Analytics Foundations

Tests of the shared analytics methodology used by both Exhaustive and Advanced Analytics, including anomaly handling and categorical encoding.

General

BetaCalibrationTests

Jube.Test.Analytics.BetaCalibrationTests — 9 tests.

Test	Kind	Verifies
Apply_ImprovesLogLossAndBrierScore	Fact	Apply improves log loss and brier score.
Apply_AlwaysReturnsAProbability	Fact	Apply always returns a probability.
Fit_ThrowsWhenScoresAreOutsideTheUnitInterval	Fact	Fit throws when scores are outside the unit interval.
Fit_ThrowsWithOnlyOneClass	Fact	Fit throws with only one class.
Fit_SilentlyDropsNaNScoresLikePlatt	Fact	Fit silently drops na n scores like platt.
Fit_ThrowsOnInfiniteScores	Fact	Fit throws on infinite scores.
ToPrimitives_FromPrimitives_RoundTrips	Fact	To primitives from primitives round trips.
FromPrimitives_RejectsAnOutOfRangeEpsilon	Fact	From primitives rejects an out of range epsilon.
FromPrimitives_RejectsTheWrongLength	Fact	From primitives rejects the wrong length.

BinaryMetricsTests

Jube.Test.Analytics.BinaryMetricsTests — 23 tests.

Test	Kind	Verifies
Confusion_KnownAnswers	Fact	Confusion known answers.
Confusion_ThresholdIsInclusive	Fact	Confusion threshold is inclusive.
ConfusionMatrix_DegenerateCountsAreDefined	Fact	Confusion matrix degenerate counts are defined.
FBeta_RejectsNonPositiveBeta	Fact	F beta rejects non positive beta.
RocAuc_WithTies_MatchesIndependentPairCount	Fact	Roc auc with ties matches independent pair count.
RocAuc_Extremes	Fact	Roc auc extremes.
RocAuc_SingleClassIsNaN	Fact	Roc auc single class is na n.
RocAuc_IsInvariantToMonotoneTransformAndDropsNaN	Fact	Roc auc is invariant to monotone transform and drops na n.
DeLong_KnownAnswer	Fact	De long known answer.
DeLong_TooFewRowsInAClassGivesNaNStandardError	Fact	De long too few rows in a class gives na n standard error.
RocCurve_GroupsTiesAndEndsAtOneOne	Fact	Roc curve groups ties and ends at one one.
RocCurve_SingleClassThrows	Fact	Roc curve single class throws.
KolmogorovSmirnov_KnownAnswer	Fact	Kolmogorov smirnov known answer.
AveragePrecision_KnownAnswer_AndPerfectRanking	Fact	Average precision known answer and perfect ranking.
LogLossAndBrier_KnownAnswers	Fact	Log loss and brier known answers.
LogLossAndBrier_WeightsMatchIndependentValuesAndDuplicatedRows	Fact	Log loss and brier weights match independent values and duplicated rows.
LogLoss_ExtremeProbabilitiesAreClippedNotInfinite	Fact	Log loss extreme probabilities are clipped not infinite.
Reliability_BinsAndExpectedCalibrationError_KnownAnswers	Fact	Reliability bins and expected calibration error known answers.
Reliability_EmptyBinsAreNaN_AndProbabilityOneLandsInLastBin	Fact	Reliability empty bins are na n and probability one lands in last bin.
PerfectlyCalibratedSet_HasZeroCalibrationError	Fact	Perfectly calibrated set has zero calibration error.
InvalidInputs_ThrowStandardExceptions	Fact	Invalid inputs throw standard exceptions.
NaNScoresAreDroppedFromEveryMetric	Fact	Na n scores are dropped from every metric.
LargeRandomSet_RankAucMatchesPairwiseDefinition	Fact	Large random set rank auc matches pairwise definition.

BivariateCopulaTests

Jube.Test.Analytics.BivariateCopulaTests — 7 tests.

Test	Kind	Verifies
GaussianCopula_RecoversItsOwnParameterFromSimulatedData	Theory	Gaussian copula recovers its own parameter from simulated data.
ClaytonCopula_RecoversItsOwnParameterFromSimulatedData	Theory	Clayton copula recovers its own parameter from simulated data.
GumbelCopula_RecoversItsOwnParameterFromSimulatedData	Theory	Gumbel copula recovers its own parameter from simulated data.
FrankCopula_RecoversItsOwnParameterFromSimulatedData	Theory	Frank copula recovers its own parameter from simulated data.
BivariateCopulaFitter_PicksGaussianForGaussianData	Fact	Bivariate copula fitter picks gaussian for gaussian data.
BivariateCopulaFitter_PicksClaytonForClaytonData	Fact	Bivariate copula fitter picks clayton for clayton data.
InverseConditionalCdf_IsAConsistentInverseOfConditionalCdf	Fact	Inverse conditional cdf is a consistent inverse of conditional cdf.

DVineCopulaTests

Jube.Test.Analytics.DVineCopulaTests — 4 tests.

Test	Kind	Verifies
Sample_RecoversPairwiseCorrelationsOfAKnownGaussianChain	Fact	Sample recovers pairwise correlations of a known gaussian chain.
Sample_ReproducesEachColumnsMarginalMeanAndStandardDeviation	Fact	Sample reproduces each columns marginal mean and standard deviation.
Fit_TruncatedAtTreeOne_StillProducesAValidSample	Fact	Fit truncated at tree one still produces a valid sample.
Sample_TwoVariables_RecoversTheSingleCorrelation	Fact	Sample two variables recovers the single correlation.

DataSplitTests

Jube.Test.Analytics.DataSplitTests — 9 tests.

Test	Kind	Verifies
Split_SetsAreDisjointAndInRange	Theory	Split sets are disjoint and in range.
Split_KnownAnswerSizes_OnExactFractions	Fact	Split known answer sizes on exact fractions.
Split_SameSeedIsReproducible_DifferentSeedDiffers	Fact	Split same seed is reproducible different seed differs.
Split_Stratified_PreservesClassProportionInEverySet	Fact	Split stratified preserves class proportion in every set.
Split_Empty_ReturnsEmptySets	Fact	Split empty returns empty sets.
Split_SingleRow_NeverDuplicatesTheRow	Fact	Split single row never duplicates the row.
Split_InvalidFractions_Throw	Theory	Split invalid fractions throw.
Split_NullLabels_Throws	Fact	Split null labels throws.
Take_GathersRowsInIndexOrder_AndRejectsOutOfRange	Fact	Take gathers rows in index order and rejects out of range.

IsotonicCalibrationTests

Jube.Test.Analytics.IsotonicCalibrationTests — 9 tests.

Test	Kind	Verifies
Apply_IsNonDecreasingAcrossTheFittedRange	Fact	Apply is non decreasing across the fitted range.
Apply_ImprovesLogLossAndBrierScore	Fact	Apply improves log loss and brier score.
Apply_ClampsToTheBoundaryOutsideTheFittedRange	Fact	Apply clamps to the boundary outside the fitted range.
Fit_ThrowsWhenAllScoresAreEqual	Fact	Fit throws when all scores are equal.
Fit_WithOnlyOneClassProducesANearConstantCalibrator	Fact	Fit with only one class produces a near constant calibrator.
Fit_ThrowsOnNonFiniteScores	Fact	Fit throws on non finite scores.
ToJson_FromJson_RoundTripsTheFittedFunction	Fact	To json from json round trips the fitted function.
FromState_RejectsNonIncreasingX	Fact	From state rejects non increasing x.
FromState_RejectsAnOutOfRangeY	Fact	From state rejects an out of range y.

PlattCalibrationTests

Jube.Test.Analytics.PlattCalibrationTests — 18 tests.

Test	Kind	Verifies
Fit_RawScore_MatchesIndependentNewton	Fact	Fit raw score matches independent newton.
Fit_LogitMode_MatchesIndependentNewton	Fact	Fit logit mode matches independent newton.
Fit_MiscalibratedSet_MatchesIndependentNewtonInBothModes	Fact	Fit miscalibrated set matches independent newton in both modes.
Apply_MiscalibratedSet_ImprovesLogLossBrierAndCalibrationError	Fact	Apply miscalibrated set improves log loss brier and calibration error.
Apply_IsStrictlyMonotoneAndPreservesAuc	Fact	Apply is strictly monotone and preserves auc.
Fit_InvertedScores_GivesNegativeSlopeThatIsStillMonotone	Fact	Fit inverted scores gives negative slope that is still monotone.
Fit_DuplicatedRowEqualsWeightTwo	Fact	Fit duplicated row equals weight two.
Fit_ZeroWeightRowIsIgnored	Fact	Fit zero weight row is ignored.
Fit_IsDeterministicAndIgnoresNaNScores	Fact	Fit is deterministic and ignores na n scores.
Fit_ExtremeScoresStayFinite	Fact	Fit extreme scores stay finite.
Fit_SeparableDataStillConvergesBecauseOfSmoothedTargets	Fact	Fit separable data still converges because of smoothed targets.
Fit_SingleClass_Throws	Fact	Fit single class throws.
Fit_ConstantScores_Throws	Fact	Fit constant scores throws.
Fit_InvalidArguments_ThrowStandardExceptions	Fact	Fit invalid arguments throw standard exceptions.
Fit_HonoursCancellation	Fact	Fit honours cancellation.
Calibrator_SnapshotsToPrimitivesAndBack	Fact	Calibrator snapshots to primitives and back.
Calibrator_FromPrimitives_RejectsBadSnapshots	Fact	Calibrator from primitives rejects bad snapshots.
Calibrator_NaNInGivesNaNOut_AndLogitModeClampsOutOfRange	Fact	Calibrator na n in gives na n out and logit mode clamps out of range.

PreprocessingTests

Jube.Test.Analytics.PreprocessingTests — 15 tests.

Test	Kind	Verifies
ZScore_FitIgnoresNaNAndMatchesIndependentValues	Fact	Z score fit ignores na n and matches independent values.
ZScore_ConstantColumnMapsToZero_SingleValueHasZeroSd	Fact	Z score constant column maps to zero single value has zero sd.
ZScore_IsFitOnTrainingOnlyAndReappliedUnchanged	Fact	Z score is fit on training only and reapplied unchanged.
MinMax_KnownAnswers_ClipAndConstant	Fact	Min max known answers clip and constant.
Robust_KnownAnswers	Fact	Robust known answers.
Robust_OutlierDoesNotMoveTheFit	Fact	Robust outlier does not move the fit.
Robust_ZeroMadMapsToZero	Fact	Robust zero mad maps to zero.
Winsoriser_QuantilesAndClamp	Fact	Winsoriser quantiles and clamp.
Winsoriser_InvalidArgumentsThrow	Fact	Winsoriser invalid arguments throw.
Imputer_MeanMedianModeConstant_KnownAnswers	Fact	Imputer mean median mode constant known answers.
Imputer_ModeTiesResolveToSmallest	Fact	Imputer mode ties resolve to smallest.
Imputer_ReplacesNaNAndInfinityAndFlagsThem	Fact	Imputer replaces na n and infinity and flags them.
Imputer_ConstantNeedsNoData_OthersNeedFiniteValues	Fact	Imputer constant needs no data others need finite values.
FitOnEmptyOrAllMissing_ThrowsForEveryScaler	Fact	Fit on empty or all missing throws for every scaler.
Pipeline_ImputeThenScale_MatchesResolveForTheMean	Fact	Pipeline impute then scale matches resolve for the mean.

PriorShiftTests

Jube.Test.Analytics.PriorShiftTests — 6 tests.

Test	Kind	Verifies
Apply_KnownAnswers	Fact	Apply known answers.
Apply_EqualPriorsIsIdentity_AndEndpointsAreFixed	Fact	Apply equal priors is identity and endpoints are fixed.
Apply_IsStrictlyMonotone	Fact	Apply is strictly monotone.
Apply_InvalidPriors_Throw	Theory	Apply invalid priors throw.
Apply_ProbabilityOutOfRange_Throws	Fact	Apply probability out of range throws.
SyntheticPopulation_BalancedFitCorrectedToKnownBaseRate	Fact	Synthetic population balanced fit corrected to known base rate.

RegressionAndThresholdTests

Jube.Test.Analytics.RegressionAndThresholdTests — 9 tests.

Test	Kind	Verifies
Regression_KnownAnswers	Fact	Regression known answers.
Regression_PerfectAndNaNAndZeroCases	Fact	Regression perfect and na n and zero cases.
Regression_NaNPairsAreDropped	Fact	Regression na n pairs are dropped.
Regression_InvalidInputsThrow	Fact	Regression invalid inputs throw.
MinimumRiskThreshold_KnownAnswers	Fact	Minimum risk threshold known answers.
MinimumRiskThreshold_IsWhereExpectedCostsAreEqual	Fact	Minimum risk threshold is where expected costs are equal.
MinimumRiskThreshold_InvalidCostsThrow	Fact	Minimum risk threshold invalid costs throw.
MinimumRiskThreshold_MinimisesEmpiricalCostOnCalibratedProbabilities	Fact	Minimum risk threshold minimises empirical cost on calibrated probabilities.
ExpectedCost_KnownAnswer	Fact	Expected cost known answer.

SplitterTests

Jube.Test.Analytics.SplitterTests — 15 tests.

Test	Kind	Verifies
ForwardChaining_KnownAnswer	Fact	Forward chaining known answer.
ForwardChaining_PreservesTimeOrderOnShuffledInput	Fact	Forward chaining preserves time order on shuffled input.
ForwardChaining_GapLeavesRowsBetweenTrainAndTest	Fact	Forward chaining gap leaves rows between train and test.
ForwardChaining_InvalidArgumentsThrow	Fact	Forward chaining invalid arguments throw.
GroupKFold_NoEntityInTwoFolds_AndIsAPartition	Fact	Group k fold no entity in two folds and is a partition.
GroupKFold_FoldSizesAreReasonablyBalanced	Fact	Group k fold fold sizes are reasonably balanced.
GroupKFold_DeterministicBySeed_AndSeedChangesAssignment	Fact	Group k fold deterministic by seed and seed changes assignment.
GroupKFold_InvalidArgumentsThrow	Fact	Group k fold invalid arguments throw.
GroupSplit_TrainValidationTest_NeverSharesAnEntity	Fact	Group split train validation test never shares an entity.
GroupSplit_DeterministicBySeed_AndValidatesFractions	Fact	Group split deterministic by seed and validates fractions.
StratifiedKFold_PartitionsAndPreservesClassRatioPerFold	Fact	Stratified k fold partitions and preserves class ratio per fold.
StratifiedKFold_UnevenClassesStayBalancedWithinOne	Fact	Stratified k fold uneven classes stay balanced within one.
KFold_SeededDeterministicAndPartitions	Fact	K fold seeded deterministic and partitions.
KFold_InvalidArgumentsThrow	Fact	K fold invalid arguments throw.
LeakageGuard_DeliberatelyLeakedIndexIsDetected	Fact	Leakage guard deliberately leaked index is detected.

StandardisationTests

Jube.Test.Analytics.StandardisationTests — 6 tests.

Test	Kind	Verifies
ZScore_KnownAnswers	Theory	Z score known answers.
Resolve_PresentValue_IsStandardisedAndNotImputed	Fact	Resolve present value is standardised and not imputed.
Resolve_PresentValue_IsUntouchedWhenNotStandardised	Fact	Resolve present value is untouched when not standardised.
Resolve_Missing_UsesTheMeanSoAStandardisedInputIsZero	Fact	Resolve missing uses the mean so a standardised input is zero.
Resolve_Missing_UsesTheRawMeanWhenNotStandardised	Fact	Resolve missing uses the raw mean when not standardised.
Resolve_NanIsTreatedAsMissing	Fact	Resolve nan is treated as missing.

Anomaly

AnomalyAutoSelectorTests

Jube.Test.Analytics.Anomaly.AnomalyAutoSelectorTests — 7 tests.

Test	Kind	Verifies
LabelledSelectionPicksTheDetectorThatFindsTheKnownPositives	Fact	Labelled selection picks the detector that finds the known positives.
SelectionIsDeterministicAndRecordsEveryCandidate	Fact	Selection is deterministic and records every candidate.
TheGuardRejectsDegenerateAndFailingDetectors	Fact	The guard rejects degenerate and failing detectors.
NothingSelectedWhenEveryCandidateIsRejected	Fact	Nothing selected when every candidate is rejected.
TheTimeBudgetSkipsCandidatesNotReached	Fact	The time budget skips candidates not reached.
CancellationIsPropagatedNotRecordedAsATimeout	Fact	Cancellation is propagated not recorded as a timeout.
TheMassCurvesPreferATightLevelSet	Fact	The mass curves prefer a tight level set.

AnomalyDetectorTests

Jube.Test.Analytics.Anomaly.AnomalyDetectorTests — 23 tests.

Test	Kind	Verifies
PlantedAnomaliesAreRankedAboveNormalRowsAcrossSeeds	Theory	Planted anomalies are ranked above normal rows across seeds.
ScoresAreDeterministicBySeed	Theory	Scores are deterministic by seed.
DifferentSeedsChangeARandomisedDetector	Fact	Different seeds change a randomised detector.
ScoringIsThreadSafe	Theory	Scoring is thread safe.
NonFiniteValuesAreImputedInFitAndScore	Theory	Non finite values are imputed in fit and score.
ACancelledTokenStopsTheFit	Theory	A cancelled token stops the fit.
ACancelledTokenStopsScoring	Fact	A cancelled token stops scoring.
DegenerateInputsThrowClearExceptions	Theory	Degenerate inputs throw clear exceptions.
ScoringARowOfTheWrongWidthIsRejected	Fact	Scoring a row of the wrong width is rejected.
UnregularisedMahalanobisRejectsMoreVariablesThanRows	Fact	Unregularised mahalanobis rejects more variables than rows.
ShrinkageMakesMahalanobisUsableWhenVariablesExceedRows	Fact	Shrinkage makes mahalanobis usable when variables exceed rows.
GaussianMixtureRejectsTooFewRowsForItsComponents	Fact	Gaussian mixture rejects too few rows for its components.
MahalanobisBeatsMarginalDetectorsWhenTheCorrelationIsBroken	Fact	Mahalanobis beats marginal detectors when the correlation is broken.
IsolationForestPathLengthMatchesTheReferenceDefinition	Fact	Isolation forest path length matches the reference definition.
HbosMatchesTheReferenceDefinition	Fact	Hbos matches the reference definition.
MahalanobisMatchesTheReferenceDefinition	Theory	Mahalanobis matches the reference definition.
LedoitWolfIntensityMatchesTheReferenceDefinition	Fact	Ledoit wolf intensity matches the reference definition.
LocalOutlierFactorMatchesTheReferenceDefinition	Fact	Local outlier factor matches the reference definition.
KnnDistanceMatchesTheReferenceDefinition	Fact	Knn distance matches the reference definition.
KnnAndLofUseARowCapDeterministically	Fact	Knn and lof use a row cap deterministically.
PcaRetainsFewerComponentsThanVariablesAndScoresOffSubspacePoints	Fact	Pca retains fewer components than variables and scores off subspace points.
FeatureBaggingDrawsDifferentSubspacesPerMemberAndIsReproducible	Fact	Feature bagging draws different subspaces per member and is reproducible.
EnsembleCombinationsAllRankPlantedAnomaliesHigh	Theory	Ensemble combinations all rank planted anomalies high.

AnomalyReportAndParametersTests

Jube.Test.Analytics.Anomaly.AnomalyReportAndParametersTests — 8 tests.

Test	Kind	Verifies
ContaminationFlagsTheTopFractionAndRecordsWhatWasApplied	Fact	Contamination flags the top fraction and records what was applied.
ContaminationOutsideTheRangeIsRejected	Theory	Contamination outside the range is rejected.
TiedScoresFlagFewerRatherThanMoreThanTheTopFraction	Fact	Tied scores flag fewer rather than more than the top fraction.
TheReportRoundTripsThroughJsonAndSummarises	Fact	The report round trips through json and summarises.
ParametersParseWithDefaultsAndRejectUnknownProperties	Fact	Parameters parse with defaults and reject unknown properties.
OutOfBoundsParametersAreReported	Theory	Out of bounds parameters are reported.
OversizedAndNumericEnumParametersAreRejected	Fact	Oversized and numeric enum parameters are rejected.
FactoryBuildsEveryNativeAlgorithmAndRefusesTheOthers	Fact	Factory builds every native algorithm and refuses the others.

AnomalyVariableReducerTests

Jube.Test.Analytics.Anomaly.AnomalyVariableReducerTests — 17 tests.

Test	Kind	Verifies
ConstantAndNearZeroVarianceVariablesAreDropped	Fact	Constant and near zero variance variables are dropped.
AFrequentlyRepeatedButVariedVariableIsNotNearZeroVariance	Fact	A frequently repeated but varied variable is not near zero variance.
VariablesAboveTheMissingRateAreDropped	Fact	Variables above the missing rate are dropped.
ExactDuplicatesAndCorrelatedGroupsKeepOneRepresentative	Fact	Exact duplicates and correlated groups keep one representative.
TheCorrelationThresholdIsConfigurable	Fact	The correlation threshold is configurable.
CompleteLinkageDoesNotChainWeaklyCorrelatedVariables	Fact	Complete linkage does not chain weakly correlated variables.
VariableClusteringKeepsARepresentativePerCluster	Fact	Variable clustering keeps a representative per cluster.
ZeroMedianAbsoluteDeviationCanBeDroppedOnRequest	Fact	Zero median absolute deviation can be dropped on request.
RobustScalingUsesMedianAndMadAndFallsBackForBinaryColumns	Fact	Robust scaling uses median and mad and falls back for binary columns.
ZScoreScalingGivesUnitVariance	Fact	Z score scaling gives unit variance.
PcaProjectionChoosesComponentsByExplainedVarianceOrParallelAnalysis	Fact	Pca projection chooses components by explained variance or parallel analysis.
ReductionIsDeterministicBySeed	Fact	Reduction is deterministic by seed.
RunningWithoutReductionKeepsEveryVariableUntouched	Fact	Running without reduction keeps every variable untouched.
MissingValuesAreImputedWithTheFittedMedianOnTransform	Fact	Missing values are imputed with the fitted median on transform.
ReducingEveryVariableAwayIsAClearError	Fact	Reducing every variable away is a clear error.
InvalidOptionsAndShapesAreRejected	Fact	Invalid options and shapes are rejected.
EveryVariableHasADecisionWithAReason	Fact	Every variable has a decision with a reason.

Categorical

CategoricalConceptTests

Jube.Test.Analytics.Categorical.CategoricalConceptTests — 3 tests. Category: Unit.

Test	Kind	Verifies
EncodedCategoryBeatsTheStringAsZeroBaselineOnHeldOutData	Fact	Encoded category beats the string as zero baseline on held out data.
TheSameStringsNoisedAtRecallGiveTheSameScores	Fact	The same strings noised at recall give the same scores.
ALevelThatOnlyAppearsAtRecallDoesNotMoveOtherRowsAndStaysFinite	Fact	A level that only appears at recall does not move other rows and stays finite.

CategoricalFieldEncoderTests

Jube.Test.Analytics.Categorical.CategoricalFieldEncoderTests — 27 tests.

Test	Kind	Verifies
Fit_RanksByFrequencyAndUsesTheMostFrequentAsReference	Fact	Fit ranks by frequency and uses the most frequent as reference.
Fit_CountsAreExact	Fact	Fit counts are exact.
Apply_KnownLevelSetsExactlyOneIndicator	Fact	Apply known level sets exactly one indicator.
Apply_ReferenceLevelIsAllZero	Fact	Apply reference level is all zero.
Apply_UnseenAndPooledLevelsMapToOther	Fact	Apply unseen and pooled levels map to other.
Apply_MissingMapsToTheMissingIndicator	Theory	Apply missing maps to the missing indicator.
Apply_WithoutOtherOrMissingIndicatorUnseenAndMissingAreAllZero	Fact	Apply without other or missing indicator unseen and missing are all zero.
Indicator_MatchesEncodeForEveryIndex	Fact	Indicator matches encode for every index.
Apply_IsAtMostOneHot	Fact	Apply is at most one hot.
Apply_NormalisationVariantsGiveTheSameIndicators	Fact	Apply normalisation variants give the same indicators.
Fit_MaxLevelsPoolsTheTail	Fact	Fit max levels pools the tail.
Fit_MinimumFrequencyPools	Fact	Fit minimum frequency pools.
Fit_MinimumFractionRaisesTheMinimum	Fact	Fit minimum fraction raises the minimum.
Fit_CoverageTargetStopsEarly	Fact	Fit coverage target stops early.
Fit_TiesAreBrokenByOrdinalKeyOrder	Fact	Fit ties are broken by ordinal key order.
Fit_ConstantFieldProducesNoEncoder	Fact	Fit constant field produces no encoder.
Fit_NoValuesProducesNoEncoder	Fact	Fit no values produces no encoder.
Fit_SingleLevelWithEnoughMissingKeepsTheMissingIndicator	Fact	Fit single level with enough missing keeps the missing indicator.
Fit_LabelIsTheMostFrequentSpelling	Fact	Fit label is the most frequent spelling.
Fit_IsDeterministicAndOrderIndependent	Fact	Fit is deterministic and order independent.
Counter_CapsTheDistinctKeysAndCountsTheOverflow	Fact	Counter caps the distinct keys and counts the overflow.
Serialisation_RoundTripsToIdenticalBehaviour	Fact	Serialisation round trips to identical behaviour.
Serialisation_HostileLabelsSurviveARoundTrip	Fact	Serialisation hostile labels survive a round trip.
FromJson_RejectsTamperedOrInvalidState	Theory	From json rejects tampered or invalid state.
FromJson_RejectsUnknownVersions	Fact	From json rejects unknown versions.
Encode_ShortDestinationThrows	Fact	Encode short destination throws.
Resolve_HostileInputNeverThrows	Fact	Resolve hostile input never throws.

CategoricalNormaliserTests

Jube.Test.Analytics.Categorical.CategoricalNormaliserTests — 8 tests.

Test	Kind	Verifies
Key_NormalisesCaseWhitespaceAndUnicode	Theory	Key normalises case whitespace and unicode.
Key_MissingShapesReturnNull	Theory	Key missing shapes return null.
Key_IsCultureInvariant	Fact	Key is culture invariant.
Key_HostileStrings_NeverThrowAndAreBounded	Fact	Key hostile strings never throw and are bounded.
Key_LongStringsSharingAPrefixCollideByDesign	Fact	Key long strings sharing a prefix collide by design.
Key_IsIdempotentOnRandomStrings	Fact	Key is idempotent on random strings.
Label_KeepsCaseButRemovesControlCharacters	Fact	Label keeps case but removes control characters.
ToRaw_UsesTheInvariantCulture	Fact	To raw uses the invariant culture.

CategoricalPlannerTests

Jube.Test.Analytics.Categorical.CategoricalPlannerTests — 9 tests.

Test	Kind	Verifies
DeclaredStringIsCategorical	Fact	Declared string is categorical.
DeclaredNumericIsNotCategoricalUnlessIncluded	Fact	Declared numeric is not categorical unless included.
UndeclaredTypeIsInferredFromTheSample	Fact	Undeclared type is inferred from the sample.
IdentifierLikeFieldsAreSkippedAndReported	Fact	Identifier like fields are skipped and reported.
ExcludedFieldsAreSkipped	Fact	Excluded fields are skipped.
ConstantAndEmptyFieldsAreSkipped	Fact	Constant and empty fields are skipped.
TotalIndicatorBudgetRemovesTheLeastFrequentLevelsAcrossFields	Fact	Total indicator budget removes the least frequent levels across fields.
TotalIndicatorBudgetIsDeterministic	Fact	Total indicator budget is deterministic.
OptionsRoundTripAndSanitise	Fact	Options round trip and sanitise.