sdl-mcp 0.6.0

package/dist/agent/types.d.ts

@@ -0,0 +1,86 @@
+export type TaskType = "debug" | "review" | "implement" | "explain";
+export type RungType = "card" | "skeleton" | "hotPath" | "raw";
+export type ActionStatus = "pending" | "inProgress" | "completed" | "failed";
+export interface AgentTask {
+    taskType: TaskType;
+    taskText: string;
+    repoId: string;
+    budget?: BudgetOptions;
+    options?: TaskOptions;
+}
+export interface BudgetOptions {
+    maxTokens?: number;
+    maxActions?: number;
+    maxDurationMs?: number;
+}
+export interface TaskOptions {
+    focusSymbols?: string[];
+    focusPaths?: string[];
+    includeTests?: boolean;
+    requireDiagnostics?: boolean;
+}
+export interface RungPath {
+    rungs: RungType[];
+    estimatedTokens: number;
+    estimatedDurationMs: number;
+    reasoning: string;
+}
+export interface PlannedExecution {
+    task: AgentTask;
+    path: RungPath;
+    sequence: Action[];
+}
+export interface Action {
+    id: string;
+    type: "getCard" | "getSkeleton" | "getHotPath" | "needWindow" | "search" | "analyze";
+    status: ActionStatus;
+    input: Record<string, unknown>;
+    output?: unknown;
+    error?: string;
+    timestamp: number;
+    durationMs: number;
+    evidence: Evidence[];
+}
+export interface Evidence {
+    type: "symbolCard" | "skeleton" | "hotPath" | "codeWindow" | "delta" | "diagnostic" | "searchResult";
+    reference: string;
+    summary: string;
+    timestamp: number;
+}
+export interface OrchestrationResult {
+    taskId: string;
+    taskType: TaskType;
+    actionsTaken: Action[];
+    path: RungPath;
+    finalEvidence: Evidence[];
+    summary: string;
+    success: boolean;
+    error?: string;
+    metrics: ExecutionMetrics;
+    answer?: string;
+    nextBestAction?: string;
+}
+export interface ExecutionMetrics {
+    totalDurationMs: number;
+    totalTokens: number;
+    totalActions: number;
+    successfulActions: number;
+    failedActions: number;
+    cacheHits: number;
+}
+export interface AgentContext {
+    repoId: string;
+    symbolIds: string[];
+    filePaths: string[];
+    diagnostics?: Diagnostic[];
+}
+export interface Diagnostic {
+    symbolId?: string;
+    filePath: string;
+    line: number;
+    column: number;
+    severity: "error" | "warning" | "info";
+    message: string;
+    code?: string;
+}
+//# sourceMappingURL=types.d.ts.map

package/dist/agent/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/agent/types.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,QAAQ,GAAG,OAAO,GAAG,QAAQ,GAAG,WAAW,GAAG,SAAS,CAAC;AAEpE,MAAM,MAAM,QAAQ,GAAG,MAAM,GAAG,UAAU,GAAG,SAAS,GAAG,KAAK,CAAC;AAE/D,MAAM,MAAM,YAAY,GAAG,SAAS,GAAG,YAAY,GAAG,WAAW,GAAG,QAAQ,CAAC;AAE7E,MAAM,WAAW,SAAS;IACxB,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,aAAa,CAAC;IACvB,OAAO,CAAC,EAAE,WAAW,CAAC;CACvB;AAED,MAAM,WAAW,aAAa;IAC5B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,WAAW;IAC1B,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC9B;AAED,MAAM,WAAW,QAAQ;IACvB,KAAK,EAAE,QAAQ,EAAE,CAAC;IAClB,eAAe,EAAE,MAAM,CAAC;IACxB,mBAAmB,EAAE,MAAM,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,gBAAgB;IAC/B,IAAI,EAAE,SAAS,CAAC;IAChB,IAAI,EAAE,QAAQ,CAAC;IACf,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,MAAM,WAAW,MAAM;IACrB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EACA,SAAS,GACT,aAAa,GACb,YAAY,GACZ,YAAY,GACZ,QAAQ,GACR,SAAS,CAAC;IACd,MAAM,EAAE,YAAY,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/B,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,QAAQ,EAAE,CAAC;CACtB;AAED,MAAM,WAAW,QAAQ;IACvB,IAAI,EACA,YAAY,GACZ,UAAU,GACV,SAAS,GACT,YAAY,GACZ,OAAO,GACP,YAAY,GACZ,cAAc,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,mBAAmB;IAClC,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,QAAQ,CAAC;IACnB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,IAAI,EAAE,QAAQ,CAAC;IACf,aAAa,EAAE,QAAQ,EAAE,CAAC;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,OAAO,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,gBAAgB,CAAC;IAC1B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,gBAAgB;IAC/B,eAAe,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,YAAY;IAC3B,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,WAAW,CAAC,EAAE,UAAU,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,OAAO,GAAG,SAAS,GAAG,MAAM,CAAC;IACvC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf"}

package/dist/agent/types.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=types.js.map

package/dist/agent/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/agent/types.ts"],"names":[],"mappings":""}

package/dist/benchmark/index.d.ts

@@ -0,0 +1,4 @@
+export { ThresholdEvaluator, loadThresholdConfig, loadBaselineMetrics, saveBaselineMetrics, type BenchmarkThresholds, type MetricThreshold, type ThresholdEvaluation, type ThresholdEvaluationResult, } from "./threshold.js";
+export { RegressionReportGenerator, type RegressionReportData, type RegressionDelta, type RegressionReport, } from "./regression.js";
+export { MetricSmoothing, runBenchmarkWithSmoothing, type SmoothingConfig, type MetricMeasurement, type SmoothingResult, } from "./smoothing.js";
+//# sourceMappingURL=index.d.ts.map

package/dist/benchmark/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/benchmark/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,kBAAkB,EAClB,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,KAAK,mBAAmB,EACxB,KAAK,eAAe,EACpB,KAAK,mBAAmB,EACxB,KAAK,yBAAyB,GAC/B,MAAM,gBAAgB,CAAC;AAExB,OAAO,EACL,yBAAyB,EACzB,KAAK,oBAAoB,EACzB,KAAK,eAAe,EACpB,KAAK,gBAAgB,GACtB,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,eAAe,EACf,yBAAyB,EACzB,KAAK,eAAe,EACpB,KAAK,iBAAiB,EACtB,KAAK,eAAe,GACrB,MAAM,gBAAgB,CAAC"}

package/dist/benchmark/index.js

@@ -0,0 +1,4 @@
+export { ThresholdEvaluator, loadThresholdConfig, loadBaselineMetrics, saveBaselineMetrics, } from "./threshold.js";
+export { RegressionReportGenerator, } from "./regression.js";
+export { MetricSmoothing, runBenchmarkWithSmoothing, } from "./smoothing.js";
+//# sourceMappingURL=index.js.map

package/dist/benchmark/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/benchmark/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,kBAAkB,EAClB,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,GAKpB,MAAM,gBAAgB,CAAC;AAExB,OAAO,EACL,yBAAyB,GAI1B,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EACL,eAAe,EACf,yBAAyB,GAI1B,MAAM,gBAAgB,CAAC"}

package/dist/benchmark/regression.d.ts

@@ -0,0 +1,40 @@
+export interface RegressionReportData {
+    timestamp: string;
+    repoId: string;
+    currentMetrics: Record<string, number>;
+    baselineMetrics?: Record<string, number>;
+    thresholds: Record<string, unknown>;
+    evaluations: import("./threshold.js").ThresholdEvaluation[];
+}
+export interface RegressionDelta {
+    metricName: string;
+    category: string;
+    currentValue: number;
+    baselineValue?: number;
+    delta?: number;
+    deltaPercent?: number;
+    trend: "improved" | "degraded" | "neutral" | "unknown";
+    impact: "high" | "medium" | "low" | "none";
+    description: string;
+}
+export interface RegressionReport {
+    summary: {
+        passed: boolean;
+        totalMetrics: number;
+        improved: number;
+        degraded: number;
+        neutral: number;
+        unknown: number;
+    };
+    deltas: RegressionDelta[];
+    failingThresholds: string[];
+    recommendations: string[];
+}
+export declare class RegressionReportGenerator {
+    generate(data: RegressionReportData): RegressionReport;
+    private generateDelta;
+    private generateRecommendations;
+    generateMarkdownReport(report: RegressionReport): string;
+    generateJsonReport(report: RegressionReport): string;
+}
+//# sourceMappingURL=regression.d.ts.map

package/dist/benchmark/regression.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"regression.d.ts","sourceRoot":"","sources":["../../src/benchmark/regression.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACvC,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACzC,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACpC,WAAW,EAAE,OAAO,gBAAgB,EAAE,mBAAmB,EAAE,CAAC;CAC7D;AAED,MAAM,WAAW,eAAe;IAC9B,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,UAAU,GAAG,UAAU,GAAG,SAAS,GAAG,SAAS,CAAC;IACvD,MAAM,EAAE,MAAM,GAAG,QAAQ,GAAG,KAAK,GAAG,MAAM,CAAC;IAC3C,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE;QACP,MAAM,EAAE,OAAO,CAAC;QAChB,YAAY,EAAE,MAAM,CAAC;QACrB,QAAQ,EAAE,MAAM,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,OAAO,EAAE,MAAM,CAAC;QAChB,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC;IACF,MAAM,EAAE,eAAe,EAAE,CAAC;IAC1B,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAC5B,eAAe,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,qBAAa,yBAAyB;IACpC,QAAQ,CAAC,IAAI,EAAE,oBAAoB,GAAG,gBAAgB;IAwCtD,OAAO,CAAC,aAAa;IAmErB,OAAO,CAAC,uBAAuB;IA4F/B,sBAAsB,CAAC,MAAM,EAAE,gBAAgB,GAAG,MAAM;IAqFxD,kBAAkB,CAAC,MAAM,EAAE,gBAAgB,GAAG,MAAM;CAGrD"}

package/dist/benchmark/regression.js

@@ -0,0 +1,199 @@
+export class RegressionReportGenerator {
+    generate(data) {
+        const deltas = [];
+        const failingThresholds = [];
+        const recommendations = [];
+        for (const evaluation of data.evaluations) {
+            const delta = this.generateDelta(evaluation);
+            deltas.push(delta);
+            if (!evaluation.passed) {
+                failingThresholds.push(`${evaluation.category}.${evaluation.metricName}`);
+            }
+        }
+        const improved = deltas.filter((d) => d.trend === "improved").length;
+        const degraded = deltas.filter((d) => d.trend === "degraded").length;
+        const neutral = deltas.filter((d) => d.trend === "neutral").length;
+        const unknown = deltas.filter((d) => d.trend === "unknown").length;
+        recommendations.push(...this.generateRecommendations(deltas, data.evaluations));
+        return {
+            summary: {
+                passed: failingThresholds.length === 0,
+                totalMetrics: deltas.length,
+                improved,
+                degraded,
+                neutral,
+                unknown,
+            },
+            deltas,
+            failingThresholds,
+            recommendations,
+        };
+    }
+    generateDelta(evaluation) {
+        const { metricName, category, currentValue, baselineValue, delta, deltaPercent, threshold, passed, } = evaluation;
+        let trend = "unknown";
+        let impact = "none";
+        let description = "";
+        if (baselineValue !== undefined &&
+            deltaPercent !== undefined &&
+            delta !== undefined) {
+            if (threshold.trend === "lower-is-better") {
+                trend = delta < 0 ? "improved" : delta > 0 ? "degraded" : "neutral";
+            }
+            else {
+                trend = delta > 0 ? "improved" : delta < 0 ? "degraded" : "neutral";
+            }
+            const absDeltaPercent = Math.abs(deltaPercent);
+            if (absDeltaPercent > 20) {
+                impact = "high";
+            }
+            else if (absDeltaPercent > 10) {
+                impact = "medium";
+            }
+            else if (absDeltaPercent > 5) {
+                impact = "low";
+            }
+        }
+        else {
+            description = "No baseline available for comparison";
+        }
+        if (!passed) {
+            impact = impact === "none" ? "low" : impact;
+            description = evaluation.message;
+        }
+        else if (impact !== "none" && trend !== "neutral") {
+            const trendStr = trend === "improved" ? "improved" : "degraded";
+            const direction = trend === "improved" ? "↓" : "↑";
+            description = `${trendStr} by ${Math.abs(deltaPercent ?? 0).toFixed(1)}% ${direction}`;
+        }
+        else if (baselineValue !== undefined) {
+            description = `${currentValue.toFixed(2)} (baseline: ${baselineValue.toFixed(2)})`;
+        }
+        else {
+            description = `${currentValue.toFixed(2)} (no baseline)`;
+        }
+        return {
+            metricName,
+            category,
+            currentValue,
+            baselineValue,
+            delta,
+            deltaPercent,
+            trend,
+            impact,
+            description,
+        };
+    }
+    generateRecommendations(deltas, evaluations) {
+        const recommendations = [];
+        const highImpactDegraded = deltas.filter((d) => d.impact === "high" && d.trend === "degraded");
+        if (highImpactDegraded.length > 0) {
+            recommendations.push(`CRITICAL: ${highImpactDegraded.length} metrics have high-impact degradation:`, ...highImpactDegraded.map((d) => `  - ${d.metricName}: ${d.description}`));
+        }
+        const mediumImpactDegraded = deltas.filter((d) => d.impact === "medium" && d.trend === "degraded");
+        if (mediumImpactDegraded.length > 0) {
+            recommendations.push(`WARNING: ${mediumImpactDegraded.length} metrics have medium-impact degradation:`, ...mediumImpactDegraded.map((d) => `  - ${d.metricName}: ${d.description}`));
+        }
+        const failingEvaluations = evaluations.filter((e) => !e.passed);
+        if (failingEvaluations.length > 0) {
+            const indexFailures = failingEvaluations.filter((e) => e.category === "indexing");
+            if (indexFailures.length > 0) {
+                recommendations.push("Indexing performance issues detected. Consider:", "  - Increasing indexing concurrency", "  - Adding large generated files to ignore list", "  - Reducing maxFileBytes threshold", "  - Checking for tree-sitter query performance issues");
+            }
+            const qualityFailures = failingEvaluations.filter((e) => e.category === "quality");
+            if (qualityFailures.length > 0) {
+                recommendations.push("Quality metrics degraded. This may indicate:", "  - Changes in symbol extraction logic", "  - Modifications to tree-sitter queries", "  - Regression in edge detection algorithms");
+            }
+            const performanceFailures = failingEvaluations.filter((e) => e.category === "performance");
+            if (performanceFailures.length > 0) {
+                recommendations.push("Runtime performance degraded. Consider:", "  - Reviewing recent database query changes", "  - Checking for N+1 query patterns", "  - Optimizing graph traversal algorithms");
+            }
+            const coverageFailures = failingEvaluations.filter((e) => e.category === "coverage");
+            if (coverageFailures.length > 0) {
+                recommendations.push("Edge coverage degraded. Verify:", "  - Call edge extraction queries are working correctly", "  - Import resolution hasn't regressed", "  - New code patterns are properly supported");
+            }
+        }
+        const improvedMetrics = deltas.filter((d) => d.trend === "improved");
+        if (improvedMetrics.length > 0 && recommendations.length === 0) {
+            recommendations.push(`All metrics passed! ${improvedMetrics.length} metrics improved.`);
+        }
+        return recommendations;
+    }
+    generateMarkdownReport(report) {
+        const lines = [];
+        lines.push("# Benchmark Regression Report");
+        lines.push("");
+        lines.push(`Generated: ${new Date().toISOString()}`);
+        lines.push("");
+        const statusEmoji = report.summary.passed ? "✅" : "❌";
+        lines.push(`## Summary ${statusEmoji}`);
+        lines.push("");
+        lines.push(`- **Status**: ${report.summary.passed ? "PASSED" : "FAILED"}`);
+        lines.push(`- **Total Metrics**: ${report.summary.totalMetrics}`);
+        lines.push(`- **Improved**: ${report.summary.improved}`);
+        lines.push(`- **Degraded**: ${report.summary.degraded}`);
+        lines.push(`- **Neutral**: ${report.summary.neutral}`);
+        lines.push(`- **Unknown**: ${report.summary.unknown}`);
+        lines.push("");
+        if (report.failingThresholds.length > 0) {
+            lines.push("## Failed Thresholds");
+            lines.push("");
+            for (const threshold of report.failingThresholds) {
+                lines.push(`- ❌ \`${threshold}\``);
+            }
+            lines.push("");
+        }
+        lines.push("## Metric Deltas");
+        lines.push("");
+        const categorized = new Map();
+        for (const delta of report.deltas) {
+            if (!categorized.has(delta.category)) {
+                categorized.set(delta.category, []);
+            }
+            categorized.get(delta.category).push(delta);
+        }
+        for (const [category, deltas] of categorized) {
+            lines.push(`### ${category}`);
+            lines.push("");
+            for (const delta of deltas) {
+                const impactEmoji = {
+                    high: "🔴",
+                    medium: "🟡",
+                    low: "🟢",
+                    none: "⚪",
+                }[delta.impact];
+                const trendEmoji = {
+                    improved: "✓",
+                    degraded: "✗",
+                    neutral: "−",
+                    unknown: "?",
+                }[delta.trend];
+                lines.push(`#### ${delta.metricName} ${impactEmoji} ${trendEmoji}`);
+                lines.push("");
+                lines.push(`- **Current**: ${delta.currentValue.toFixed(2)}`);
+                if (delta.baselineValue !== undefined) {
+                    lines.push(`- **Baseline**: ${delta.baselineValue.toFixed(2)}`);
+                }
+                if (delta.deltaPercent !== undefined) {
+                    lines.push(`- **Change**: ${delta.deltaPercent.toFixed(2)}%`);
+                }
+                lines.push(`- **Impact**: ${delta.impact}`);
+                lines.push(`- **Description**: ${delta.description}`);
+                lines.push("");
+            }
+        }
+        if (report.recommendations.length > 0) {
+            lines.push("## Recommendations");
+            lines.push("");
+            for (const recommendation of report.recommendations) {
+                lines.push(recommendation);
+            }
+            lines.push("");
+        }
+        return lines.join("\n");
+    }
+    generateJsonReport(report) {
+        return JSON.stringify(report, null, 2);
+    }
+}
+//# sourceMappingURL=regression.js.map

package/dist/benchmark/regression.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"regression.js","sourceRoot":"","sources":["../../src/benchmark/regression.ts"],"names":[],"mappings":"AAmCA,MAAM,OAAO,yBAAyB;IACpC,QAAQ,CAAC,IAA0B;QACjC,MAAM,MAAM,GAAsB,EAAE,CAAC;QACrC,MAAM,iBAAiB,GAAa,EAAE,CAAC;QACvC,MAAM,eAAe,GAAa,EAAE,CAAC;QAErC,KAAK,MAAM,UAAU,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;YAC1C,MAAM,KAAK,GAAG,IAAI,CAAC,aAAa,CAAC,UAAU,CAAC,CAAC;YAC7C,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YAEnB,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC;gBACvB,iBAAiB,CAAC,IAAI,CACpB,GAAG,UAAU,CAAC,QAAQ,IAAI,UAAU,CAAC,UAAU,EAAE,CAClD,CAAC;YACJ,CAAC;QACH,CAAC;QAED,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,UAAU,CAAC,CAAC,MAAM,CAAC;QACrE,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,UAAU,CAAC,CAAC,MAAM,CAAC;QACrE,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;QACnE,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;QAEnE,eAAe,CAAC,IAAI,CAClB,GAAG,IAAI,CAAC,uBAAuB,CAAC,MAAM,EAAE,IAAI,CAAC,WAAW,CAAC,CAC1D,CAAC;QAEF,OAAO;YACL,OAAO,EAAE;gBACP,MAAM,EAAE,iBAAiB,CAAC,MAAM,KAAK,CAAC;gBACtC,YAAY,EAAE,MAAM,CAAC,MAAM;gBAC3B,QAAQ;gBACR,QAAQ;gBACR,OAAO;gBACP,OAAO;aACR;YACD,MAAM;YACN,iBAAiB;YACjB,eAAe;SAChB,CAAC;IACJ,CAAC;IAEO,aAAa,CACnB,UAAwD;QAExD,MAAM,EACJ,UAAU,EACV,QAAQ,EACR,YAAY,EACZ,aAAa,EACb,KAAK,EACL,YAAY,EACZ,SAAS,EACT,MAAM,GACP,GAAG,UAAU,CAAC;QAEf,IAAI,KAAK,GAA6B,SAAS,CAAC;QAChD,IAAI,MAAM,GAA8B,MAAM,CAAC;QAC/C,IAAI,WAAW,GAAG,EAAE,CAAC;QAErB,IACE,aAAa,KAAK,SAAS;YAC3B,YAAY,KAAK,SAAS;YAC1B,KAAK,KAAK,SAAS,EACnB,CAAC;YACD,IAAI,SAAS,CAAC,KAAK,KAAK,iBAAiB,EAAE,CAAC;gBAC1C,KAAK,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YACtE,CAAC;iBAAM,CAAC;gBACN,KAAK,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YACtE,CAAC;YAED,MAAM,eAAe,GAAG,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YAC/C,IAAI,eAAe,GAAG,EAAE,EAAE,CAAC;gBACzB,MAAM,GAAG,MAAM,CAAC;YAClB,CAAC;iBAAM,IAAI,eAAe,GAAG,EAAE,EAAE,CAAC;gBAChC,MAAM,GAAG,QAAQ,CAAC;YACpB,CAAC;iBAAM,IAAI,eAAe,GAAG,CAAC,EAAE,CAAC;gBAC/B,MAAM,GAAG,KAAK,CAAC;YACjB,CAAC;QACH,CAAC;aAAM,CAAC;YACN,WAAW,GAAG,sCAAsC,CAAC;QACvD,CAAC;QAED,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,GAAG,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC;YAC5C,WAAW,GAAG,UAAU,CAAC,OAAO,CAAC;QACnC,CAAC;aAAM,IAAI,MAAM,KAAK,MAAM,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;YACpD,MAAM,QAAQ,GAAG,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC;YAChE,MAAM,SAAS,GAAG,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;YACnD,WAAW,GAAG,GAAG,QAAQ,OAAO,IAAI,CAAC,GAAG,CAAC,YAAY,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,SAAS,EAAE,CAAC;QACzF,CAAC;aAAM,IAAI,aAAa,KAAK,SAAS,EAAE,CAAC;YACvC,WAAW,GAAG,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;QACrF,CAAC;aAAM,CAAC;YACN,WAAW,GAAG,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,gBAAgB,CAAC;QAC3D,CAAC;QAED,OAAO;YACL,UAAU;YACV,QAAQ;YACR,YAAY;YACZ,aAAa;YACb,KAAK;YACL,YAAY;YACZ,KAAK;YACL,MAAM;YACN,WAAW;SACZ,CAAC;IACJ,CAAC;IAEO,uBAAuB,CAC7B,MAAyB,EACzB,WAA2D;QAE3D,MAAM,eAAe,GAAa,EAAE,CAAC;QAErC,MAAM,kBAAkB,GAAG,MAAM,CAAC,MAAM,CACtC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU,CACrD,CAAC;QACF,IAAI,kBAAkB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAClC,eAAe,CAAC,IAAI,CAClB,aAAa,kBAAkB,CAAC,MAAM,wCAAwC,EAC9E,GAAG,kBAAkB,CAAC,GAAG,CACvB,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,WAAW,EAAE,CAC/C,CACF,CAAC;QACJ,CAAC;QAED,MAAM,oBAAoB,GAAG,MAAM,CAAC,MAAM,CACxC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,KAAK,KAAK,UAAU,CACvD,CAAC;QACF,IAAI,oBAAoB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACpC,eAAe,CAAC,IAAI,CAClB,YAAY,oBAAoB,CAAC,MAAM,0CAA0C,EACjF,GAAG,oBAAoB,CAAC,GAAG,CACzB,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,WAAW,EAAE,CAC/C,CACF,CAAC;QACJ,CAAC;QAED,MAAM,kBAAkB,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAChE,IAAI,kBAAkB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAClC,MAAM,aAAa,GAAG,kBAAkB,CAAC,MAAM,CAC7C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,UAAU,CACjC,CAAC;YACF,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC7B,eAAe,CAAC,IAAI,CAClB,iDAAiD,EACjD,qCAAqC,EACrC,iDAAiD,EACjD,qCAAqC,EACrC,uDAAuD,CACxD,CAAC;YACJ,CAAC;YAED,MAAM,eAAe,GAAG,kBAAkB,CAAC,MAAM,CAC/C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,SAAS,CAChC,CAAC;YACF,IAAI,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC/B,eAAe,CAAC,IAAI,CAClB,8CAA8C,EAC9C,wCAAwC,EACxC,0CAA0C,EAC1C,6CAA6C,CAC9C,CAAC;YACJ,CAAC;YAED,MAAM,mBAAmB,GAAG,kBAAkB,CAAC,MAAM,CACnD,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,aAAa,CACpC,CAAC;YACF,IAAI,mBAAmB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACnC,eAAe,CAAC,IAAI,CAClB,yCAAyC,EACzC,6CAA6C,EAC7C,qCAAqC,EACrC,2CAA2C,CAC5C,CAAC;YACJ,CAAC;YAED,MAAM,gBAAgB,GAAG,kBAAkB,CAAC,MAAM,CAChD,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,UAAU,CACjC,CAAC;YACF,IAAI,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAChC,eAAe,CAAC,IAAI,CAClB,iCAAiC,EACjC,wDAAwD,EACxD,wCAAwC,EACxC,8CAA8C,CAC/C,CAAC;YACJ,CAAC;QACH,CAAC;QAED,MAAM,eAAe,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,KAAK,UAAU,CAAC,CAAC;QACrE,IAAI,eAAe,CAAC,MAAM,GAAG,CAAC,IAAI,eAAe,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC/D,eAAe,CAAC,IAAI,CAClB,uBAAuB,eAAe,CAAC,MAAM,oBAAoB,CAClE,CAAC;QACJ,CAAC;QAED,OAAO,eAAe,CAAC;IACzB,CAAC;IAED,sBAAsB,CAAC,MAAwB;QAC7C,MAAM,KAAK,GAAa,EAAE,CAAC;QAE3B,KAAK,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;QAC5C,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,cAAc,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;QACrD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAEf,MAAM,WAAW,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QACtD,KAAK,CAAC,IAAI,CAAC,cAAc,WAAW,EAAE,CAAC,CAAC;QACxC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,iBAAiB,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC3E,KAAK,CAAC,IAAI,CAAC,wBAAwB,MAAM,CAAC,OAAO,CAAC,YAAY,EAAE,CAAC,CAAC;QAClE,KAAK,CAAC,IAAI,CAAC,mBAAmB,MAAM,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC,CAAC;QACzD,KAAK,CAAC,IAAI,CAAC,mBAAmB,MAAM,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC,CAAC;QACzD,KAAK,CAAC,IAAI,CAAC,kBAAkB,MAAM,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;QACvD,KAAK,CAAC,IAAI,CAAC,kBAAkB,MAAM,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;QACvD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAEf,IAAI,MAAM,CAAC,iBAAiB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;YACnC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACf,KAAK,MAAM,SAAS,IAAI,MAAM,CAAC,iBAAiB,EAAE,CAAC;gBACjD,KAAK,CAAC,IAAI,CAAC,SAAS,SAAS,IAAI,CAAC,CAAC;YACrC,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACjB,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;QAC/B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAEf,MAAM,WAAW,GAAG,IAAI,GAAG,EAA6B,CAAC;QACzD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;YAClC,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;gBACrC,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;YACtC,CAAC;YACD,WAAW,CAAC,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC/C,CAAC;QAED,KAAK,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,IAAI,WAAW,EAAE,CAAC;YAC7C,KAAK,CAAC,IAAI,CAAC,OAAO,QAAQ,EAAE,CAAC,CAAC;YAC9B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAEf,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;gBAC3B,MAAM,WAAW,GAAG;oBAClB,IAAI,EAAE,IAAI;oBACV,MAAM,EAAE,IAAI;oBACZ,GAAG,EAAE,IAAI;oBACT,IAAI,EAAE,GAAG;iBACV,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;gBAEhB,MAAM,UAAU,GAAG;oBACjB,QAAQ,EAAE,GAAG;oBACb,QAAQ,EAAE,GAAG;oBACb,OAAO,EAAE,GAAG;oBACZ,OAAO,EAAE,GAAG;iBACb,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;gBAEf,KAAK,CAAC,IAAI,CAAC,QAAQ,KAAK,CAAC,UAAU,IAAI,WAAW,IAAI,UAAU,EAAE,CAAC,CAAC;gBACpE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;gBACf,KAAK,CAAC,IAAI,CAAC,kBAAkB,KAAK,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;gBAC9D,IAAI,KAAK,CAAC,aAAa,KAAK,SAAS,EAAE,CAAC;oBACtC,KAAK,CAAC,IAAI,CAAC,mBAAmB,KAAK,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;gBAClE,CAAC;gBACD,IAAI,KAAK,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;oBACrC,KAAK,CAAC,IAAI,CAAC,iBAAiB,KAAK,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;gBAChE,CAAC;gBACD,KAAK,CAAC,IAAI,CAAC,iBAAiB,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC5C,KAAK,CAAC,IAAI,CAAC,sBAAsB,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;gBACtD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACjB,CAAC;QACH,CAAC;QAED,IAAI,MAAM,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtC,KAAK,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACf,KAAK,MAAM,cAAc,IAAI,MAAM,CAAC,eAAe,EAAE,CAAC;gBACpD,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YAC7B,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACjB,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAED,kBAAkB,CAAC,MAAwB;QACzC,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;IACzC,CAAC;CACF"}

package/dist/benchmark/smoothing.d.ts

@@ -0,0 +1,40 @@
+export interface SmoothingConfig {
+    warmupRuns: number;
+    sampleRuns: number;
+    outlierMethod: "iqr" | "stddev" | "none";
+    iqrMultiplier: number;
+}
+export interface MetricMeasurement {
+    metricName: string;
+    values: number[];
+}
+export interface SmoothingResult {
+    metricName: string;
+    originalValues: number[];
+    smoothedValue: number;
+    outliersRemoved: number[];
+    stats: {
+        mean: number;
+        median: number;
+        stddev: number;
+        min: number;
+        max: number;
+    };
+}
+export declare class MetricSmoothing {
+    private config;
+    constructor(config: SmoothingConfig);
+    filterWarmup(measurements: Record<string, number[]>): Record<string, number[]>;
+    detectOutliers(values: number[]): number[];
+    private detectIQROutliers;
+    private detectStdDevOutliers;
+    private percentile;
+    smooth(measurements: Record<string, number[]>): Record<string, number>;
+    smoothMetric(metricName: string, values: number[]): SmoothingResult;
+    private calculateStats;
+}
+export declare function runBenchmarkWithSmoothing(config: SmoothingConfig, benchmarkFn: () => Promise<Record<string, number>>): Promise<{
+    smoothed: Record<string, number>;
+    raw: Record<string, number[]>;
+}>;
+//# sourceMappingURL=smoothing.d.ts.map

package/dist/benchmark/smoothing.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"smoothing.d.ts","sourceRoot":"","sources":["../../src/benchmark/smoothing.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,eAAe;IAC9B,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACzC,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB;AAED,MAAM,WAAW,eAAe;IAC9B,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,KAAK,EAAE;QACL,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,GAAG,EAAE,MAAM,CAAC;QACZ,GAAG,EAAE,MAAM,CAAC;KACb,CAAC;CACH;AAED,qBAAa,eAAe;IAC1B,OAAO,CAAC,MAAM,CAAkB;gBAEpB,MAAM,EAAE,eAAe;IAInC,YAAY,CACV,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,GACrC,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC;IAc3B,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,EAAE;IAkB1C,OAAO,CAAC,iBAAiB;IAYzB,OAAO,CAAC,oBAAoB;IAU5B,OAAO,CAAC,UAAU;IAalB,MAAM,CAAC,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC;IAWtE,YAAY,CAAC,UAAU,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,eAAe;IAenE,OAAO,CAAC,cAAc;CAyBvB;AAED,wBAAsB,yBAAyB,CAC7C,MAAM,EAAE,eAAe,EACvB,WAAW,EAAE,MAAM,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,GACjD,OAAO,CAAC;IACT,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CAC/B,CAAC,CAsBD"}

package/dist/benchmark/smoothing.js

@@ -0,0 +1,121 @@
+export class MetricSmoothing {
+    config;
+    constructor(config) {
+        this.config = config;
+    }
+    filterWarmup(measurements) {
+        const filtered = {};
+        for (const [metricName, values] of Object.entries(measurements)) {
+            if (values.length > this.config.warmupRuns) {
+                filtered[metricName] = values.slice(this.config.warmupRuns);
+            }
+            else {
+                filtered[metricName] = values;
+            }
+        }
+        return filtered;
+    }
+    detectOutliers(values) {
+        if (this.config.outlierMethod === "none") {
+            return [];
+        }
+        if (values.length < 3) {
+            return [];
+        }
+        if (this.config.outlierMethod === "iqr") {
+            return this.detectIQROutliers(values);
+        }
+        else if (this.config.outlierMethod === "stddev") {
+            return this.detectStdDevOutliers(values);
+        }
+        return [];
+    }
+    detectIQROutliers(values) {
+        const sorted = [...values].sort((a, b) => a - b);
+        const q1 = this.percentile(sorted, 25);
+        const q3 = this.percentile(sorted, 75);
+        const iqr = q3 - q1;
+        const lowerBound = q1 - this.config.iqrMultiplier * iqr;
+        const upperBound = q3 + this.config.iqrMultiplier * iqr;
+        return values.filter((v) => v < lowerBound || v > upperBound);
+    }
+    detectStdDevOutliers(values) {
+        const mean = values.reduce((sum, v) => sum + v, 0) / values.length;
+        const variance = values.reduce((sum, v) => sum + Math.pow(v - mean, 2), 0) / values.length;
+        const stddev = Math.sqrt(variance);
+        const threshold = 2 * stddev;
+        return values.filter((v) => Math.abs(v - mean) > threshold);
+    }
+    percentile(sorted, p) {
+        const index = (p / 100) * (sorted.length - 1);
+        const lower = Math.floor(index);
+        const upper = Math.ceil(index);
+        const weight = index - lower;
+        if (upper >= sorted.length) {
+            return sorted[sorted.length - 1];
+        }
+        return sorted[lower] * (1 - weight) + sorted[upper] * weight;
+    }
+    smooth(measurements) {
+        const results = {};
+        for (const [metricName, values] of Object.entries(measurements)) {
+            const result = this.smoothMetric(metricName, values);
+            results[metricName] = result.smoothedValue;
+        }
+        return results;
+    }
+    smoothMetric(metricName, values) {
+        const outliers = this.detectOutliers(values);
+        const filteredValues = values.filter((v) => !outliers.includes(v));
+        const stats = this.calculateStats(filteredValues);
+        return {
+            metricName,
+            originalValues: values,
+            smoothedValue: stats.median,
+            outliersRemoved: outliers,
+            stats,
+        };
+    }
+    calculateStats(values) {
+        if (values.length === 0) {
+            return {
+                mean: 0,
+                median: 0,
+                stddev: 0,
+                min: 0,
+                max: 0,
+            };
+        }
+        const sorted = [...values].sort((a, b) => a - b);
+        const sum = values.reduce((acc, v) => acc + v, 0);
+        const mean = sum / values.length;
+        const variance = values.reduce((acc, v) => acc + Math.pow(v - mean, 2), 0) / values.length;
+        return {
+            mean,
+            median: this.percentile(sorted, 50),
+            stddev: Math.sqrt(variance),
+            min: sorted[0],
+            max: sorted[sorted.length - 1],
+        };
+    }
+}
+export async function runBenchmarkWithSmoothing(config, benchmarkFn) {
+    const measurements = {};
+    for (let i = 0; i < config.warmupRuns + config.sampleRuns; i++) {
+        const result = await benchmarkFn();
+        for (const [metricName, value] of Object.entries(result)) {
+            if (!measurements[metricName]) {
+                measurements[metricName] = [];
+            }
+            measurements[metricName].push(value);
+        }
+    }
+    const smoother = new MetricSmoothing(config);
+    const filtered = smoother.filterWarmup(measurements);
+    const smoothed = smoother.smooth(filtered);
+    return {
+        smoothed,
+        raw: measurements,
+    };
+}
+//# sourceMappingURL=smoothing.js.map

package/dist/benchmark/smoothing.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"smoothing.js","sourceRoot":"","sources":["../../src/benchmark/smoothing.ts"],"names":[],"mappings":"AA0BA,MAAM,OAAO,eAAe;IAClB,MAAM,CAAkB;IAEhC,YAAY,MAAuB;QACjC,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IACvB,CAAC;IAED,YAAY,CACV,YAAsC;QAEtC,MAAM,QAAQ,GAA6B,EAAE,CAAC;QAE9C,KAAK,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;YAChE,IAAI,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,CAAC;gBAC3C,QAAQ,CAAC,UAAU,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;YAC9D,CAAC;iBAAM,CAAC;gBACN,QAAQ,CAAC,UAAU,CAAC,GAAG,MAAM,CAAC;YAChC,CAAC;QACH,CAAC;QAED,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED,cAAc,CAAC,MAAgB;QAC7B,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,KAAK,MAAM,EAAE,CAAC;YACzC,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,KAAK,KAAK,EAAE,CAAC;YACxC,OAAO,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC;QACxC,CAAC;aAAM,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,KAAK,QAAQ,EAAE,CAAC;YAClD,OAAO,IAAI,CAAC,oBAAoB,CAAC,MAAM,CAAC,CAAC;QAC3C,CAAC;QAED,OAAO,EAAE,CAAC;IACZ,CAAC;IAEO,iBAAiB,CAAC,MAAgB;QACxC,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACjD,MAAM,EAAE,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACvC,MAAM,EAAE,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACvC,MAAM,GAAG,GAAG,EAAE,GAAG,EAAE,CAAC;QAEpB,MAAM,UAAU,GAAG,EAAE,GAAG,IAAI,CAAC,MAAM,CAAC,aAAa,GAAG,GAAG,CAAC;QACxD,MAAM,UAAU,GAAG,EAAE,GAAG,IAAI,CAAC,MAAM,CAAC,aAAa,GAAG,GAAG,CAAC;QAExD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,UAAU,IAAI,CAAC,GAAG,UAAU,CAAC,CAAC;IAChE,CAAC;IAEO,oBAAoB,CAAC,MAAgB;QAC3C,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QACnE,MAAM,QAAQ,GACZ,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QAC5E,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAEnC,MAAM,SAAS,GAAG,CAAC,GAAG,MAAM,CAAC;QAC7B,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,SAAS,CAAC,CAAC;IAC9D,CAAC;IAEO,UAAU,CAAC,MAAgB,EAAE,CAAS;QAC5C,MAAM,KAAK,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC9C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QAChC,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,MAAM,GAAG,KAAK,GAAG,KAAK,CAAC;QAE7B,IAAI,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;YAC3B,OAAO,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACnC,CAAC;QAED,OAAO,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC;IAC/D,CAAC;IAED,MAAM,CAAC,YAAsC;QAC3C,MAAM,OAAO,GAA2B,EAAE,CAAC;QAE3C,KAAK,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;YAChE,MAAM,MAAM,GAAG,IAAI,CAAC,YAAY,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;YACrD,OAAO,CAAC,UAAU,CAAC,GAAG,MAAM,CAAC,aAAa,CAAC;QAC7C,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED,YAAY,CAAC,UAAkB,EAAE,MAAgB;QAC/C,MAAM,QAAQ,GAAG,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC;QAC7C,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;QAEnE,MAAM,KAAK,GAAG,IAAI,CAAC,cAAc,CAAC,cAAc,CAAC,CAAC;QAElD,OAAO;YACL,UAAU;YACV,cAAc,EAAE,MAAM;YACtB,aAAa,EAAE,KAAK,CAAC,MAAM;YAC3B,eAAe,EAAE,QAAQ;YACzB,KAAK;SACN,CAAC;IACJ,CAAC;IAEO,cAAc,CAAC,MAAgB;QACrC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACxB,OAAO;gBACL,IAAI,EAAE,CAAC;gBACP,MAAM,EAAE,CAAC;gBACT,MAAM,EAAE,CAAC;gBACT,GAAG,EAAE,CAAC;gBACN,GAAG,EAAE,CAAC;aACP,CAAC;QACJ,CAAC;QAED,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACjD,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;QAClD,MAAM,IAAI,GAAG,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC;QACjC,MAAM,QAAQ,GACZ,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QAE5E,OAAO;YACL,IAAI;YACJ,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC;YACnC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC;YAC3B,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC;YACd,GAAG,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;SAC/B,CAAC;IACJ,CAAC;CACF;AAED,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,MAAuB,EACvB,WAAkD;IAKlD,MAAM,YAAY,GAA6B,EAAE,CAAC;IAElD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,EAAE,CAAC,EAAE,EAAE,CAAC;QAC/D,MAAM,MAAM,GAAG,MAAM,WAAW,EAAE,CAAC;QAEnC,KAAK,MAAM,CAAC,UAAU,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;YACzD,IAAI,CAAC,YAAY,CAAC,UAAU,CAAC,EAAE,CAAC;gBAC9B,YAAY,CAAC,UAAU,CAAC,GAAG,EAAE,CAAC;YAChC,CAAC;YACD,YAAY,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACvC,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,eAAe,CAAC,MAAM,CAAC,CAAC;IAC7C,MAAM,QAAQ,GAAG,QAAQ,CAAC,YAAY,CAAC,YAAY,CAAC,CAAC;IACrD,MAAM,QAAQ,GAAG,QAAQ,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAE3C,OAAO;QACL,QAAQ;QACR,GAAG,EAAE,YAAY;KAClB,CAAC;AACJ,CAAC"}

package/dist/benchmark/threshold.d.ts

@@ -0,0 +1,64 @@
+export interface MetricThreshold {
+    minValue?: number;
+    maxMs?: number;
+    maxTokens?: number;
+    minPercent?: number;
+    trend: "higher-is-better" | "lower-is-better";
+    allowableIncreasePercent?: number;
+    allowableDecreasePercent?: number;
+}
+export interface ThresholdCategory {
+    [metricName: string]: MetricThreshold;
+}
+export interface BenchmarkThresholds {
+    version: string;
+    description: string;
+    thresholds: {
+        indexing: ThresholdCategory;
+        quality: ThresholdCategory;
+        performance: ThresholdCategory;
+        tokenEfficiency: ThresholdCategory;
+        coverage: ThresholdCategory;
+    };
+    smoothing: {
+        warmupRuns: number;
+        sampleRuns: number;
+        outlierMethod: "iqr" | "stddev" | "none";
+        iqrMultiplier: number;
+    };
+    baseline: {
+        filePath: string;
+        autoUpdateOnPass: boolean;
+    };
+}
+export interface ThresholdEvaluation {
+    metricName: string;
+    category: string;
+    currentValue: number;
+    baselineValue?: number;
+    threshold: MetricThreshold;
+    passed: boolean;
+    delta?: number;
+    deltaPercent?: number;
+    message: string;
+}
+export interface ThresholdEvaluationResult {
+    passed: boolean;
+    evaluations: ThresholdEvaluation[];
+    summary: {
+        total: number;
+        passed: number;
+        failed: number;
+        warnings: number;
+    };
+}
+export declare class ThresholdEvaluator {
+    private thresholds;
+    constructor(thresholds: BenchmarkThresholds);
+    evaluate(currentMetrics: Record<string, number>, baselineMetrics?: Record<string, number>): ThresholdEvaluationResult;
+    private evaluateMetric;
+}
+export declare function loadThresholdConfig(configPath: string): BenchmarkThresholds;
+export declare function loadBaselineMetrics(baselinePath: string): Record<string, number> | undefined;
+export declare function saveBaselineMetrics(baselinePath: string, results: unknown): void;
+//# sourceMappingURL=threshold.d.ts.map

package/dist/benchmark/threshold.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"threshold.d.ts","sourceRoot":"","sources":["../../src/benchmark/threshold.ts"],"names":[],"mappings":"AAGA,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,kBAAkB,GAAG,iBAAiB,CAAC;IAC9C,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC,wBAAwB,CAAC,EAAE,MAAM,CAAC;CACnC;AAED,MAAM,WAAW,iBAAiB;IAChC,CAAC,UAAU,EAAE,MAAM,GAAG,eAAe,CAAC;CACvC;AAED,MAAM,WAAW,mBAAmB;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE;QACV,QAAQ,EAAE,iBAAiB,CAAC;QAC5B,OAAO,EAAE,iBAAiB,CAAC;QAC3B,WAAW,EAAE,iBAAiB,CAAC;QAC/B,eAAe,EAAE,iBAAiB,CAAC;QACnC,QAAQ,EAAE,iBAAiB,CAAC;KAC7B,CAAC;IACF,SAAS,EAAE;QACT,UAAU,EAAE,MAAM,CAAC;QACnB,UAAU,EAAE,MAAM,CAAC;QACnB,aAAa,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;QACzC,aAAa,EAAE,MAAM,CAAC;KACvB,CAAC;IACF,QAAQ,EAAE;QACR,QAAQ,EAAE,MAAM,CAAC;QACjB,gBAAgB,EAAE,OAAO,CAAC;KAC3B,CAAC;CACH;AAED,MAAM,WAAW,mBAAmB;IAClC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,eAAe,CAAC;IAC3B,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,yBAAyB;IACxC,MAAM,EAAE,OAAO,CAAC;IAChB,WAAW,EAAE,mBAAmB,EAAE,CAAC;IACnC,OAAO,EAAE;QACP,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,QAAQ,EAAE,MAAM,CAAC;KAClB,CAAC;CACH;AAED,qBAAa,kBAAkB;IAC7B,OAAO,CAAC,UAAU,CAAsB;gBAE5B,UAAU,EAAE,mBAAmB;IAI3C,QAAQ,CACN,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EACtC,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GACvC,yBAAyB;IAuC5B,OAAO,CAAC,cAAc;CAyGvB;AAED,wBAAgB,mBAAmB,CAAC,UAAU,EAAE,MAAM,GAAG,mBAAmB,CAG3E;AAED,wBAAgB,mBAAmB,CACjC,YAAY,EAAE,MAAM,GACnB,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,SAAS,CAmDpC;AAED,wBAAgB,mBAAmB,CACjC,YAAY,EAAE,MAAM,EACpB,OAAO,EAAE,OAAO,GACf,IAAI,CAON"}