Interface ForgeStats

Per-run forge reliability rollup snapshot.

interface ForgeStats {
    attempts: number;
    approved: number;
    rejected: number;
    approvedConfidenceSum: number;
    uniqueNames: number;
    uniqueApproved: number;
    uniqueTerminalRejections: number;
    rejectionReasons: Record<ForgeRejectionCategory, number>;
}

Properties

attempts: number

Total forge attempts (approved + rejected combined).

approved: number

Attempts the judge approved.

rejected: number

Attempts the judge or shape validator rejected.

approvedConfidenceSum: number

Sum of confidence across approved attempts. Divide by approved for avg.

uniqueNames: number

Count of unique tool names seen this run (union of approved + rejected). A tool rejected then re-forged under the same name counts once.

uniqueApproved: number

Count of unique tool names that landed approved at least once this run.

uniqueTerminalRejections: number

Count of unique tool names that were ONLY rejected (never approved) this run. The retry loop did not recover these. Actionable signal for "real quality failures" vs retry churn.

rejectionReasons: Record<ForgeRejectionCategory, number>

Histogram of rejection reasons, classified via classifyForgeRejection. Keys match ForgeRejectionCategory.