Member Junction
    Preparing search index...

    Module @memberjunction/predictive-studio

    @memberjunction/predictive-studio

    The server-side engine of MemberJunction Predictive Studio — assemble features, train models, score records, run experiment searches, keep models honest over time, and expose all of it through Actions + Remote Operations.

    What — the engine that turns a client's own data into a trained, scored predictive model. Its core is four service classes — FeatureAssemblyExecutor, TrainingEngine, MLModelInferenceProcessor, ExperimentOrchestrator — surrounded by feature-pipeline discovery, ongoing maintenance, and two thin invocation surfaces (Actions and Remote Operations) so agents, UI, and workflows can drive it.

    Why — predictive modeling needs feature assembly, anti-skew correctness, honest validation, batch scoring, and a budgeted search. Each of those already has a home in MJ's substrate; this engine composes onto them (Record Set Processing, entities/RunView, Remote Operations, Agents, vectors) plus the Python ML sidecar — rather than re-implementing batching, audit, or inference. The cardinal correctness rule: one feature-assembly code path per generation of model, used identically for training and scoring, so there is no train/serve skew by construction.

    How it fits — in the four-layer architecture (data → feature → model → inference) this package owns the feature, model, and inference layers in TypeScript, sitting above the type contracts and the MLSidecar client, and is consumed by the Studio dashboard, the (planned) Model Development Agent, and any agent/workflow via its Actions + Remote Operations. See How it fits the whole.

    For the full architecture, read the Predictive Studio Guide (§4–§12 cover this package); for the design record, plans/predictive-studio.md.

    pnpm add @memberjunction/predictive-studio
    

    Depends on @memberjunction/predictive-studio-core (type contracts), @memberjunction/predictive-studio-sidecar (the MLSidecar client), @memberjunction/record-set-processor-base (the scoring work-type seam), and the MJ core/global/core-entities packages.

    Component Where What it does
    FeatureAssemblyExecutor src/feature-assembly/feature-assembly-executor.ts (record set, frozen FeatureSteps) → raw matrix + preprocessing recipe. The score-time assembly path ('on-demand') of models trained before Prediction Units; training and the scoring of models trained under a Prediction Unit use AsOfFeatureAssembler (src/feature-assembly/as-of/), which reads every feature as of each unit row's decision date. It refuses windowed-aggregate and series steps, which need a Prediction Unit pipeline. Splits the graph into data-assembly steps (built as raw columns in TS) vs. preprocessing steps (emitted as a PreprocessingOp[] recipe for the sidecar to fit/apply). Enforces the as-of point-in-time cutoff (as-of.ts) and the leakage deny-list (leakage-guard.ts).
    TrainingEngine src/training/training-engine.ts trainModel(input, deps) — refuse a pipeline without a Prediction Unit (unit-required) → read the unit rows → assemble as of each decision date → freeze a Training Snapshot → train through the sidecar trainer (/trainer) under the bound Evaluation Protocol → decide the gate in TypeScript → store the shipped artifact and create an immutable, versioned MJ: ML Models row + the MJ: ML Training Runs row. Built on injectable seams (IEntityFactory / IRecordLoader / ITrainerEvaluator / IArtifactStore).
    MLModelInferenceProcessor src/scoring/ml-model-inference-processor.ts A Record Set Processing work type (@RegisterClass(MLModelInferenceProcessor, 'ML Model'), alias 'MLModelInference') for batch + single-record scoring. Warm-loads the model once per run, assembles features transform-only (frozen — never re-fits), calls the sidecar /predict, returns an MLInferenceResultPayload. Ephemeral by default; the substrate's WriteBackProcessor applies an OutputMapping when present. An untrained model refuses the run in Preflight (or, with the opt-in autoTrain, is trained first — never in a dry run, which reports MODEL_NEEDS_TRAINING instead). Registers without forking the substrate (src/scoring/register.ts).
    ExperimentOrchestrator src/experiment/experiment-orchestrator.ts runSession(plan, deps, options) — deterministic execution of an approved ModelingPlanSpec in an in-process pool (production experiment sessions are switched off by ExperimentSessions.Enabled in @memberjunction/predictive-studio-core, and the Run Experiment Session action and Start Experiment Session operation refuse with ExperimentSessionsUnavailableError; see Guide §7): generate wave (IWaveStrategist) → train (bounded concurrency) → evaluate → leaderboard → prune → budget-gate → next wave. Wires the generic MJ: Experiments / Experiment Sessions / Experiment Session Iterations primitives.

    The MLSidecar client (the self-managing Python sidecar) lives in its own package, @memberjunction/predictive-studio-sidecar — import it directly from there; it is intentionally not re-exported here.

    Beyond the four core engines, the package's src/ is organized into focused folders, each its own barrel export:

    Folder Exports Role Plan ref
    feature-assembly/ FeatureAssemblyExecutor + as-of / leakage-guard / vision-llm / data-access seams The single skew-free assembly path and its correctness primitives (point-in-time cutoff, deny-list, vision-LLM-as-feature) §5 / §6 / §11
    training/ TrainingEngine + seams (IEntityFactory / IRecordLoader / ITrainerEvaluator / IArtifactStore) + artifact-store, evaluation-protocol-store, trainer-request / trainer-result Train under the bound Prediction Unit and Evaluation Protocol → decide the gate → persist an immutable, versioned MJ: ML Models row with full lineage when the gate ships one §3 / §4.3
    prediction-unit/ PredictionUnitRowReader, UnitTargetValidator, RunViewPredictionUnitStore, DecisionDateAdvance Read a unit's rows (censored and label-known rows excluded and counted), validate its target, store unit versions, and advance a fixed decision date to the run's day §4.3 / §5
    snapshot/ TrainingSnapshotBuilder, HoldoutCarvers, ISnapshotFileStore, SnapshotHash, RunViewTrainingSnapshotStore Freeze the assembled rows into an immutable, hashed Training Snapshot (matrix, row manifest, provenance files) with the protocol's locked holdout carved out §5
    protocol/ ProtocolSearchGuard The checks that bind a protocol to a unit and a pipeline: the candidate budget, a time index under a time-respecting holdout, protocol-versus-unit fit §5.1
    migration/ PipelineMigration The dry-run-first migration of existing pipelines to version-1 units and protocols (run by the Migrate Pipelines Action) §10.2
    scoring/ MLModelInferenceProcessor, LoadMLModelInferenceProcessor, the 'ML Model' work-type register, artifact-loader, scoring-binding Batch + single-record inference as a Record Set Processing work type, registered without forking the substrate §10
    experiment/ ExperimentOrchestrator + leaderboard / concurrency / wave-strategist / seams Deterministic, wave-based execution of an approved ModelingPlanSpec with leaderboard, pruning, and a budget gate §8 / §9
    feature-pipelines/ FeaturePipelineEngine (a BaseEngine cache) + projection types Discover/monitor Feature Pipelines (categorized MJ: Record Processes rows) — "what pipelines exist, what each writes to, when each last ran" §5.4 / SP6
    maintenance/ MaintenanceEngine, RetrainingPolicy + defaults, the honest RowCountProxyDriftDetector, seams Staleness detection, scheduled re-scoring, retraining triggers, and challenger-vs-incumbent promotion recommendation (built, not wired — see Guide §14) §12 / SP10
    scheduling/ createScheduledModelScoring (the Phase 2 north-star helper) One call binds a model to write its prediction into a target column on a recurring cron (auto-creates the owned Scheduled Job) §15 / PS2-6
    actions/ PredictiveStudio{TrainModel,ScoreRecordSet,RunExperiment,PromoteModel,ScheduleModelScoring,MigratePipelines}Action + LoadPredictiveStudioActions Invocation surface A — thin MJ Actions over the engines (see below) §12
    operations/ PredictiveStudio{TrainModel,ScoreRecordSet,RunFeaturePipeline,StartExperimentSession,ControlExperimentSession,PromoteModel,CreateScoringProcess}ServerOperation + LoadPredictiveStudioOperations Invocation surface B — typed Remote Operations over the engines (see below) §12

    Phase 2 extends the model's reach without new ML: the same scorer (MLModelInferenceProcessor) and the same entry point (ScoreRecordSet) plumbed into every MJ surface where a prediction is useful, each through a registry / ClassFactory seam so no base/types package depends on Predictive Studio. Full write-up: Predictive Studio Guide §15.

    Item Module What it adds
    PS2-1 scoring as a saved/scheduled Record Process scoring/startup-register.ts (PredictiveStudioScoringStartup), scoring/register.ts (registerMLScoringProcessor) Registers the 'ML Model' work type into RecordProcessorRegistry at MJAPI boot so a saved MJ: Record Processes row routes to the scorer. (Migration V202606280215… drops the closed WorkType CHECK so the registry governs work types.)
    PS2-2 per-model Action on Publish actions/model-scoring-action-generator.ts (ModelScoringActionGenerator), hooked from actions/promote-model.gate.ts Generates an idempotent child Action Score with <pipeline> v<n> (Type=Custom, parent's DriverClass, ParentID set, ModelID DefaultValue baked) — so every published model is a discoverable Action.
    PS2-3 scored-query enrichment scoring/ml-model-score-enricher.ts (MLModelScoreEnricher, @RegisterClass(QueryResultEnricherBase, 'ML Model Score')) Appends a model prediction column to a RunQuery result via the decoupled RunQueryParams.Enrichment seam in @memberjunction/core.
    PS2-6 scheduled model scoring (north-star) scheduling/scheduled-model-scoring.ts (createScheduledModelScoring), actions/schedule-model-scoring.action.ts One call creates an Active 'ML Model' Record Process with ScheduleEnabled + monthly cron + OutputMapping; saving it auto-creates the owned Scheduled Job.

    PS2-4 (the Interactive Components utilities.ml capability) lives outside this package — SimpleMLTools in @memberjunction/interactive-component-types, built by RuntimeUtilities.CreateSimpleMLTools() in @memberjunction/ng-react, marshaling ScoreRecordSet over GraphQL to this engine.

    Each item ships with an MJServer integration script (ps-live-recordprocess-scoring, ps-live-modelaction-generation, ps-inproc-scored-query, ps-inproc-scheduled-scoring).

    The business-user experience phase (the conversational north-star made real) builds on this engine without new ML: createScheduledModelScoring now, after saving the scheduled Record Process, also records an MJ: ML Model Scoring Binding (via upsertScoringBinding — MLModelID / RecordProcessID / TargetEntityID / TargetColumn / Mode='Scheduled'). That binding is the lineage seam the Angular surfaces key off — the generic, binding-driven risk-card form panel and the "Models in Production" Studio section both light up from it. On the agent side, the Model Development Agent now proactively offers to operationalize after a clean promote ("score everyone + refresh monthly?") and, on a yes, calls the Schedule Model Scoring action. Full plan + commit lineage: plans/predictive-studio-business-experience.md; architecture write-up: Predictive Studio Guide §16.

    The engine service classes are the real logic; both surfaces are thin and share one delegation path (operations/delegation.ts), so they train / score / experiment / promote through the same code — never duplicating it. Concretely: the Train action and the TrainModel operation build their dependencies with TrainingDelegation.BuildTrainingDeps, refuse the retired inputs by name (the operation: asOf, labelEventDates, maxRows and primaryKeyField; the action: MaxRows), and return the run's stored gate decision (outcome, branch, reason, shipped member, snapshot and protocol) through the one TrainModelMapping; the Score Record Set action and the ScoreRecordSet operation both score with BuildScoreRecordSetRunner() (the production model loader, sidecar and StorageProviderArtifactLoader); the Promote action and the PromoteModel operation both run ProductionModelPromotionGate; and the Run Experiment Session action and the StartExperimentSession operation both refuse first with EXPERIMENT_SESSIONS_UNAVAILABLE, before reading any input. Both surfaces return Predictive Studio refusals as typed result codes through PredictiveStudioRefusals.

    The Score Record Set scope reads are complete or refused: view and list scopes read every row; a filter scope with no maxRows (or null, or <= 0) reads every matching row, and an explicit maxRows that the scope exceeds is refused with READ_CAP_EXCEEDED instead of scoring a sample. A single scope must name every primary-key field of the entity, and a listId must be a UUID.

    • Actions (actions/) — discoverable, metadata-driven boundaries for agents, low-code workflows, and the visual designer. Six driver-class keys (PredictiveStudioTrainModelAction, …ScoreRecordSetAction, …RunExperimentAction, …PromoteModelAction, …ScheduleModelScoringAction, …MigratePipelinesAction, which has no Remote Operation) matching the metadata/actions/predictive-studio/ rows. Call LoadPredictiveStudioActions() from a server bootstrap to keep the @RegisterClass registrations from being tree-shaken.
    • Remote Operations (operations/) — the typed peers, invoked by stable key from client or server with framework-applied scope/permission gates. Seven keys: PredictiveStudio.TrainModel, .ScoreRecordSet, .RunFeaturePipeline, .StartExperimentSession, .ControlExperimentSession, .PromoteModel, .CreateScoringProcess. These are the Manual-mode server bodies for the CodeGen-emitted operation bases in @memberjunction/core-entities. Call LoadPredictiveStudioOperations() from a server bootstrap to anchor them.

    Which to reach for: Actions when an agent/workflow needs to discover and chain capabilities; Remote Operations when client or server code needs a typed, gated call by key. (Internal engine-to-engine calls use neither — they import the service class directly. See the Transport-Layer and Remote Operations guides.)

    import { TrainingEngine, FeatureAssemblyExecutor } from '@memberjunction/predictive-studio';
    import { MLModelInferenceProcessor, LoadMLModelInferenceProcessor } from '@memberjunction/predictive-studio';

    // 1. Ensure the scoring work-type registration survives bundling (call from a server bootstrap)
    LoadMLModelInferenceProcessor();

    // 2. Train under the pipeline's Prediction Unit + Evaluation Protocol — produces a new immutable MJ: ML Models row when the gate ships one
    const engine = new TrainingEngine();
    const result = await engine.trainModel({ pipelineId: retentionPipelineId }, productionDeps);
    // result.model is null when the stored gate decision ships no artifact (no-signal)

    // 3. Score — runs as the 'ML Model' Record Set Processing work type (on-demand or scheduled)
    // See the Record Set Processing Guide for RunNow / scope-override mechanics.
    predictive-studio-core (types)            ─┐
    predictive-studio-sidecar (MLSidecar) ─┤
    record-set-processor / -base (substrate) ─┤
    @memberjunction/actions + core-entities ─┤
    ▼
    @memberjunction/predictive-studio (this package)
    feature-assembly · prediction-unit · snapshot · protocol · training · migration · scoring ·
    experiment · feature-pipelines · maintenance · actions · operations
    │ exposed to / consumed by
    ┌───────────────────────┬───────────┴───────────┬────────────────────┐
    Actions (agents/ Remote Operations Studio dashboard (planned) Model
    workflows/low-code) (typed, by key) (ng-dashboards) Development Agent

    For one-time setup on a fresh clone (building the workspace and the Python environment), see Setting up the bundled Python environment.

    pnpm run build              # tsc && tsc-alias -f (after the workspace dependencies are built)
    pnpm test # vitest run — unit tests (all seams mocked; no DB, no live sidecar)
    pnpm run test:integration # opt-in end-to-end live test (PS_INTEGRATION=1) — spawns the real
    # managed Python sidecar; skips gracefully if the venv is missing.
    # Prereq: pnpm --filter @memberjunction/predictive-studio-sidecar run setup:python

    Classes

    AsOfFeatureAssembler
    AsOfSettingHasNoEffectError
    AssemblyRefused
    AutoTrainShippedNoModelError
    BasePredictiveStudioAction
    CodeVersionInfo
    CompleteReader
    ComputePricing
    DateCarvedCarver
    DecisionDateAdvance
    DefaultDriftAlertEmitter
    EvaluateFeatureAndPredictionDriftAction
    EvaluationProtocolSpecs
    EvaluationProtocolVersioning
    ExperimentBudget
    ExperimentOrchestrator
    ExperimentScores
    ExperimentSessionsUnavailableError
    FeatureAssemblyExecutor
    FeaturePipelineEngine
    FeatureResolver
    FeatureValueShadowAuditSink
    HoldoutCarvers
    InMemoryArtifactLoader
    InMemoryArtifactStore
    InMemorySnapshotFileStore
    KeyedReader
    LeakageGuardEnforcer
    LockedHoldoutSplitter
    MaintenanceEngine
    MetadataEntityFactory
    MetadataExperimentEntityFactory
    MJSidecarPredictor
    MJSidecarTrainer
    MLModelChallengerScoreRunner
    MLModelInferenceProcessor
    MLModelScoreEnricher
    ModelNeedsTrainingError
    ModelNotScorableError
    ModelNotServableError
    ModelReadiness
    ModelScoringActionGenerator
    PipelineAutoTrainer
    PipelineMigration
    PlanOrderWaveStrategist
    PreAuthDownloads
    PredictionUnitRowReader
    PredictionUnitVersioning
    PredictiveStudioControlExperimentSessionServerOperation
    PredictiveStudioCreateScoringProcessServerOperation
    PredictiveStudioMigratePipelinesAction
    PredictiveStudioModelDevAgent
    PredictiveStudioPipelineBuilder
    PredictiveStudioPipelineBuilderAgent
    PredictiveStudioPromoteModelAction
    PredictiveStudioPromoteModelServerOperation
    PredictiveStudioRefusals
    PredictiveStudioRunExperimentAction
    PredictiveStudioRunFeaturePipelineServerOperation
    PredictiveStudioScheduleModelScoringAction
    PredictiveStudioScoreRecordSetAction
    PredictiveStudioScoreRecordSetServerOperation
    PredictiveStudioScoringStartup
    PredictiveStudioStartExperimentSessionServerOperation
    PredictiveStudioStorage
    PredictiveStudioTrainModelAction
    PredictiveStudioTrainModelServerOperation
    ProductionModelPromotionGate
    ProductionRescoreRunner
    ProductionScoreRecordSetRunner
    ProtocolDefaults
    ProtocolSearchGuard
    ProviderAssemblyMetadata
    ProviderMigrationReads
    PSIDriftDetector
    ReadCapExceededError
    RecordHistory
    RollingOriginCarver
    RowCountProxyDriftDetector
    RunViewDataAccess
    RunViewEvaluationProtocolStore
    RunViewMaintenanceLoader
    RunViewMLModelLoader
    RunViewPipelineUnitLookup
    RunViewPredictionUnitStore
    RunViewRecordHistoryReader
    RunViewRecordLoader
    RunViewRowCounter
    RunViewTrainingRowsLookup
    RunViewTrainingSnapshotStore
    SeededRandomCarver
    SeriesFeatures
    ShadowScoringCoordinator
    SnapshotFileExistsError
    SnapshotFileFormatError
    SnapshotFiles
    SnapshotHash
    SnapshotIntegrityError
    SnapshotManifest
    SnapshotNonFiniteCellError
    SnapshotObjectKeys
    SnapshotRetrainer
    SnapshotWriteOnce
    StorageProviderArtifactLoader
    StorageProviderArtifactStore
    StorageProviderSnapshotFileStore
    SystemClock
    TrainerLineageRefusedError
    TrainerModelLineageReader
    TrainerRequestBuilder
    TrainerRunRecorder
    TrainingAttemptFailedError
    TrainingDelegation
    TrainingEngine
    TrainingEngineExperimentTrainer
    TrainingKeys
    TrainingRefusedError
    TrainingSnapshotBuilder
    UnitModelLineageParser
    UnitTargetValidator
    UnresolvedPipelineResolver
    VisionFeatureExtractor
    WindowedAggregates

    Interfaces

    AsOfAssemblyInput
    AssembledColumn
    AssembledRow
    AssemblyEntityInfo
    AssemblyFieldInfo
    AssemblyRowRef
    AutoTrainRequest
    Budget
    BuildGateDecision
    BuildPredictionInput
    BuildPredictionResult
    BuiltSnapshot
    CalibrationHold
    ChallengerComparison
    ChallengerScoreOutput
    ChampionChallengerPair
    ChildRow
    CompleteReadParams
    ConfidenceAndDivergenceInput
    ConfidenceAndDivergenceMetrics
    CountRowsResult
    CreateScoringProcessOptions
    CreateScoringProcessResult
    DatedFeatureSpec
    DatedRow
    DatedSourceSpec
    DecisionDateAdvanceDeps
    DeletedChildRecord
    DeletedRecordRead
    DispatchShadowScoreOptions
    DominanceResult
    DriftContext
    DriftResult
    EntityHistory
    EphemeralPrediction
    ExperimentDeps
    ExperimentLeaderboardEntry
    ExperimentRunOptions
    ExperimentScore
    ExperimentSessionResult
    FeatureAndPredictionDriftInput
    FeatureAndPredictionDriftReport
    FeatureAssemblyParams
    FeatureAssemblyResult
    FeaturePipelineSummary
    FeaturePSIResult
    FetchRowsParams
    FetchRowsResult
    HoldoutCarveOutcome
    IArtifactLoader
    IArtifactStore
    IAssemblyMetadata
    IChallengerScoreRunner
    IClock
    ICountingDataAccess
    IDriftAlertEmitter
    IDriftDetector
    IEntityFactory
    IEvaluationProtocolStore
    IExperimentEntityFactory
    IExperimentTrainer
    IFeatureDataAccess
    IFeaturePipelineRunner
    IHoldoutCarver
    IMaintenanceClock
    IMaintenanceLoader
    IMLModelLoader
    IModelAutoTrainer
    IModelPromotionGate
    InMemorySnapshotFile
    IPipelineMigrationReads
    IPipelineResolver
    IPipelineUnitLookup
    IPredictionUnitStore
    IRecordHistoryReader
    IRecordLoader
    IRescoreRunner
    IRowCounter
    IScoreRecordSetRunner
    IShadowScoreAuditSink
    IShadowWorkQueuePublisher
    ISidecarPredictor
    ISnapshotFileStore
    ITrainerEvaluator
    ITrainingRowsLookup
    ITrainingSnapshotStore
    IVisionPromptRunner
    IWaveStrategist
    LoadDriftFromFeatureValuesOptions
    LoadedModel
    LoadedProtocol
    LoadedUnitModel
    LockedHoldoutSplit
    LoggedTrainerRun
    MaintenanceDeps
    MaintenancePassEntry
    MaintenancePassOptions
    MaintenancePassResult
    MetricDifference
    MigrationDeps
    MigrationReportRow
    MigrationRun
    MLAutoTrainConfig
    MLFeatureImportancePayload
    MLInferenceDeps
    MLInferenceResultPayload
    MLLeaderboardEntryPayload
    MLModelInferenceProcessorOptions
    MLModelScoreEnricherConfig
    MLScoringConfiguration
    ModelArtifactRef
    ModelServingRef
    ModelTrainingLineage
    PipelineAutoTrainerOptions
    PipelineConfig
    PredictiveStudioBuilderPayload
    PredictiveStudioBuildOutcome
    PresentRowsByKey
    ProcessDriftSummaryItem
    ProductionScoreRecordSetRunnerOptions
    PromoteModelRequest
    RecordChangeRow
    RecordProcessDriftLoad
    RecordSetDescriptor
    RescoreRequest
    RescoreResult
    ResolvedPipeline
    RetrainingPolicy
    RetrainOutcome
    RetrainResult
    RunBoundedOptions
    ScheduledModelScoringResult
    ScheduleModelScoringOptions
    ScoreRecordSetRequest
    ScoreRecordSetResult
    ScoringBindingInput
    ScoringScope
    ScoringSeams
    SeriesPoint
    ShadowDispatchResult
    ShadowFeatureValueAuditRecord
    ShadowScoreEnvelope
    ShadowScoringCoordinatorOptions
    SnapshotBuildInput
    SnapshotContent
    SnapshotContentHashInput
    SnapshotFileBytes
    SnapshotHashes
    StalenessReason
    StalenessResult
    StorageDriverBinding
    StoredFileLocation
    StoredFileRow
    StoredSnapshotFile
    StoredSnapshotFiles
    TrainerModelLineage
    TrainerRequestInput
    TrainExperimentInput
    TrainExperimentResult
    TrainingDeps
    TrainingLeakageAudit
    TrainingRowCarve
    TrainModelInput
    TrainModelResult
    UnitAssemblyInputs
    UnitModelLineage
    UnitReadDeps
    UnitRow
    UnitTrainingAssembly
    UserInfo
    VisionExtractionResult
    WaveStrategistContext

    Type Aliases

    AlgorithmRow
    AsOfAssembledMatrix
    AsOfAssemblyResult
    AsOfSettingSource
    AssemblyContext
    AssemblyTrainingRefusal
    AutoTrainTrainFn
    CanonicalLabelValue
    CarvedRowRef
    CatalogStatus
    ChildRowIndex
    CompleteReadOutcome
    DriftSeverity
    EvaluationProtocolVersionRow
    FeatureKind
    FeaturePipelineRunStatus
    FeatureResolution
    FeatureValueRole
    HistoryEntity
    HistoryValue
    JsonObject
    MaintenanceBindingMode
    MatrixCell
    ProblemType
    PromotableStatus
    PromoteModelOutcome
    PromotionRecommendation
    ProposedExperiment
    ProtocolDefaultsReadDeps
    ProtocolRequirement
    ReadByKeysOutcome
    ResolvedAlgorithm
    ScheduledScoringScope
    ScheduleModelScoringFn
    ScoringBindingMode
    ScoringCadence
    ScoringValueKind
    SeriesFeatureRef
    SeriesHistory
    SessionStopReason
    SnapshotIntegrityTarget
    SnapshotManifestRef
    SnapshotRowSave
    SourceRow
    StoredEvaluationProtocol
    StoredFileRowSave
    TrainerRequestReissue
    TrainerRunTarget
    UnitBinding
    UnitRowReadResult
    UnitRowsRead
    VisionPromptResolver
    WriteBackDirective

    Variables

    CADENCE_CRON
    CATEGORICAL_CARDINALITY_CAP
    COUNT_SIGNATURE_MAX_DISTINCT
    DECISION_DATE_ADVANCE_ACTOR
    DEFAULT_DRIFT_MAX_ROWS
    DEFAULT_MIN_SAMPLE_SIZE
    DEFAULT_PSI_EPSILON
    DEFAULT_RETRAINING_POLICY
    DEFAULT_SPLIT_SEED
    DEFAULT_WAVE_CONCURRENCY
    DRIFT_SWEEP_BATCH_SIZE
    EVALUATE_DRIFT_DRIVER_CLASS
    FEATURE_PIPELINE_CATEGORY_NAME
    HELD_MARKER_PREFIX
    MaintenanceSystemClock
    MIGRATE_PIPELINES_DRIVER_CLASS
    MIGRATION_ACTOR
    ML_INFERENCE_WORK_TYPE
    ML_INFERENCE_WORK_TYPE_ALIAS
    ML_INFERENCE_WORK_TYPES
    ML_MODEL_SCORE_ENRICHER_KEY
    OTHER_CATEGORY_LEVEL
    PIPELINE_BUILDER_SUBAGENT_NAME
    PredictiveStudioEngineVersion
    PROMOTABLE_STATUSES
    PROMOTE_MODEL_DRIVER_CLASS
    PSI_CRITICAL_THRESHOLD
    PSI_WARNING_THRESHOLD
    READ_BY_KEYS_BATCH
    RESERVED_CANDIDATE_NAMES
    RESERVED_CANDIDATE_PREFIX
    RUN_EXPERIMENT_DRIVER_CLASS
    SCHEDULE_MODEL_SCORING_DRIVER_CLASS
    SCORE_RECORD_SET_DRIVER_CLASS
    SHADOW_NOT_DISPATCHED
    TRAIN_MODEL_DRIVER_CLASS
    UNSHOWN_IMPORTANCE_KIND_NOTE

    Functions

    ActivityCountAsOf
    bestEntry
    BestEntry
    BuildOutcomeMessage
    buildOutputInstruction
    BuildOutputInstruction
    buildProductionExperimentDeps
    BuildProductionExperimentDeps
    BuildProductionMLInferenceDeps
    BuildScoreRecordSetRunner
    buildVisionUserMessage
    BuildVisionUserMessage
    cadenceToCron
    CadenceToCron
    CatalogStatusOf
    ClassifyPSISeverity
    ComputeCategoricalFeaturePSI
    ComputeConfidenceAndChallengerDivergence
    ComputeNumericFeaturePSI
    ComputePopulationStabilityIndex
    createScheduledModelScoring
    CreateScheduledModelScoring
    createScoringProcess
    CreateScoringProcess
    DaysSinceLastActivityAsOf
    detectSingleFeatureDominance
    DetectSingleFeatureDominance
    EvaluateFeatureAndPredictionDrift
    FilterAsOf
    GenerateMarkdownReport
    HiddenImportanceNote
    isMLInferenceWorkType
    IsMLInferenceWorkType
    IsModelingPlanApproved
    LoadFeatureAndPredictionDriftForRecordProcesses
    LoadFeatureAndPredictionDriftFromFeatureValues
    LoadMLModelInferenceProcessor
    LoadMLModelScoreEnricher
    LoadPredictiveStudioActions
    LoadPredictiveStudioControlExperimentSessionOperation
    LoadPredictiveStudioModelDevAgent
    LoadPredictiveStudioOperations
    LoadPredictiveStudioPipelineBuilderAgent
    LoadPredictiveStudioPromoteModelOperation
    LoadPredictiveStudioRunFeaturePipelineOperation
    LoadPredictiveStudioScoreRecordSetOperation
    LoadPredictiveStudioStartExperimentSessionOperation
    LoadPredictiveStudioTrainModelOperation
    LocalArtifactPath
    MatrixToFeatureRows
    ModelArtifactMissingMessage
    ModelHasTrainedArtifact
    ModelingPlanToPipelineConfig
    ModelNeedsTrainingMessage
    ParseFeatureImportance
    parseVisionOutput
    ParseVisionOutput
    PromoteModelViaGate
    rankLeaderboard
    RankLeaderboard
    readImageRef
    ReadImageRef
    ReadMetric
    recordScoringRun
    RecordScoringRun
    registerMLScoringProcessor
    RegisterMLScoringProcessor
    ResolveAsOfDate
    ResolveComparisonMetric
    ResolveEntity
    ResolveLocalArtifactBaseDir
    resolveMLInferenceProcessor
    ResolveMLInferenceProcessor
    runBounded
    RunBounded
    RunExperimentSessionViaOrchestrator
    RunMaintenancePass
    ScoreRecordSetViaRunner
    selectPrunedIterationIds
    SelectPrunedIterationIds
    shouldForceBuild
    ShouldForceBuild
    sortByPriority
    SortByPriority
    SummarizeBuildResult
    upsertScoringBinding
    UpsertScoringBinding
    ValidateDriftThresholds
    WasTrainingLeakageFlagged