Member Junction
    Preparing search index...

    Interface PredictiveStudioBuilderPayload

    The agent payload = the modeling plan the conversation accumulated, plus the builder's outcome.

    interface PredictiveStudioBuilderPayload {
        Approved?: boolean;
        BuildAttemptUserMessageCount?: number;
        BuildResult?: PredictiveStudioBuildOutcome;
        CandidateFeatures: {
            Kind: "embedding" | "llm-derived" | "numeric" | "categorical";
            Name: string;
            SourceRef: string;
            Why: string;
        }[];
        CandidateSources: {
            Kind: | "Entity"
            | "Query"
            | "ExternalEntity"
            | "VectorSet"
            | "FeaturePipeline";
            Ref: string;
            Why: string;
        }[];
        Goal: string;
        Leaderboard?: { IterationID: string; Metric: number; ModelID?: string }[];
        LeakageNotes: { Action: "exclude" | "allow"; Field: string; Risk: string }[];
        ProposedBudget: {
            MaxComputeCost?: number;
            MaxRuns?: number;
            MaxWallclockMinutes?: number;
        };
        ProposedExperiments: {
            AlgorithmName: string;
            FeatureSet: string[];
            Hyperparameters?: Record<string, unknown>;
            Label: string;
            Priority: number;
            Rationale: string;
        }[];
        TargetDefinition: {
            AsOfStrategy?: {
                Column?: string;
                Mode: "none"
                | "column"
                | "offset";
                OffsetDays?: number;
            };
            EntityName: string;
            ProblemType: "classification"
            | "regression";
            SuccessMetric: string;
            TargetVariable: string;
        };
        ValidationStrategy: {
            K?: number;
            LockedHoldoutFraction: number;
            Strategy: "train_test_split"
            | "kfold"
            | "holdout";
            TestSize?: number;
        };
    }

    Hierarchy (View Summary)

    Index

    Properties

    Approved? BuildAttemptUserMessageCount? BuildResult? CandidateFeatures CandidateSources Goal Leaderboard? LeakageNotes ProposedBudget ProposedExperiments TargetDefinition ValidationStrategy

    Properties

    Approved?: boolean

    User approval gate — execution does not begin until this is true.

    BuildAttemptUserMessageCount?: number

    The count of USER messages in the conversation at the moment the orchestrator last FORCED a build (stamped by PredictiveStudioModelDevAgent.determineNextStep). After a failed build this is what distinguishes the stale "build it" message that triggered the failed attempt (same count → no re-force, the no-loop guard) from a FRESH user request to retry (higher count → deterministic rebuild).

    BuildResult?: PredictiveStudioBuildOutcome
    CandidateFeatures: {
        Kind: "embedding" | "llm-derived" | "numeric" | "categorical";
        Name: string;
        SourceRef: string;
        Why: string;
    }[]

    Candidate features proposed by the Data Scout, each with rationale.

    CandidateSources: {
        Kind:
            | "Entity"
            | "Query"
            | "ExternalEntity"
            | "VectorSet"
            | "FeaturePipeline";
        Ref: string;
        Why: string;
    }[]

    Candidate feed-in sources proposed by the Data Scout, each with rationale.

    Goal: string

    Business objective, refined from the user's initial goal.

    Leaderboard?: { IterationID: string; Metric: number; ModelID?: string }[]

    Execution-phase leaderboard — one entry per Experiment Session Iteration.

    LeakageNotes: { Action: "exclude" | "allow"; Field: string; Risk: string }[]

    Leakage risks identified by the Data Scout and the chosen action per field.

    ProposedBudget: {
        MaxComputeCost?: number;
        MaxRuns?: number;
        MaxWallclockMinutes?: number;
    }

    Proposed resource budget for the experiment session.

    ProposedExperiments: {
        AlgorithmName: string;
        FeatureSet: string[];
        Hyperparameters?: Record<string, unknown>;
        Label: string;
        Priority: number;
        Rationale: string;
    }[]

    Ranked experiments proposed by the Experiment Designer (feature combos × algorithms × hyperparameters).

    TargetDefinition: {
        AsOfStrategy?: {
            Column?: string;
            Mode: "none" | "column" | "offset";
            OffsetDays?: number;
        };
        EntityName: string;
        ProblemType: "classification"
        | "regression";
        SuccessMetric: string;
        TargetVariable: string;
    }

    Precise definition of what is being predicted.

    Type Declaration

    • OptionalAsOfStrategy?: { Column?: string; Mode: "none" | "column" | "offset"; OffsetDays?: number }

      Optional point-in-time assembly strategy.

    • EntityName: string

      Training-unit entity (e.g. "Members").

    • ProblemType: "classification" | "regression"

      Classification or regression.

    • SuccessMetric: string

      The deterministic success metric driving the search.

    • TargetVariable: string

      Label expression/column.

    ValidationStrategy: {
        K?: number;
        LockedHoldoutFraction: number;
        Strategy: "train_test_split" | "kfold" | "holdout";
        TestSize?: number;
    }

    Validation strategy for the search.

    Settings

    Member Visibility

    On This Page

    Properties