRateTests

Jube.Test.Exhaustive.ExhaustiveBaseRateTests — 9 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
Archive_TheBaseRateIsTheCountOfFilterMatchesOverThePopulationAsync	Fact	Archive the base rate is the count of filter matches over the population.
Archive_TheSampleCapsDoNotDistortTheBaseRateAsync	Fact	Archive the sample caps do not distort the base rate.
Archive_OtherModelsAndTheMockTableAreNotCountedAsync	Fact	Archive other models and the mock table are not counted.
Demo_TheMockArchiveIsCountedInsteadOfTheArchiveAsync	Fact	Demo the mock archive is counted instead of the archive.
Archive_AFilterThatMatchesNothingOrEverythingHasNoUsableBaseRateAsync	Fact	Archive a filter that matches nothing or everything has no usable base rate.
Archive_AFailureIsLoggedAndFallsBackToTheSamplePriorAsync	Fact	Archive a failure is logged and falls back to the sample prior.
Dataset_TheClassProportionIsCountedOverTheWholeDatasetAsync	Fact	Dataset the class proportion is counted over the whole dataset.
Dataset_TheClassProportionIsCountedForBothStoredDatasetKindsAsync	Theory	Dataset the class proportion is counted for both stored dataset kinds.
TheCountSql_IsScopedToTheKeyAndTheFilter	Fact	The count sql is scoped to the key and the filter.

ExhaustiveCalibrationConceptTests

Jube.Test.Exhaustive.ExhaustiveCalibrationConceptTests — 5 tests. Category: Unit.

Test	Kind	Verifies
AtATwoPercentBaseRate_TheCalibratedMeanMatchesAndTheRawMeanDoesNotAsync	Fact	At a two percent base rate the calibrated mean matches and the raw mean does not.
AtATwoPercentBaseRate_EceBrierAndLogLossImproveOnTheDeploymentPopulation_AndAucIsUnchangedAsync	Fact	At a two percent base rate ece brier and log loss improve on the deployment population and auc is unchanged.
TheStoredQuality_ShowsTheImprovementOnTheBalancedTestSplitAndAtTheBaseRateAsync	Fact	The stored quality shows the improvement on the balanced test split and at the base rate.
TheOptimismOfCalibratingOnTheSelectionSplitIsSmallComparedWithACalibratorFittedOnTheTestSplitItselfAsync	Fact	The optimism of calibrating on the selection split is small compared with a calibrator fitted on the test split itself.
TheCalibratorIsFittedOnTheValidationSplit_NotOnTheTestLabelsAsync	Fact	The calibrator is fitted on the validation split not on the test labels.

ExhaustiveCalibrationFitterTests

Jube.Test.Exhaustive.ExhaustiveCalibrationFitterTests — 26 tests. Category: Unit.

Test	Kind	Verifies
Fit_SigmoidOutput_UsesTheLogitInputAndMatchesTheIndependentReference	Fact	Fit sigmoid output uses the logit input and matches the independent reference.
Fit_LinearOutput_UsesTheRawInput	Fact	Fit linear output uses the raw input.
Fit_BoundedAndUnboundedNonSigmoidFamilies_UseTheRawInput	Theory	Fit bounded and unbounded non sigmoid families use the raw input.
ApplyingTheStoredColumns_ReproducesTheReferenceProbabilitiesAtTheBaseRate	Fact	Applying the stored columns reproduces the reference probabilities at the base rate.
Fit_BeforeAndAfterQuality_MatchesTheIndependentReference	Fact	Fit before and after quality matches the independent reference.
Fit_QualityRoundTripsThroughItsJson	Fact	Fit quality round trips through its json.
Fit_AnUnusableBaseRate_LeavesTheProbabilityAtTheSamplePrior	Fact	Fit an unusable base rate leaves the probability at the sample prior.
Fit_HardOutputFamilies_AreSkippedWithARecordedReason	Theory	Fit hard output families are skipped with a recorded reason.
Fit_SingleClassValidation_IsSkipped	Fact	Fit single class validation is skipped.
Fit_ConstantScores_AreSkipped	Fact	Fit constant scores are skipped.
Fit_ScoresWithTwoDistinctValues_AreSkippedAsAHardOutput	Fact	Fit scores with two distinct values are skipped as a hard output.
Fit_IsotonicRequested_FitsAMonotoneStepFunctionAndImprovesLogLoss	Fact	Fit isotonic requested fits a monotone step function and improves log loss.
Fit_BetaRequested_OnASigmoidFamily_FitsAndImprovesLogLoss	Fact	Fit beta requested on a sigmoid family fits and improves log loss.
Fit_BetaRequested_OnANonSigmoidFamily_IsSkippedWithAReason	Fact	Fit beta requested on a non sigmoid family is skipped with a reason.
Fit_Automatic_PicksACandidateAtLeastAsGoodAsPlattOnValidationBrier	Fact	Fit automatic picks a candidate at least as good as platt on validation brier.
Fit_Automatic_OnANonSigmoidFamily_NeverPicksBeta	Fact	Fit automatic on a non sigmoid family never picks beta.
Fit_AnUnknownRequestedMethod_FallsBackToAutomatic	Fact	Fit an unknown requested method falls back to automatic.
ApplyingStoredIsotonicColumns_RoundTripsThroughFromColumns	Fact	Applying stored isotonic columns round trips through from columns.
ApplyingStoredBetaColumns_RoundTripsThroughFromColumns	Fact	Applying stored beta columns round trips through from columns.
FromColumns_AnUnknownMethod_IsInvalidAndThrowsOnApply	Fact	From columns an unknown method is invalid and throws on apply.
Fit_NonFiniteScores_AreSkipped	Fact	Fit non finite scores are skipped.
Fit_TooFewPositives_IsSkippedAndTheMinimumIsTwenty	Fact	Fit too few positives is skipped and the minimum is twenty.
Fit_BetweenTwentyAndFortyNinePositives_FitsWithALowSampleWarning	Fact	Fit between twenty and forty nine positives fits with a low sample warning.
Fit_ScoresThatRankBackwards_AreSkippedRatherThanInverted	Fact	Fit scores that rank backwards are skipped rather than inverted.
Fit_FlippingTheTestLabels_DoesNotChangeTheCalibrator_ButChangingTheValidationLabelsDoes	Fact	Fit flipping the test labels does not change the calibrator but changing the validation labels does.
BaseRateResolution_FromCountsOverrideAndAnomalyLabels	Fact	Base rate resolution from counts override and anomaly labels.

ExhaustiveCalibrationRecallPathsTests

Jube.Test.Exhaustive.ExhaustiveCalibrationRecallPathsTests — 7 tests. Category: Unit.

Test	Kind	Verifies
ScoreRecallAndTheInPipelineStage_ReturnTheSameCalibratedValue	Theory	Score recall and the in pipeline stage return the same calibrated value.
AModelWithoutACalibrator_BehavesByteForByteAsBefore	Theory	A model without a calibrator behaves byte for byte as before.
AnInvalidCalibrator_DoesNotServeRawScoresOnAnyPath	Fact	An invalid calibrator does not serve raw scores on any path.
Sync_BuildsTheCalibrationFromThePromotedTrialRow	Fact	Sync builds the calibration from the promoted trial row.
Sync_ForAModelPromotedBeforeCalibration_LoadsTheIdentity	Fact	Sync for a model promoted before calibration loads the identity.
Sync_ForASkippedCalibration_LoadsTheIdentityAndKeepsScoringRaw	Fact	Sync for a skipped calibration loads the identity and keeps scoring raw.
Sync_ForCorruptColumns_RegistersTheModelLogsAnErrorAndMarksItInvalid	Fact	Sync for corrupt columns registers the model logs an error and marks it invalid.

ExhaustiveCalibrationValueTests

Jube.Test.Exhaustive.ExhaustiveCalibrationValueTests — 8 tests. Category: Unit.

Test	Kind	Verifies
NoStoredMethod_IsTheIdentityAndReturnsTheRawScoreByteForByte	Theory	No stored method is the identity and returns the raw score byte for byte.
Apply_IsPlattThenThePriorShiftToTheBaseRate	Fact	Apply is platt then the prior shift to the base rate.
Apply_WithoutABaseRate_StaysAtTheSamplePrior	Fact	Apply without a base rate stays at the sample prior.
Apply_IsStrictlyMonotoneInTheRawScore	Fact	Apply is strictly monotone in the raw score.
Apply_RawInputMode_UsesTheScoreDirectly	Fact	Apply raw input mode uses the score directly.
Apply_NonFiniteRawValuesPassThroughUnchanged	Fact	Apply non finite raw values pass through unchanged.
AnUnknownMethod_IsInvalid	Theory	An unknown method is invalid.
CorruptColumns_AreInvalidAndNeverSilentlyRaw	Fact	Corrupt columns are invalid and never silently raw.

ExhaustiveRecallTests

Jube.Test.Exhaustive.ExhaustiveRecallTests — 5 tests.

Test	Kind	Verifies
BuildScoreInputs_AllPresent_StandardisesType2Only	Fact	Build score inputs all present standardises type2 only.
BuildScoreInputs_MissingKeys_AreMeanImputed_NotZero	Fact	Build score inputs missing keys are mean imputed not zero.
BuildScoreInputs_MissingMatchesInPipelineFallback	Fact	Build score inputs missing matches in pipeline fallback.
BuildScoreInputs_JsonNullAndUnconvertibleValues_AreTreatedAsMissing	Fact	Build score inputs json null and unconvertible values are treated as missing.
BuildScoreInputs_MatchIsCaseSensitiveAndExact	Fact	Build score inputs match is case sensitive and exact.

InactiveModelRecallHttpTests

Jube.Test.Exhaustive.InactiveModelRecallHttpTests — 4 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
AnInactiveInstanceStillRecallsThroughTheDirectInvokePathAsync	Fact	An inactive instance still recalls through the direct invoke path.
AnInactiveInstanceStillScoresThroughTheBatchInvokePathAsync	Fact	An inactive instance still scores through the batch invoke path.
ATenantForeignUserIsForbiddenOnTheDirectInvokePathForAnInactiveInstanceAsync	Fact	A tenant foreign user is forbidden on the direct invoke path for an inactive instance.
ATenantUserWithoutTheModelRoleOrPermissionIsForbiddenForAnInactiveInstanceAsync	Fact	A tenant user without the model role or permission is forbidden for an inactive instance.

LoadDatabaseTests

Jube.Test.Exhaustive.LoadDatabaseTests — 23 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
RealFields_OnlyTtlCounterAbstractionAndAdaptationFieldsSurviveInQueryOrderAsync	Fact	Real fields only ttl counter abstraction and adaptation fields survive in query order.
RealFields_InactiveDeletedAndOtherTenantFieldsAreExcludedAsync	Fact	Real fields inactive deleted and other tenant fields are excluded.
RealFields_ValuesAreExtractedByTheJsonPathAndMissingIsZeroAsync	Fact	Real fields values are extracted by the json path and missing is zero.
RealSample_IsOrderedByEntryGuidAndCappedAtTenThousandAsync	Fact	Real sample is ordered by entry guid and capped at ten thousand.
RealSample_ExactlyAtAndJustBelowTheCapIsNotTruncatedAsync	Fact	Real sample exactly at and just below the cap is not truncated.
RealSample_OtherModelsAndTheMockTableAreNotMixedInAsync	Fact	Real sample other models and the mock table are not mixed in.
RealSample_ModelWithNoArchiveRowsGivesVariablesAndNoRowsAsync	Fact	Real sample model with no archive rows gives variables and no rows.
MockSample_ReadsMockArchiveWithTheFixedVariableListIgnoringModelFieldsAsync	Fact	Mock sample reads mock archive with the fixed variable list ignoring model fields.
MockSample_IsOrderedByEntryGuidAndCappedAtTenThousandAsync	Fact	Mock sample is ordered by entry guid and capped at ten thousand.
ClassData_ReturnsOnlyRowsMatchingTheFilterLabelledOneAsync	Fact	Class data returns only rows matching the filter labelled one.
ClassData_WorksWithTheSelectOnlyGuardEnabledAsync	Fact	Class data works with the select only guard enabled.
ClassData_ClassCapIsTenThousandAsync	Fact	Class data class cap is ten thousand.
ClassData_MockTableIsUsedWhenMockDataIsSetAsync	Fact	Class data mock table is used when mock data is set.
NegativeSample_IsTheComplementOfTheFilterAndDisjointFromTheClassAsync	Fact	Negative sample is the complement of the filter and disjoint from the class.
NegativeSample_OfAStrictGreaterThanFilter_KeepsTheBoundaryValueAsync	Fact	Negative sample of a strict greater than filter keeps the boundary value.
NegativeSample_RowsMissingTheFilteredKeyAreInNeitherClassAsync	Fact	Negative sample rows missing the filtered key are in neither class.
NegativeSample_CapIsTenThousandAndAppliesIndependentlyOfTheClassCapAsync	Fact	Negative sample cap is ten thousand and applies independently of the class cap.
NegativeSample_MockDataFilterUsesTheMockTableAndFixedVariablesAsync	Fact	Negative sample mock data filter uses the mock table and fixed variables.
LoadMock_PopulatesMockArchiveOncePerModelFromTheEmbeddedCsvAsync	Fact	Load mock populates mock archive once per model from the embedded csv.
LoadMock_RoundTripsEveryCsvRowThroughJsonBackToTheSameFeatureMatrixAsync	Fact	Load mock round trips every csv row through json back to the same feature matrix.
LoadMock_TagsFraudExactlyWhenTheDependentColumnIsPositiveAsync	Fact	Load mock tags fraud exactly when the dependent column is positive.
LoadMock_DoesNotTouchAnotherModelAndSkipsOnlyTheModelAlreadyLoadedAsync	Fact	Load mock does not touch another model and skips only the model already loaded.
LoadMock_SkipsWhenTheModelAlreadyHasAnyMockRowAsync	Fact	Load mock skips when the model already has any mock row.

LoadJsonExtractionTests

Jube.Test.Exhaustive.LoadJsonExtractionTests — 13 tests.

Test	Kind	Verifies
ExtractValue_ReadsTheNumberAtThePath	Theory	Extract value reads the number at the path.
ExtractValue_MissingNonNumericOrMalformedPathIsZero	Theory	Extract value missing non numeric or malformed path is zero.
ExtractValue_ConsecutiveDotsInThePathAreTolerated	Fact	Extract value consecutive dots in the path are tolerated.
ExtractValue_MissingAndGenuineZeroAreIndistinguishable	Fact	Extract value missing and genuine zero are indistinguishable.
ExtractValue_BooleanAndDateTokensConvertTheWayNewtonsoftConvertsThem	Fact	Extract value boolean and date tokens convert the way newtonsoft converts them.
BuildRows_OneRowPerJsonWithOneColumnPerVariableInVariableOrder	Fact	Build rows one row per json with one column per variable in variable order.
BuildRows_MissingValuesBecomeZeroWithoutDroppingTheRowOrShorteningIt	Fact	Build rows missing values become zero without dropping the row or shortening it.
BuildRows_ExtraJsonFieldsAreIgnored	Fact	Build rows extra json fields are ignored.
BuildRows_EmptyJsonListGivesNoRows	Fact	Build rows empty json list gives no rows.
BuildRows_NoVariablesGivesEmptyRowsOnePerJson	Fact	Build rows no variables gives empty rows one per json.
BuildRows_UnparseableJsonThrowsRatherThanBeingSkipped	Fact	Build rows unparseable json throws rather than being skipped.
BuildRows_VariableKeysAreReadAsAZeroBasedSequence	Fact	Build rows variable keys are read as a zero based sequence.
BuildRows_ManyRowsKeepInputOrder	Fact	Build rows many rows keep input order.

LoadSampleSqlTests

Jube.Test.Exhaustive.LoadSampleSqlTests — 6 tests.

Test	Kind	Verifies
SampleLimitIsTenThousand	Fact	Sample limit is ten thousand.
RealArchiveSample_SelectsFromArchiveOrderedByEntryGuidWithModelIdAndLimitAsTheLastTwoParameters	Fact	Real archive sample selects from archive ordered by entry guid with model id and limit as the last two parameters.
MockArchiveSample_SelectsFromMockArchive	Fact	Mock archive sample selects from mock archive.
ParameterNumbersFollowTheTokenCount	Theory	Parameter numbers follow the token count.
NegatedFilterIsWrappedSoTheNegationCoversTheWholeClause	Fact	Negated filter is wrapped so the negation covers the whole clause.
OrderIsAlwaysByEntryGuidAndNeverRandom	Fact	Order is always by entry guid and never random.

LoadVariableSelectionTests

Jube.Test.Exhaustive.LoadVariableSelectionTests — 8 tests.

Test	Kind	Verifies
BuildVariablesFromFields_KeepsOnlyProcessingTypesThreeFiveAndSeven	Theory	Build variables from fields keeps only processing types three five and seven.
BuildVariablesFromFields_PreservesQueryOrderAndNumbersKeysFromZeroWithoutGaps	Fact	Build variables from fields preserves query order and numbers keys from zero without gaps.
BuildVariablesFromFields_CarriesNameProcessingTypeAndJsonPath	Fact	Build variables from fields carries name processing type and json path.
BuildVariablesFromFields_NoFieldsOrNoEligibleFieldsGivesNoVariables	Fact	Build variables from fields no fields or no eligible fields gives no variables.
BuildMockVariables_IsTwentySixAbstractionsInCsvHeaderOrderEachOnce	Fact	Build mock variables is twenty six abstractions in csv header order each once.
BuildMockVariables_EveryVariableIsProcessingTypeFiveWithADollarAbstractionPath	Fact	Build mock variables every variable is processing type five with a dollar abstraction path.
BuildMockVariables_ReturnsAFreshDictionaryEachTime	Fact	Build mock variables returns a fresh dictionary each time.
BuildMockVariables_MockJsonWithEveryFeatureProducesAFullyPopulatedRow	Fact	Build mock variables mock json with every feature produces a fully populated row.

MockCategoricalDatasetTests

Jube.Test.Exhaustive.MockCategoricalDatasetTests — 3 tests.

Test	Kind	Verifies
GenerateResourcesWhenRequested	Fact	Generate resources when requested.
TheCommittedFileIsRowAlignedWithMockCsvAndEveryCategoryIsAKnownLabel	Fact	The committed file is row aligned with mock csv and every category is a known label.
EveryCategoricalColumnUsesEveryOneOfItsLabelsAtLeastOnce	Fact	Every categorical column uses every one of its labels at least once.

MockVariableListTests

Jube.Test.Exhaustive.MockVariableListTests — 1 test.

Test	Kind	Verifies
MockVariables_MatchTheCsvFeatureColumnsOnceEach_Async	Fact	Mock variables match the csv feature columns once each.

NetworkRoundTripTests

Jube.Test.Exhaustive.NetworkRoundTripTests — 1 test. Category: Unit.

Test	Kind	Verifies
ATrainedNetworkOfEveryActivationFunctionSurvivesTheEngineRoundTrip	Theory	A trained network of every activation function survives the engine round trip.

ParserJsonToSqlEntityAnalysisModelTests

Jube.Test.Exhaustive.ParserJsonToSqlEntityAnalysisModelTests — 12 tests. Category: Unit.

Test	Kind	Verifies
SingleEqualRule_ProducesOneParameterisedComparison	Fact	Single equal rule produces one parameterised comparison.
EveryComparisonOperator_MapsToItsSqlToken	Theory	Every comparison operator maps to its sql token.
NotEqual_IsNegated	Fact	Not equal is negated.
NestedGroup_WrapsInParenthesesAndJoinsWithCondition	Fact	Nested group wraps in parentheses and joins with condition.
InOperator_WithAnArrayValue_ProducesOnePlaceholderPerValue	Fact	In operator with an array value produces one placeholder per value.
NotInOperator_IsNegated	Fact	Not in operator is negated.
InOperator_WithASingleScalarValue_StillWorks	Fact	In operator with a single scalar value still works.
InOperator_WithNoValues_Throws	Fact	In operator with no values throws.
MixedGroup_NumbersTokensAcrossAnInRuleAndAScalarRuleCorrectly	Fact	Mixed group numbers tokens across an in rule and a scalar rule correctly.
UnknownField_Throws	Fact	Unknown field throws.
UnknownOperator_Throws	Fact	Unknown operator throws.
ANullRule_ProducesNoSql	Fact	A null rule produces no sql.

RecallBatchHttpTests

Jube.Test.Exhaustive.RecallBatchHttpTests — 3 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
EveryRowIsScoredThroughTheSameRecallPathAsASingleRowAsync	Fact	Every row is scored through the same recall path as a single row.
AnUnparseableValueIsReportedAsARowErrorNotABatchFailureAsync	Fact	An unparseable value is reported as a row error not a batch failure.
AColumnWithNoMatchingTrainedVariableIsReportedAsUnmappedAsync	Fact	A column with no matching trained variable is reported as unmapped.

RecallChainTests

Jube.Test.Exhaustive.RecallChainTests — 1 test. Category: Service. Collection: Database.

Test	Kind	Verifies
ATrainedAndPromotedModelIsSyncedAndRecalledThroughBothPathsAsync	Fact	A trained and promoted model is synced and recalled through both paths.

RecallExplainabilityTests

Jube.Test.Exhaustive.RecallExplainabilityTests — 11 tests.

Test	Kind	Verifies
Explain_OneLayerIdentityNetwork_ConnectionWeightIsWeightTimesInput	Fact	Explain one layer identity network connection weight is weight times input.
Explain_OneLayerIdentityNetwork_LrpRelevanceAndBaseValueReconstructTheOutputExactly	Fact	Explain one layer identity network lrp relevance and base value reconstruct the output exactly.
Explain_AnIgnoredInputWithZeroWeightContributesZeroToBothMeasures	Fact	Explain an ignored input with zero weight contributes zero to both measures.
Explain_TwoLayerSigmoidNetwork_LrpRelevancePlusBaseValueReconstructsTheOutput	Fact	Explain two layer sigmoid network lrp relevance plus base value reconstructs the output.
Explain_TwoLayerSigmoidNetwork_ConnectionWeightIsSignedByThePathWeightProduct	Fact	Explain two layer sigmoid network connection weight is signed by the path weight product.
Explain_ReconstructionHoldsRegardlessOfWhichActivationFunctionTheTopologySearchPicked	Theory	Explain reconstruction holds regardless of which activation function the topology search picked.
Explain_ADeadRectifiedLinearHiddenNeuronPropagatesZeroRelevanceThroughItAndStillReconstructs	Fact	Explain a dead rectified linear hidden neuron propagates zero relevance through it and still reconstructs.
Explain_NodeRelevanceAtTheInputBoundaryMatchesEachVariablesLrpRelevance	Fact	Explain node relevance at the input boundary matches each variables lrp relevance.
Explain_NodeRelevanceAtTheOutputBoundaryIsTheRawNetworkOutput	Fact	Explain node relevance at the output boundary is the raw network output.
Explain_EdgeRelevanceHasTheSameShapeAsEachLayersWeightMatrix	Fact	Explain edge relevance has the same shape as each layers weight matrix.
Explain_WithZeroThresholdsEachNeuronsIncomingEdgeRelevanceSumsToItsOwnNodeRelevance	Fact	Explain with zero thresholds each neurons incoming edge relevance sums to its own node relevance.

RecallHttpTests

Jube.Test.Exhaustive.RecallHttpTests — 13 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
ResponseIsTheScoreRoundedToFourDecimalPlacesAsync	Theory	Response is the score rounded to four decimal places.
ACalibratedModelReturnsTheCalibratedProbabilityRoundedToFourDecimalsAsync	Theory	A calibrated model returns the calibrated probability rounded to four decimals.
ASmallProbabilityKeepsItsResolutionAtFourDecimalsAsync	Fact	A small probability keeps its resolution at four decimals.
ResponseIsNotAnUnroundedScoreAsync	Fact	Response is not an unrounded score.
ANonFiniteScoreIsA500WithNoBodyAsync	Fact	A non finite score is a 500 with no body.
ANetworkThatThrowsIsSwallowedToAZeroScoreWith200Async	Fact	A network that throws is swallowed to a zero score with200.
AMissingKeyIsImputedWithTheMeanAsInThePipelineAsync	Fact	A missing key is imputed with the mean as in the pipeline.
AStandardisedVariableUsesTheStoredMeanAndSdAsync	Fact	A standardised variable uses the stored mean and sd.
VariablesAreConsumedInStoredOrderNotBodyOrderAsync	Fact	Variables are consumed in stored order not body order.
ExtraKeysInTheBodyAreIgnoredAsync	Fact	Extra keys in the body are ignored.
TheResponseCarriesPerVariableRelevanceThatReconstructsTheRawScoreAsync	Fact	The response carries per variable relevance that reconstructs the raw score.
TypeGateRejectionsAre400WithAPlainMessageAsync	Theory	Type gate rejections are400 with a plain message.
AnUnknownGuidIs404Async	Fact	An unknown guid is404.

RecallInPipelineTests

Jube.Test.Exhaustive.RecallInPipelineTests — 22 tests. Category: Unit.

Test	Kind	Verifies
AbsentValue_WithoutStandardisation_FallsBackToTheRawMeanForEveryProcessingType	Theory	Absent value without standardisation falls back to the raw mean for every processing type.
AbsentValue_WithStandardisation_ImputesTheMeanAndStandardisesToExactlyZero	Theory	Absent value with standardisation imputes the mean and standardises to exactly zero.
AValueHeldInADifferentSourceThanTheProcessingTypeIsNotSeenAndTheMeanIsImputed	Theory	A value held in a different source than the processing type is not seen and the mean is imputed.
APresentZeroIsUsedAsZeroAndNotTreatedAsAbsent	Theory	A present zero is used as zero and not treated as absent.
PresentValue_WithStandardisation_IsStandardisedAgainstTheStoredMeanAndSd	Theory	Present value with standardisation is standardised against the stored mean and sd.
PresentValue_WithConstantColumnSd_StandardisesToZeroRatherThanNonFinite	Fact	Present value with constant column sd standardises to zero rather than non finite.
InputVectorFollowsTheStoredVariableOrderAndNotTheNameOrder	Fact	Input vector follows the stored variable order and not the name order.
EachVariableIsResolvedFromItsOwnProcessingTypeInOneVector	Fact	Each variable is resolved from its own processing type in one vector.
OnlyTheSecondDotSegmentIsUsedAsTheLookupKey	Theory	Only the second dot segment is used as the lookup key.
ADottedNameIsNotFoundUnderItsFullPrefixedKey	Fact	A dotted name is not found under its full prefixed key.
LookupIsCaseSensitive	Fact	Lookup is case sensitive.
ANetworkThatThrowsOmitsTheEntryAndDoesNotRaise	Fact	A network that throws omits the entry and does not raise.
AModelWithoutANetworkOmitsTheEntryButStillRecordsTiming	Fact	A model without a network omits the entry but still records timing.
AFailingModelDoesNotStopTheNextOneFromBeingRecalled	Fact	A failing model does not stop the next one from being recalled.
ANonFiniteScoreIsStoredUnchangedWithoutAnyFinitenessCheck	Fact	A non finite score is stored unchanged without any finiteness check.
TheScoreIsNotRoundedInThePipeline	Fact	The score is not rounded in the pipeline.
ThePayloadValueIsReadAsADoubleWhenTheStoredValueIsAnInteger	Fact	The payload value is read as a double when the stored value is an integer.
ModelsAreKeyedByNameAndEachScoredIndependently	Fact	Models are keyed by name and each scored independently.
ExhaustedBudget_SkipsEveryModelAndLeavesNoScore	Fact	Exhausted budget skips every model and leaves no score.
GenerousBudget_ScoresEveryModel	Fact	Generous budget scores every model.
ExhaustedBudget_DoesNotRecordTimingForSkippedModels	Fact	Exhausted budget does not record timing for skipped models.
WithoutEnvironment_TheBudgetIsTheDefault	Fact	Without environment the budget is the default.

RecallPermissionTests

Jube.Test.Exhaustive.RecallPermissionTests — 4 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
ThePermissionToUseExhaustiveAuthorisesRecallWithoutAModelApiKeyLinkAsync	Fact	The permission to use exhaustive authorises recall without a model api key link.
AnyOtherPermissionDoesNotAuthoriseRecallAsync	Theory	Any other permission does not authorise recall.
TheTenantOfTheModelStillBoundsRecallForAHolderOfThePermissionAsync	Fact	The tenant of the model still bounds recall for a holder of the permission.
ARevokedPermissionStopsAuthorisingRecallAsync	Fact	A revoked permission stops authorising recall.

RecallScoreTests

Jube.Test.Exhaustive.RecallScoreTests — 19 tests. Category: Unit.

Test	Kind	Verifies
Score_AppliesTheNetworkToTheStandardisedInputsInVariableOrder	Fact	Score applies the network to the standardised inputs in variable order.
Score_ReturnsTheFirstOutputOnly	Fact	Score returns the first output only.
Score_WithAllValuesMissing_UsesTheImputedMeans	Fact	Score with all values missing uses the imputed means.
Score_WhenTheNetworkCannotBeComputed_Throws	Fact	Score when the network cannot be computed throws.
Score_WithANullNetwork_Throws	Fact	Score with a null network throws.
Score_WithFewerVariablesThanNetworkInputs_Throws	Fact	Score with fewer variables than network inputs throws.
BuildScoreInputs_ConvertsEveryAcceptedTokenShape	Theory	Build score inputs converts every accepted token shape.
BuildScoreInputs_UnconvertibleTokenIsMissingAndImputesTheMean	Theory	Build score inputs unconvertible token is missing and imputes the mean.
BuildScoreInputs_NestedObjectsAreNotSearched	Fact	Build score inputs nested objects are not searched.
BuildScoreInputs_TraceReportsImputationAndConversionFailure	Fact	Build score inputs trace reports imputation and conversion failure.
BuildScoreInputs_WithNoVariables_ReturnsAnEmptyVector	Fact	Build score inputs with no variables returns an empty vector.
BuildScoreInputs_StringNumbersAreReadInvariantlyWhateverTheCurrentCulture	Fact	Build score inputs string numbers are read invariantly whatever the current culture.
RecallExhaustive_ForAKnownGuid_ReturnsTheScore	Fact	Recall exhaustive for a known guid returns the score.
RecallExhaustive_ForAnUnknownGuid_ReturnsZeroAndLogsTheError	Fact	Recall exhaustive for an unknown guid returns zero and logs the error.
RecallExhaustive_WhenTheNetworkThrows_SwallowsTheExceptionAndReturnsZero	Fact	Recall exhaustive when the network throws swallows the exception and returns zero.
RecallExhaustive_WhenTheNetworkIsNull_ReturnsZero	Fact	Recall exhaustive when the network is null returns zero.
RecallExhaustive_ANonFiniteScoreIsReturnedUnchanged	Fact	Recall exhaustive a non finite score is returned unchanged.
RecallExhaustive_FindsTheInstanceAcrossModels	Fact	Recall exhaustive finds the instance across models.
RecallBodyHasValidTypes_Boundaries	Theory	Recall body has valid types boundaries.

RecallTimelineTests

Jube.Test.Exhaustive.RecallTimelineTests — 1 test. Category: Service. Collection: Database.

Test	Kind	Verifies
PromotedModelBecomesRecallableWhileTheEngineIsRunningAsync	Fact	Promoted model becomes recallable while the engine is running.

SupervisedRepairTests

Jube.Test.Exhaustive.SupervisedRepairTests — 9 tests.

Test	Kind	Verifies
SplitData_TrainCrossValidationAndTest_AreDisjointAndAligned	Theory	Split data train cross validation and test are disjoint and aligned.
ReduceDataForOnlyTrialVariables_PopulatesEveryTrialColumnInOrder	Fact	Reduce data for only trial variables populates every trial column in order.
DrawVariableIndices_AreDistinctAndInRange	Theory	Draw variable indices are distinct and in range.
DrawVariableIndices_MinimumSubset_CanReachTheLastVariable	Fact	Draw variable indices minimum subset can reach the last variable.
LeastSensitiveVariableKey_EliminationRemovesLeastSensitiveEveryTimeAndNeverRunsOffTheEnd	Fact	Least sensitive variable key elimination removes least sensitive every time and never runs off the end.
Sensitivity_PerturbsByAFactorOfEightyToOneTwentyPercent_NotBySquaringTheValue	Fact	Sensitivity perturbs by a factor of eighty to one twenty percent not by squaring the value.
InterquartileRange_IsNonNegativeAndInOriginalUnits	Fact	Interquartile range is non negative and in original units.
TryGrowLayers_WidensUntilWideThenAddsALayerUntilTheDepthLimit	Fact	Try grow layers widens until wide then adds a layer until the depth limit.
Performance_Score_RewardsGoodModelsAndDoesNotRewardInvertedOnes	Fact	Performance score rewards good models and does not reward inverted ones.

SyncBuildModelTests

Jube.Test.Exhaustive.SyncBuildModelTests — 12 tests. Category: Unit.

Test	Kind	Verifies
IdentityAndGuidComeFromTheRecord	Fact	Identity and guid come from the record.
SpacesInTheNameBecomeUnderscores	Fact	Spaces in the name become underscores.
ANullNameDefaultsToExhaustiveUnderscoreId	Fact	A null name defaults to exhaustive underscore id.
ResponsePayloadAndReportTableAreTrueOnlyForOne	Theory	Response payload and report table are true only for one.
TheNetworkJsonIsDeserialisedToAnEquivalentNetwork	Fact	The network json is deserialised to an equivalent network.
VariablesKeepTheOrderSuppliedAndCarryTheStatisticsPerVariable	Fact	Variables keep the order supplied and carry the statistics per variable.
TheLoadedVariablesStandardiseWithTheStoredStatistics	Fact	The loaded variables standardise with the stored statistics.
NoVariableRowsLeavesTheNetworkLoadedWithAnEmptyVariableList	Fact	No variable rows leaves the network loaded with an empty variable list.
ANullJsonReportsTheLoadErrorAndStillReturnsTheModelWithoutNetworkOrVariables	Fact	A null json reports the load error and still returns the model without network or variables.
AnEmptyJsonStringGivesANullNetworkAndNoError	Fact	An empty json string gives a null network and no error.
GarbageJsonNeverEscapesAsAnException	Fact	Garbage json never escapes as an exception.
AnOnErrorCallbackIsOptional	Fact	An on error callback is optional.

SyncExhaustiveSearchInstancesTests

Jube.Test.Exhaustive.SyncExhaustiveSearchInstancesTests — 11 tests. Category: Service. Collection: Database.

Test	Kind	Verifies
LoadsTheNetworkVariablesAndNameOfAPromotedActiveInstanceAsync	Fact	Loads the network variables and name of a promoted active instance.
VariablesAreOrderedByVariableSequenceNotInsertionOrderAsync	Fact	Variables are ordered by variable sequence not insertion order.
RemovedVariablesAreExcludedAndOnlyThePromotedTrialsVariablesAreLoadedAsync	Fact	Removed variables are excluded and only the promoted trials variables are loaded.
StatisticsAndProcessingTypeAreLoadedPerVariableAsync	Fact	Statistics and processing type are loaded per variable.
TheLastActivePromotedTrialWinsAndInactiveOnesAreIgnoredAsync	Fact	The last active promoted trial wins and inactive ones are ignored.
AnInstanceWithoutAnActivePromotedTrialIsNotRegisteredAsync	Fact	An instance without an active promoted trial is not registered.
InactiveAndDeletedInstancesAreSkippedAndOthersStillLoadAsync	Fact	Inactive and deleted instances are skipped and others still load.
ANullNameDefaultsToExhaustiveUnderscoreIdAsync	Fact	A null name defaults to exhaustive underscore id.
InstancesAreRegisteredInIdOrderAsync	Fact	Instances are registered in id order.
ACorruptPromotedJsonDoesNotStopTheNextInstanceLoadingAsync	Fact	A corrupt promoted json does not stop the next instance loading.
ResyncReplacesTheCollectionAndPicksUpANewlyPromotedTrialAsync	Fact	Resync replaces the collection and picks up a newly promoted trial.

TrainClassHistogramEndToEndTests

Jube.Test.Exhaustive.TrainClassHistogramEndToEndTests — 2 tests. Collection: Database.

Test	Kind	Verifies
ClassHistogramTableRepositoryRoundTripsSoftDeletesAndIsTenantScopedAsync	Fact	Class histogram table repository round trips soft deletes and is tenant scoped.
MigrationCreatesTheClassHistogramTableWithEveryColumnAsync	Fact	Migration creates the class histogram table with every column.

TrainClassHistogramTests

Jube.Test.Exhaustive.TrainClassHistogramTests — 10 tests.

Test	Kind	Verifies
HandCheckedTenValuesSplitOnTheGivenPopulationEdges	Fact	Hand checked ten values split on the given population edges.
BoundaryValuesGoToTheUpperBinAndTheMaximumToTheLastBin	Fact	Boundary values go to the upper bin and the maximum to the last bin.
EveryRowLandsInExactlyOneBinSoClassCountsSumToTheRowCount	Fact	Every row lands in exactly one bin so class counts sum to the row count.
IsDeterministicForTheSameInput	Fact	Is deterministic for the same input.
FilterModeKeepsThePopulationAsTheNegativeClassAndClampsPositivesOutsideTheEdges	Fact	Filter mode keeps the population as the negative class and clamps positives outside the edges.
ConstantColumnIsOneBinAndEveryRowLandsInIt	Fact	Constant column is one bin and every row lands in it.
NoBinsYieldsNothingAndNaNValuesFallInTheFirstBin	Fact	No bins yields nothing and na n values fall in the first bin.
WriterInsertsNegativeThenPositiveBinsWithTheirLabelsAsync	Fact	Writer inserts negative then positive bins with their labels.
WriteAllProcessesEveryVariableWithItsOwnColumnAndItsOwnStoredEdgesAsync	Fact	Write all processes every variable with its own column and its own stored edges.
WriteAllSkipsAVariableWhoseStoredPopulationHistogramIsEmptyAsync	Fact	Write all skips a variable whose stored population histogram is empty.

TrainDbRecorderTests

Jube.Test.Exhaustive.TrainDbRecorderTests — 8 tests. Collection: Database.

Test	Kind	Verifies
IsStopRequested_IsTrueOnlyForStoppingAndStoppedStatusesAsync	Theory	Is stop requested is true only for stopping and stopped statuses.
InstanceUpdates_PersistScoreModelsAndModelsSinceBestAsync	Fact	Instance updates persist score models and models since best.
Recorder_InsertsEveryTrialArtefactAndReturnsTheGeneratedIdsAsync	Fact	Recorder inserts every trial artefact and returns the generated ids.
TrialVariables_InsertThenMarkRemovedAsync	Fact	Trial variables insert then mark removed.
MarkTrialVariableRemoved_ForAnUnknownPair_ThrowsAsync	Fact	Mark trial variable removed for an unknown pair throws.
DeleteTrialVariables_SoftDeletesTheRowsOfThatTrialOnlyAsync	Fact	Delete trial variables soft deletes the rows of that trial only.
PromotedVariableHistogram_IsInsertedAgainstThePromotedVariableAsync	Fact	Promoted variable histogram is inserted against the promoted variable.
PromotedVariableHistogram_ForAnUnknownPromotedVariable_ViolatesTheForeignKeyAsync	Fact	Promoted variable histogram for an unknown promoted variable violates the foreign key.

TrainInstanceRunTests

Jube.Test.Exhaustive.TrainInstanceRunTests — 13 tests. Collection: Database.

Test	Kind	Verifies
RunOneInstance_StopRequested_MovesStraightToStatusNineteenAndReturnsFalseAsync	Fact	Run one instance stop requested moves straight to status nineteen and returns false.
RunOneInstance_AlreadyStopped_StaysAtStatusNineteenAsync	Fact	Run one instance already stopped stays at status nineteen.
RunOneInstance_CancelledToken_ThrowsBeforeTouchingTheStatusAsync	Fact	Run one instance cancelled token throws before touching the status.
RunOneInstance_AnomalyMockData_RunsEveryStageToCompletionAtStatusSixteenAsync	Fact	Run one instance anomaly mock data runs every stage to completion at status sixteen.
RunOneInstance_AnomalyMockData_StoresClassHistogramsOnThePopulationBinEdgesAsync	Fact	Run one instance anomaly mock data stores class histograms on the population bin edges.
RunOneInstance_TheDefaultAlgorithmRecordsAnSvmReportWithNoReductionAsync	Fact	Run one instance the default algorithm records an svm report with no reduction.
RunOneInstance_ANativeAlgorithmLabelsTheContaminationFractionAndStoresTheReductionAsync	Fact	Run one instance a native algorithm labels the contamination fraction and stores the reduction.
RunOneInstance_AutomaticRecordsEveryCandidateEvaluationAsync	Fact	Run one instance automatic records every candidate evaluation.
RunOneInstance_AnomalyThatFlagsNothing_CompletesWithStatusSeventeenAndNoTrialsAsync	Fact	Run one instance anomaly that flags nothing completes with status seventeen and no trials.
StartAsync_PollsRunsTheNextInstanceThenKeepsPollingUntilCancelledAsync	Fact	Start async polls runs the next instance then keeps polling until cancelled.
StartAsync_AnAnomalyInstanceThatFlagsNothing_CompletesAtStatusSeventeenAndPollingContinuesAsync	Fact	Start async an anomaly instance that flags nothing completes at status seventeen and polling continues.
StartAsync_ARunThatThrows_IsSurvivedAndPollingContinuesAsync	Fact	Start async a run that throws is survived and polling continues.
StartAsync_CancelledBeforeStart_ReturnsImmediatelyWithoutPollingAsync	Fact	Start async cancelled before start returns immediately without polling.

TrainLoopTests

Jube.Test.Exhaustive.TrainLoopTests — 27 tests.

Test	Kind	Verifies
TrainAsync_RunsTrialsLimitMinusOneTrials_AndReturnsTrueAsync	Fact	Train async runs trials limit minus one trials and returns true.
TrainAsync_TrialsLimitOfOne_RunsNoTrialsAtAllAsync	Fact	Train async trials limit of one runs no trials at all.
TrainAsync_RecordsEveryTrialInStageOrderAsync	Fact	Train async records every trial in stage order.
TrainAsync_ActivationSearch_RecordsTenCandidatesPerVariableEliminationIterationAsync	Fact	Train async activation search records ten candidates per variable elimination iteration.
TrainAsync_TopologyTrials_GrowOneNeuronAtATimeAndEndWithExactlyOneFinalisationRowAsync	Fact	Train async topology trials grow one neuron at a time and end with exactly one finalisation row.
TrainAsync_TopologySinceImprovementLimit_BoundsTheTopologySearchAsync	Fact	Train async topology since improvement limit bounds the topology search.
TrainAsync_Promotion_IsStrictlyIncreasingAndStartsFromZeroAsync	Fact	Train async promotion is strictly increasing and starts from zero.
TrainAsync_NonPromotedTrials_IncrementModelsSinceBest_AndPromotionDoesNotReportItAsync	Fact	Train async non promoted trials increment models since best and promotion does not report it.
TrainAsync_PerPromotion_WritesTwentyRocRowsWithTheFixedThresholdGridAsync	Fact	Train async per promotion writes twenty roc rows with the fixed threshold grid.
TrainAsync_PerPromotion_WritesPredictedVersusActualForEveryTestRowAsync	Fact	Train async per promotion writes predicted versus actual for every test row.
TrainAsync_PerPromotion_WritesOnePromotedSensitivityPerVariableWithDenseRanksAsync	Fact	Train async per promotion writes one promoted sensitivity per variable with dense ranks.
TrainAsync_PerPromotion_PersistsPrescriptionHistogramsLinkedToTheInsertedPrescriptionAsync	Fact	Train async per promotion persists prescription histograms linked to the inserted prescription.
TrainAsync_PerPromotion_WritesTheModelJsonThatReproducesTheRecordedScoresAsync	Fact	Train async per promotion writes the model json that reproduces the recorded scores.
TrainAsync_PromotedConfusion_IsRecordedAtTheHalfThresholdAsync	Fact	Train async promoted confusion is recorded at the half threshold.
TrainAsync_RemovedVariables_AreOnlyVariablesThatWereSelectedInThatTrialAsync	Fact	Train async removed variables are only variables that were selected in that trial.
TrainAsync_VariableSelection_ClearsThePreviousSelectionBeforeEveryDrawAsync	Fact	Train async variable selection clears the previous selection before every draw.
TrainAsync_SelectedVariableCount_RespectsMinimumAndMaximumAsync	Fact	Train async selected variable count respects minimum and maximum.
TrainAsync_MinimumAboveAvailableVariables_KeepsEveryVariableAndCompletesAsync	Fact	Train async minimum above available variables keeps every variable and completes.
TrainAsync_DrawEqualToTheMinimum_RemovesNothingAndCompletesAsync	Fact	Train async draw equal to the minimum removes nothing and completes.
TrainAsync_Elimination_NeverLeavesFewerThanTheMinimumVariablesAsync	Fact	Train async elimination never leaves fewer than the minimum variables.
TrainAsync_StopRequestedBeforeTheFirstTrial_ReturnsFalseAndRecordsNothingAsync	Fact	Train async stop requested before the first trial returns false and records nothing.
TrainAsync_StopRequestedBetweenTrials_FinishesTheRunningTrialThenStopsAsync	Fact	Train async stop requested between trials finishes the running trial then stops.
TrainAsync_CancellationToken_PropagatesAsOperationCanceledNotSwallowedPerTrialAsync	Fact	Train async cancellation token propagates as operation canceled not swallowed per trial.
TrainAsync_CancellationMidTrial_PropagatesFromTheRecorderAsync	Fact	Train async cancellation mid trial propagates from the recorder.
TrainAsync_AnErrorInsideOneTrial_IsSwallowedAndTheNextTrialStillRunsAsync	Fact	Train async an error inside one trial is swallowed and the next trial still runs.
TrainAsync_AnErrorDuringPromotionStorage_SkipsTheRestOfThatPromotionButKeepsTheBestScoreAsync	Fact	Train async an error during promotion storage skips the rest of that promotion but keeps the best score.
TrainAsync_SeparableData_ReachesAPerfectScoreAndStopsBeforeTheTrialLimitAsync	Fact	Train async separable data reaches a perfect score and stops before the trial limit.

TrainMetricsTests

Jube.Test.Exhaustive.TrainMetricsTests — 34 tests.

Test	Kind	Verifies
CalculatePerformance_CountsTheConfusionMatrixAtTheThreshold	Fact	Calculate performance counts the confusion matrix at the threshold.
CalculatePerformance_ThresholdIsStrictlyGreaterThan	Fact	Calculate performance threshold is strictly greater than.
CalculatePerformance_ScoreIsTheMeanOfAccuracyAndSpearmanOfPredictedVersusActual	Fact	Calculate performance score is the mean of accuracy and spearman of predicted versus actual.
CalculatePerformance_InvertedClassifier_ScoresMinusOneHalf_WithTiedBinaryRanks	Fact	Calculate performance inverted classifier scores minus one half with tied binary ranks.
CalculatePerformance_PerfectClassifier_ScoresOne	Fact	Calculate performance perfect classifier scores one.
CalculatePerformance_ASingleRow_HasNaNCorrelationTreatedAsZero	Fact	Calculate performance a single row has na n correlation treated as zero.
CalculatePerformance_NoRows_ScoresNaN_Characterisation	Fact	Calculate performance no rows scores na n characterisation.
CalculatePerformance_NaNScoreIsPredictedNegative_AndInfinityHonoursTheThreshold	Fact	Calculate performance na n score is predicted negative and infinity honours the threshold.
CalculatePerformance_ResetsCountersOnEveryCall	Fact	Calculate performance resets counters on every call.
CalculateScores_ComputesTheFirstOutputNeuronForEveryRow	Fact	Calculate scores computes the first output neuron for every row.
RocPoints_IsTwentyThresholdsFromFiveHundredthsToOne	Fact	Roc points is twenty thresholds from five hundredths to one.
RocPoints_MatchesAnIndependentSweepAtEveryThreshold	Fact	Roc points matches an independent sweep at every threshold.
RocPoints_AtTheHalfThreshold_AgreesWithCalculatePerformance	Fact	Roc points at the half threshold agrees with calculate performance.
RocPoints_ScoresAllAboveOne_PredictEveryRowPositiveEvenAtTheLastThreshold	Fact	Roc points scores all above one predict every row positive even at the last threshold.
RocPoints_DoesNotDisturbAnExistingPerformanceInstance	Fact	Roc points does not disturb an existing performance instance.
Spearman_PerfectAgreementAndDisagreement	Fact	Spearman perfect agreement and disagreement.
Spearman_IsRankBased_SoAMonotoneNonLinearRelationIsOne	Fact	Spearman is rank based so a monotone non linear relation is one.
Spearman_KnownValueWithoutTies	Fact	Spearman known value without ties.
Spearman_WithTies_IsThePearsonCorrelationOfAverageRanks	Theory	Spearman with ties is the pearson correlation of average ranks.
Spearman_IsSymmetric	Fact	Spearman is symmetric.
Spearman_LengthMismatchThrows	Fact	Spearman length mismatch throws.
Spearman_AConstantSeriesHasNoCorrelation_AndSingleElementAndEmptyAreNaN	Fact	Spearman a constant series has no correlation and single element and empty are na n.
Spearman_ConstantAgainstVarying_IsFiniteNotNaN	Fact	Spearman constant against varying is finite not na n.
Sensitivity_DoesNotMutateTheInputRows_AndIsRepeatableForTheSameSeed	Fact	Sensitivity does not mutate the input rows and is repeatable for the same seed.
Sensitivity_AVariableTheModelIgnoresHasZeroSensitivity_AndAZeroValueCannotBePerturbed	Fact	Sensitivity a variable the model ignores has zero sensitivity and a zero value cannot be perturbed.
BuildPrescriptions_KeepsOnlyActivationsAboveHalfAndEmitsOneRowPerVariable	Fact	Build prescriptions keeps only activations above half and emits one row per variable.
BuildPrescriptionsWithHistograms_HistogramCoversTheSampleAndConstantVariablesGetNone	Fact	Build prescriptions with histograms histogram covers the sample and constant variables get none.
BuildPrescriptions_NetworkThatNeverActivatesPositively_ReturnsOnlyTheNegativeClassPrescription	Fact	Build prescriptions network that never activates positively returns only the negative class prescription.
BuildPrescriptions_SplitsSimulationsIntoAPositiveAndANegativeClassPrescriptionPerVariable	Fact	Build prescriptions splits simulations into a positive and a negative class prescription per variable.
BuildPrescriptions_BinaryVariablesAreSnappedToZeroOrOne_AndMissingDistributionsBecomeZero	Fact	Build prescriptions binary variables are snapped to zero or one and missing distributions become zero.
GetRandomVariableCount_StaysWithinMinimumAndMaximumAndBelowTheAvailableCount	Theory	Get random variable count stays within minimum and maximum and below the available count.
GetRandomVariableCount_FewerAvailableThanTheMinimum_ReturnsEveryAvailableVariable	Fact	Get random variable count fewer available than the minimum returns every available variable.
Unsupervised_OneClassModel_ScoresAnOutlierLowerThanTheClusterAndTheClassifierFlagsIt	Fact	Unsupervised one class model scores an outlier lower than the cluster and the classifier flags it.
GetClassVariableByAnomaly_IsStrictlyBelowTheProbability	Fact	Get class variable by anomaly is strictly below the probability.

TrainRecallExhaustiveTests

Jube.Test.Exhaustive.TrainRecallExhaustiveTests — 10 tests.

Test	Kind	Verifies
RecallExhaustive_ScoresTheMatchingInstanceWithZScoredInputsInVariableOrder	Fact	Recall exhaustive scores the matching instance with z scored inputs in variable order.
RecallExhaustive_ChoosesTheInstanceByGuidAcrossModels	Fact	Recall exhaustive chooses the instance by guid across models.
RecallExhaustive_AMissingOrNullOrUnparseableValue_IsImputedWithTheTrainingMean	Fact	Recall exhaustive a missing or null or unparseable value is imputed with the training mean.
RecallExhaustive_UnknownGuid_ReturnsZeroBecauseTheNotFoundErrorIsSwallowed	Fact	Recall exhaustive unknown guid returns zero because the not found error is swallowed.
RecallExhaustive_NoActiveModels_ReturnsZero	Fact	Recall exhaustive no active models returns zero.
RecallExhaustive_ANetworkWhoseInputCountDisagreesWithTheVariables_ReturnsZeroNotAnException	Fact	Recall exhaustive a network whose input count disagrees with the variables returns zero not an exception.
RecallExhaustive_AnInstanceWithNoNetwork_ReturnsZero	Fact	Recall exhaustive an instance with no network returns zero.
RecallExhaustiveWithExplainability_MatchesTheScoreOfRecallExhaustive	Fact	Recall exhaustive with explainability matches the score of recall exhaustive.
RecallExhaustiveWithExplainability_LrpRelevancePlusBaseValueReconstructsTheRawOutput	Fact	Recall exhaustive with explainability lrp relevance plus base value reconstructs the raw output.
RecallExhaustiveWithExplainability_UnknownGuid_ReturnsZeroAndNoRelevance	Fact	Recall exhaustive with explainability unknown guid returns zero and no relevance.

TrainShuffleSymmetryTests

Jube.Test.Exhaustive.TrainShuffleSymmetryTests — 20 tests.

Test	Kind	Verifies
Shuffle_IsAPermutationThatKeepsEveryRowPairedWithItsLabel	Fact	Shuffle is a permutation that keeps every row paired with its label.
Shuffle_SameSeedGivesTheSameOrder_DifferentSeedsGiveDifferentOrders	Fact	Shuffle same seed gives the same order different seeds give different orders.
Shuffle_DoesNotMutateItsInputs_AndReusesTheRowInstances	Fact	Shuffle does not mutate its inputs and reuses the row instances.
Shuffle_EmptyAndSingleRow_AreReturnedUnchanged	Theory	Shuffle empty and single row are returned unchanged.
Shuffle_IsUniformEnoughThatEveryRowVisitsEveryPositionOverManySeeds	Fact	Shuffle is uniform enough that every row visits every position over many seeds.
Shuffle_DoesNotPreserveClassStratification_ClassRatioOfTheFirstHalfVaries	Fact	Shuffle does not preserve class stratification class ratio of the first half varies.
DatasetSymmetry_RemovesOnlyMajorityRows_KeepsOrder_AndIsSeedDeterministic	Fact	Dataset symmetry removes only majority rows keeps order and is seed deterministic.
DatasetSymmetry_MajorityPositive_RemovesPositivesInstead	Fact	Dataset symmetry majority positive removes positives instead.
DatasetSymmetry_SingleClass_RemovesEveryRow_Characterisation	Theory	Dataset symmetry single class removes every row characterisation.
DatasetSymmetry_DoesNotMutateItsInputs_AndKeepsTheRowInstances	Fact	Dataset symmetry does not mutate its inputs and keeps the row instances.
DatasetSymmetry_DifferentSeedsPickDifferentSurvivors	Fact	Dataset symmetry different seeds pick different survivors.
DatasetSymmetry_ExplicitTargetBelowBothClasses_TrimsBothClassesToTheTarget	Fact	Dataset symmetry explicit target below both classes trims both classes to the target.
DatasetSymmetry_ExplicitTargetAboveMinority_TrimsOnlyTheMajority_LeavingMinorityForBootstrapToTopUp	Fact	Dataset symmetry explicit target above minority trims only the majority leaving minority for bootstrap to top up.
DatasetSymmetry_ExplicitTargetAboveBothClasses_LeavesBothClassesUntouched	Fact	Dataset symmetry explicit target above both classes leaves both classes untouched.
NormaliseData_OnlyZScoresTypeTwoColumns_AndMutatesInPlace	Fact	Normalise data only z scores type two columns and mutates in place.
NormaliseData_UsesTheFittedStatisticsForEveryRow_SoARowIsNotAffectedByOthers	Fact	Normalise data uses the fitted statistics for every row so a row is not affected by others.
NormaliseData_MatrixWithFewerColumnsThanVariables_Throws	Fact	Normalise data matrix with fewer columns than variables throws.
NormaliseData_FitOnPopulationIsAppliedToAllRows_LeakageOfTestRowsIntoTheStatistics_H6Async	Fact	Normalise data fit on population is applied to all rows leakage of test rows into the statistics h 6.
OnlyOutputWithClassificationForData_ReturnsThePositiveRowsInOrder_AndThrowsIfTheCountIsTooLow	Fact	Only output with classification for data returns the positive rows in order and throws if the count is too low.
OnlyOutputWithClassificationForData_CountTooHigh_LeavesNullRows_Characterisation	Fact	Only output with classification for data count too high leaves null rows characterisation.

TrainStatisticsTests

Jube.Test.Exhaustive.TrainStatisticsTests — 19 tests.

Test	Kind	Verifies
Population_ComputesMeanMinMaxAndSampleStandardDeviationAsync	Fact	Population computes mean min max and sample standard deviation.
Population_AssignsNormalisationTypeByShapeOfTheColumnAsync	Fact	Population assigns normalisation type by shape of the column.
Population_ConstantNonZeroColumn_IsZScoreTypeWithZeroSd_AndTheStatisticsRowIsStillWrittenAsync	Fact	Population constant non zero column is z score type with zero sd and the statistics row is still written.
Population_TriangularDistribution_IsBuiltInNormalisedSpaceOnlyWhenMinBelowMaxAsync	Fact	Population triangular distribution is built in normalised space only when min below max.
Population_ModeZeroAndMeanNonZero_UsesAFixedSmallModeOfPointZeroOneAsync	Fact	Population mode zero and mean non zero uses a fixed small mode of point zero one.
Population_ColumnWithOnlyNonPositiveValuesAndAZeroMode_ClampsTheModeIntoTheRangeAsync	Fact	Population column with only non positive values and a zero mode clamps the mode into the range.
Population_ColumnWithMaximumZeroAndNegativeValues_IsTypedAsNoNormalisationLikeAnAllZeroColumn_CharacterisationAsync	Fact	Population column with maximum zero and negative values is typed as no normalisation like an all zero column characterisation.
Population_ConstantColumn_StoresZeroKurtosisAndSkewnessInsteadOfNaNAsync	Fact	Population constant column stores zero kurtosis and skewness instead of na n.
Population_ColumnContainingNaN_StoresZeroedStatisticsAndHasNoDistributionAsync	Fact	Population column containing na n stores zeroed statistics and has no distribution.
BuildTriangularDistribution_ForADegenerateRange_ReturnsNull	Theory	Build triangular distribution for a degenerate range returns null.
BuildTriangularDistribution_ClampsTheModeIntoTheRange	Theory	Build triangular distribution clamps the mode into the range.
Population_StoredBinCount_FollowsTheSturgesLikeRuleAndCapsAtTwentyAsync	Fact	Population stored bin count follows the sturges like rule and caps at twenty.
Population_HistogramRows_AreWrittenPerBinInSequenceForTheInsertedVariableIdAsync	Fact	Population histogram rows are written per bin in sequence for the inserted variable id.
Population_HistogramBinCount_IgnoresTheComputedBinCount_H17Async	Fact	Population histogram bin count ignores the computed bin count h 17.
Classification_WritesClassificationRowsLinkedToTheExistingVariableIds_AndLeavesVariablesAloneAsync	Fact	Classification writes classification rows linked to the existing variable ids and leaves variables alone.
Anomaly_WritesAnomalyRowsLinkedToTheExistingVariableIdsAsync	Fact	Anomaly writes anomaly rows linked to the existing variable ids.
TheOriginalSignature_ResolvesTheTargetFromTheRepositoryTypeAsync	Fact	The original signature resolves the target from the repository type.
OnlyTheFirstVariablesCountRowsAreCovered_WhenTheMatrixHasMoreColumnsAsync	Fact	Only the first variables count rows are covered when the matrix has more columns.
MockPopulation_ClassifiesEveryColumnAndBuildsADistributionForEveryNonConstantColumnAsync	Fact	Mock population classifies every column and builds a distribution for every non constant column.

TrainingRepairTests

Jube.Test.Exhaustive.TrainingRepairTests — 6 tests.

Test	Kind	Verifies
NormaliseData_NormalisesEveryRowIncludingTheLast	Fact	Normalise data normalises every row including the last.
NormaliseData_ConstantNonBinaryColumn_ProducesZeroNotNaN	Fact	Normalise data constant non binary column produces zero not na n.
DatasetSymmetry_BalancesClassesExactlyAndCanKeepTheLastRow	Fact	Dataset symmetry balances classes exactly and can keep the last row.
DatasetSymmetry_Balanced_ReturnsEveryRow	Fact	Dataset symmetry balanced returns every row.
OnlyOutputWithClassificationForData_KeepsAPositiveFinalRow	Fact	Only output with classification for data keeps a positive final row.
CalculateStatistics_ProducesStatisticsForEveryVariableIncludingTheLastAsync	Fact	Calculate statistics produces statistics for every variable including the last.

Datasets

DatasetCategoricalSupportTests

Jube.Test.Exhaustive.Datasets.DatasetCategoricalSupportTests — 4 tests.

Test	Kind	Verifies
ColumnWithTwoWellPopulatedLevels_IsFittedAsCategoricalButExcludedByDefault	Fact	Column with two well populated levels is fitted as categorical but excluded by default.
IdentifierLikeColumn_IsExcludedWithAReason	Fact	Identifier like column is excluded with a reason.
DistinctValues_AreCountedAndSortedByFrequency	Fact	Distinct values are counted and sorted by frequency.
RawDistinctTracking_TruncatesBeyondTheBudget	Fact	Raw distinct tracking truncates beyond the budget.

DatasetCsvParserTests

Jube.Test.Exhaustive.Datasets.DatasetCsvParserTests — 33 tests.

Test	Kind	Verifies
SimpleFile_IsParsedWithProfilesAsync	Fact	Simple file is parsed with profiles.
Bom_IsSkippedAndHeaderIsCleanAsync	Fact	Bom is skipped and header is clean.
Delimiter_IsDetectedAsync	Theory	Delimiter is detected.
SemicolonHeaderWithQuotedCommas_UsesSemicolonAsync	Fact	Semicolon header with quoted commas uses semicolon.
MixedLineEndings_AreAcceptedAsync	Fact	Mixed line endings are accepted.
Quoting_EscapedQuotesAndEmbeddedDelimitersAndNewlinesAsync	Fact	Quoting escaped quotes and embedded delimiters and newlines.
MalformedQuoting_IsRejectedAsync	Theory	Malformed quoting is rejected.
InjectionCells_AreStoredAsDataNeverEvaluatedAsync	Fact	Injection cells are stored as data never evaluated.
MissingTokens_AreCountedAsync	Fact	Missing tokens are counted.
Numbers_UseInvariantCultureAndRejectThousandsSeparatorsAsync	Fact	Numbers use invariant culture and reject thousands separators.
NonFiniteNumbers_AreNotAcceptedAsNumbersAsync	Fact	Non finite numbers are not accepted as numbers.
TextColumn_TooFewRowsToFitAsCategorical_IsExcludedByDefaultAsync	Fact	Text column too few rows to fit as categorical is excluded by default.
ConstantAndEmptyColumns_AreExcludedByDefaultAsync	Fact	Constant and empty columns are excluded by default.
RaggedRow_IsRejectedWithRowNumberAsync	Fact	Ragged row is rejected with row number.
RaggedRow_TooManyValues_IsRejectedAsync	Fact	Ragged row too many values is rejected.
DuplicateHeaders_AreRejectedAsync	Theory	Duplicate headers are rejected.
BadHeaders_AreRejectedAsync	Theory	Bad headers are rejected.
HugeCell_IsRejectedAsync	Fact	Huge cell is rejected.
HugeHeader_IsRejectedAsync	Fact	Huge header is rejected.
NonUtf8Bytes_AreRejectedAsync	Fact	Non utf8 bytes are rejected.
Utf16WithBom_IsRejectedAsync	Fact	Utf16 with bom is rejected.
CompressedAndBinaryContent_IsRejectedAsync	Theory	Compressed and binary content is rejected.
ControlCharacters_AreRejectedAsync	Fact	Control characters are rejected.
RowCap_IsEnforcedAsync	Fact	Row cap is enforced.
RowCap_AtTheLimitIsAcceptedAsync	Fact	Row cap at the limit is accepted.
ColumnCap_IsEnforcedBeforeAnyRowAsync	Fact	Column cap is enforced before any row.
ByteCap_IsEnforcedWhileStreamingAsync	Fact	Byte cap is enforced while streaming.
EmptyFile_HeaderOnly_AndBlankLines_AreHandledAsync	Fact	Empty file header only and blank lines are handled.
Cancellation_StopsParsingAsync	Fact	Cancellation stops parsing.
LargeStreamedFile_DoesNotBufferRowsAsync	Fact	Large streamed file does not buffer rows.
SuggestName_ProducesSafeNames	Theory	Suggest name produces safe names.
SuggestName_DeduplicatesAndTruncates	Fact	Suggest name deduplicates and truncates.
RowJson_UsesIndexKeysAndNullsAndText	Fact	Row json uses index keys and nulls and text.

DatasetJsonParserTests

Jube.Test.Exhaustive.Datasets.DatasetJsonParserTests — 57 tests.

Test	Kind	Verifies
ArrayOfObjects_IsParsedWithProfilesAsync	Fact	Array of objects is parsed with profiles.
ColumnarForm_IsParsedWithProfilesAsync	Fact	Columnar form is parsed with profiles.
BothShapes_GiveIdenticalReportsAndRowsAsync	Fact	Both shapes give identical reports and rows.
ObjectForm_UnionOfKeysDefinesColumns_AndMissingKeyIsMissingAsync	Fact	Object form union of keys defines columns and missing key is missing.
Booleans_BecomeOneAndZero_WithAWarningAsync	Fact	Booleans become one and zero with a warning.
NullIsMissing_AndCountedAsync	Fact	Null is missing and counted.
Strings_FlagNonNumericColumns_ExcludedByDefault_AndKeepTheTextAsync	Fact	Strings flag non numeric columns excluded by default and keep the text.
MixedTypesInAColumn_AreMixedAndExcludedAsync	Fact	Mixed types in a column are mixed and excluded.
NumbersAsStrings_AreTextAndWarnedAsync	Fact	Numbers as strings are text and warned.
NegativeZero_IsStoredAsZeroAsync	Fact	Negative zero is stored as zero.
IntegersBeyondTwoToThe53_AreWarned_AndExactOnesAreNotAsync	Fact	Integers beyond two to the53 are warned and exact ones are not.
OverflowingNumbers_AreRejectedAsync	Theory	Overflowing numbers are rejected.
LargeButFiniteNumber_IsAcceptedAsync	Fact	Large but finite number is accepted.
UnicodeColumnNames_AreAcceptedAndGetSafeSuggestedNamesAsync	Fact	Unicode column names are accepted and get safe suggested names.
NestedObject_IsRejectedWithAPrecisePathAsync	Fact	Nested object is rejected with a precise path.
NestedArray_InColumnarRows_IsRejectedWithAPrecisePathAsync	Fact	Nested array in columnar rows is rejected with a precise path.
DuplicateKeys_InOneObject_AreRejectedAsync	Fact	Duplicate keys in one object are rejected.
KeysDifferingOnlyByCase_AreRejectedLikeCsvHeadersAsync	Fact	Keys differing only by case are rejected like csv headers.
DuplicateColumnNames_InColumnarForm_AreRejectedAsync	Fact	Duplicate column names in columnar form are rejected.
UnsafeNames_AreRejectedWithTheCsvHeaderRulesAsync	Theory	Unsafe names are rejected with the csv header rules.
NameLongerThanTheHeaderCap_IsRejectedAsync	Fact	Name longer than the header cap is rejected.
TopLevelScalars_AreRejectedAsync	Theory	Top level scalars are rejected.
EmptyArray_IsRejectedAsNoRowsAsync	Fact	Empty array is rejected as no rows.
ColumnarWithNoRows_IsRejectedAsync	Fact	Columnar with no rows is rejected.
EmptyObject_AndEmptyInput_AreRejectedAsync	Fact	Empty object and empty input are rejected.
MalformedJson_IsRejectedAndNothingSucceedsAsync	Theory	Malformed json is rejected and nothing succeeds.
TrailingContent_IsRejectedAsync	Theory	Trailing content is rejected.
TrailingWhitespace_IsAcceptedAsync	Fact	Trailing whitespace is accepted.
CsvContent_WithAJsonName_IsRejectedAsync	Fact	Csv content with a json name is rejected.
ByteOrderMark_IsRejectedAsync	Fact	Byte order mark is rejected.
BinaryAndUtf16_AreRejectedAsync	Theory	Binary and utf16 are rejected.
InvalidUtf8_InAString_IsRejectedAsync	Fact	Invalid utf8 in a string is rejected.
LoneSurrogateEscape_IsRejectedAsync	Fact	Lone surrogate escape is rejected.
LoneSurrogateInAName_IsRejectedAsync	Fact	Lone surrogate in a name is rejected.
ValidSurrogatePair_IsAcceptedAsync	Fact	Valid surrogate pair is accepted.
NulInAStringValue_IsRejectedAsync	Fact	Nul in a string value is rejected.
DeepNesting_IsRejectedAsync	Fact	Deep nesting is rejected.
NestingBeyondTheDepthLimit_InAValue_IsRejectedAsync	Fact	Nesting beyond the depth limit in a value is rejected.
HugeString_IsRejected_WithoutBufferingItAsync	Fact	Huge string is rejected without buffering it.
HugeNumber_IsRejectedAsync	Fact	Huge number is rejected.
StringLongerThanTheCellCap_IsRejectedAsync	Fact	String longer than the cell cap is rejected.
LongStoredText_IsTruncatedTo64CharactersAsync	Fact	Long stored text is truncated to64 characters.
NonObjectRow_InArrayForm_IsRejectedWithAPathAsync	Fact	Non object row in array form is rejected with a path.
RaggedRow_InColumnarForm_IsRejectedAsync	Fact	Ragged row in columnar form is rejected.
ColumnarShapeErrors_AreRejectedAsync	Fact	Columnar shape errors are rejected.
ColumnCap_IsEnforced_ForBothShapesAsync	Fact	Column cap is enforced for both shapes.
RowCap_IsEnforced_ForBothShapesAsync	Fact	Row cap is enforced for both shapes.
SizeCap_IsEnforcedAsync	Fact	Size cap is enforced.
LargeFile_WithinCaps_IsStreamedAndCountedAsync	Fact	Large file within caps is streamed and counted.
TokensSplitAcrossReads_AreParsedIdenticallyAsync	Fact	Tokens split across reads are parsed identically.
ManyRowsOfEmptyObjects_AreBatchedAndDoNotExhaustMemoryAsync	Fact	Many rows of empty objects are batched and do not exhaust memory.
SinkFailure_PropagatesAsync	Fact	Sink failure propagates.
Cancellation_IsHonouredAsync	Fact	Cancellation is honoured.
ClassBalance_AndConstantColumns_MatchTheCsvInspectionAsync	Fact	Class balance and constant columns match the csv inspection.
SameDataAsCsv_GivesTheSameColumnsRowsAndReportAsync	Fact	Same data as csv gives the same columns rows and report.
CsvParser_NamesJsonContentAsync	Fact	Csv parser names json content.
LargeIntegerLiteral_Detection	Fact	Large integer literal detection.

DatasetMappingRulesTests

Jube.Test.Exhaustive.Datasets.DatasetMappingRulesTests — 12 tests.

Test	Kind	Verifies
ValidMapping_HasNoIssues	Fact	Valid mapping has no issues.
MissingColumns_AreRejected	Fact	Missing columns are rejected.
ColumnCountAndIndexProblems_AreRejected	Fact	Column count and index problems are rejected.
CategoricalTextVariable_IsAcceptedAsAnIndependentVariable	Fact	Categorical text variable is accepted as an independent variable.
TextAndConstantVariables_AreRejected	Fact	Text and constant variables are rejected.
UnsafeNames_AreRejected	Theory	Unsafe names are rejected.
OverlongName_IsRejected	Fact	Overlong name is rejected.
DuplicateNames_AreRejectedCaseInsensitively	Fact	Duplicate names are rejected case insensitively.
UnknownProcessingTypes_AreRejected	Theory	Unknown processing types are rejected.
NoIncludedVariable_IsRejected	Fact	No included variable is rejected.
ExcludedColumnsAreNotValidated	Fact	Excluded columns are not validated.
ProcessingTypeHelpers_Behave	Fact	Processing type helpers behave.