Member Junction
    Preparing search index...

    Interface JSONRepairInfo

    Tracks how a malformed JSON response was repaired during prompt execution. Stored transiently on the prompt run entity and persisted into ValidationSummary JSON.

    interface JSONRepairInfo {
        method: "JSON5" | "LexicalEscaping" | "AIRepair";
        originalError: string;
        rawOutputPrefix: string;
        repaired: true;
        repairedOffsets?: number[];
        repairPromptRunId?: string;
    }
    Index

    Properties

    method originalError rawOutputPrefix repaired repairedOffsets? repairPromptRunId?

    Properties

    method: "JSON5" | "LexicalEscaping" | "AIRepair"

    Which repair strategy succeeded

    originalError: string

    The original JSON parse error message

    rawOutputPrefix: string

    First 200 characters of the raw LLM output (for diagnostics)

    repaired: true

    Whether the JSON was repaired

    repairedOffsets?: number[]

    Offsets escaped by the lexical repair, in fix order (LexicalEscaping only).

    Recorded because that repair infers intent: it assumes a quote inside a string value was meant to be escaped. That is nearly always right, but it is a judgment, so it should never be invisible. A rising count here means model output is drifting and the prompt likely needs attention.

    repairPromptRunId?: string

    The prompt run ID of the repair prompt execution (AI repair only)

    Settings

    Member Visibility

    On This Page

    Properties