anantham-v2 2.0.0-alpha.1

package/dist/evaluation/evaluation-harness.js

@@ -0,0 +1,203 @@
+import { randomUUID } from "node:crypto";
+import { ProjectRepository } from "../persistence/repositories/project-repository.js";
+import { SessionRepository } from "../persistence/repositories/session-repository.js";
+import { EvaluationRunSchema, EvaluationCaseResultSchema, } from "../domain/evaluation.js";
+import { EvidenceCollector } from "./evidence-collector.js";
+import { AssertionEvaluator } from "./assertion-evaluator.js";
+/**
+ * Evaluation Harness.
+ * PRD Part 3 Section 92.
+ */
+export class EvaluationHarness {
+    engine;
+    eventStore;
+    projectRepo;
+    sessionRepo;
+    gitCommit;
+    runtimeVersion;
+    constructor(options) {
+        this.engine = options.engine;
+        this.eventStore = options.eventStore;
+        this.projectRepo = new ProjectRepository(options.engine);
+        this.sessionRepo = new SessionRepository(options.engine);
+        this.gitCommit = options.gitCommit ?? "de6bf08";
+        this.runtimeVersion = options.runtimeVersion ?? "2.0.0-alpha.1";
+    }
+    /**
+     * Execute an evaluation run over a benchmark dataset.
+     */
+    async executeRun(dataset, executors) {
+        const runId = `eval_run_${randomUUID().slice(0, 8)}`;
+        const startedAt = new Date().toISOString();
+        const caseResults = [];
+        // Create isolated project & session for the evaluation run
+        const evalProjectId = `eval_proj_${runId}`;
+        const evalSessionId = `eval_sess_${runId}`;
+        this.projectRepo.save({
+            id: evalProjectId,
+            name: `Evaluation Project (${runId})`,
+            rootPath: `/tmp/eval/${runId}`,
+            status: "active",
+            tags: ["evaluation"],
+            modelProfile: "default",
+            memoryNamespace: "eval",
+            orchestrationProfile: "default",
+            trustProfile: "safe",
+            createdAt: startedAt,
+            lastOpenedAt: startedAt,
+            lastActivityAt: startedAt,
+            metadata: { runId },
+        });
+        this.sessionRepo.save({
+            id: evalSessionId,
+            projectId: evalProjectId,
+            name: `Evaluation Session (${runId})`,
+            branch: "main",
+            status: "active",
+            modelProfile: "default",
+            keyPoolProfile: "default",
+            mode: "interactive",
+            permissions: {},
+            createdAt: startedAt,
+            updatedAt: startedAt,
+            metadata: { runId },
+        });
+        try {
+            for (const benchCase of dataset.cases) {
+                const result = await this.executeCase(benchCase, {
+                    projectId: evalProjectId,
+                    sessionId: evalSessionId,
+                    executor: executors?.[benchCase.caseId],
+                });
+                caseResults.push(result);
+            }
+        }
+        finally {
+            // Mark evaluation session and project completed
+            try {
+                this.sessionRepo.updateStatus(evalSessionId, "archived");
+                this.projectRepo.updateStatus(evalProjectId, "archived");
+            }
+            catch {
+                // Ignore archive error in cleanup
+            }
+        }
+        const completedAt = new Date().toISOString();
+        const passedCases = caseResults.filter((r) => r.status === "PASS").length;
+        const failedCases = caseResults.filter((r) => r.status === "FAIL").length;
+        const partialCases = caseResults.filter((r) => r.status === "PARTIAL").length;
+        const inconclusiveCases = caseResults.filter((r) => r.status === "INCONCLUSIVE").length;
+        const totalCases = caseResults.length;
+        const totalScore = caseResults.reduce((acc, r) => acc + r.score, 0);
+        const overallScore = totalCases > 0 ? Math.round((totalScore / totalCases) * 100) / 100 : 0;
+        return EvaluationRunSchema.parse({
+            id: runId,
+            datasetId: dataset.datasetId,
+            datasetVersion: dataset.version,
+            status: failedCases === 0 ? "COMPLETED" : "FAILED",
+            summary: {
+                totalCases,
+                passedCases,
+                failedCases,
+                partialCases,
+                inconclusiveCases,
+                overallScore,
+            },
+            results: caseResults,
+            provenance: {
+                runtimeVersion: this.runtimeVersion,
+                gitCommit: this.gitCommit,
+                datasetId: dataset.datasetId,
+                datasetVersion: dataset.version,
+                modelProfile: "default",
+                environment: "node-test",
+            },
+            createdAt: startedAt,
+            completedAt,
+        });
+    }
+    /**
+     * Execute an individual benchmark case.
+     */
+    async executeCase(benchCase, options) {
+        const caseStart = Date.now();
+        const startedAt = new Date().toISOString();
+        const evidenceCollector = new EvidenceCollector();
+        // Listen to EventStore for this session/project
+        const unsubscribe = this.eventStore.subscribe({ projectId: options.projectId }, (event) => {
+            evidenceCollector.recordEvent(event);
+        });
+        let failureClassification = "NONE";
+        try {
+            if (options.executor) {
+                await options.executor(benchCase, {
+                    projectId: options.projectId,
+                    sessionId: options.sessionId,
+                    evidenceCollector,
+                    engine: this.engine,
+                    eventStore: this.eventStore,
+                });
+            }
+        }
+        catch (err) {
+            const msg = (err.message ?? "").toLowerCase();
+            if (msg.includes("timeout")) {
+                failureClassification = "TIMEOUT";
+            }
+            else if (msg.includes("policy") || msg.includes("denied")) {
+                failureClassification = "POLICY";
+            }
+            else if (msg.includes("security") || msg.includes("forbidden")) {
+                failureClassification = "SECURITY";
+            }
+            else {
+                failureClassification = "RUNTIME";
+            }
+        }
+        finally {
+            unsubscribe();
+        }
+        const durationMs = Date.now() - caseStart;
+        const completedAt = new Date().toISOString();
+        const evidence = evidenceCollector.getEvidence();
+        // Evaluate assertions with physical engine verification
+        const assertionResults = benchCase.assertions.map((a) => AssertionEvaluator.evaluate(a, evidence, { engine: this.engine }));
+        const criticalFailed = assertionResults.some((r) => r.criticality === "CRITICAL" && !r.passed);
+        const optionalFailed = assertionResults.some((r) => r.criticality === "OPTIONAL" && !r.passed);
+        const allPassed = assertionResults.every((r) => r.passed);
+        let status = "PASS";
+        let score = 100;
+        if (criticalFailed) {
+            status = "FAIL";
+            score = 0;
+            if (failureClassification === "NONE") {
+                failureClassification = "ASSERTION";
+            }
+        }
+        else if (optionalFailed) {
+            status = "PARTIAL";
+            const passedCount = assertionResults.filter((r) => r.passed).length;
+            score = Math.round((passedCount / assertionResults.length) * 100);
+        }
+        else if (!allPassed) {
+            status = "INCONCLUSIVE";
+            score = 50;
+        }
+        return EvaluationCaseResultSchema.parse({
+            caseId: benchCase.caseId,
+            status,
+            score,
+            assertionResults,
+            evidence: {
+                eventsCount: evidence.events.length,
+                policyDecisions: evidence.policyDecisions,
+                stateKeys: Object.keys(evidence.stateSnapshots),
+            },
+            failureClassification,
+            durationMs,
+            startedAt,
+            completedAt,
+        });
+    }
+}
+//# sourceMappingURL=evaluation-harness.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-harness.js","sourceRoot":"","sources":["../../src/evaluation/evaluation-harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,OAAO,EAAE,iBAAiB,EAAE,MAAM,mDAAmD,CAAC;AACtF,OAAO,EAAE,iBAAiB,EAAE,MAAM,mDAAmD,CAAC;AACtF,OAAO,EAKL,mBAAmB,EACnB,0BAA0B,GAC3B,MAAM,yBAAyB,CAAC;AACjC,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAoB9D;;;GAGG;AACH,MAAM,OAAO,iBAAiB;IACX,MAAM,CAAe;IACrB,UAAU,CAAa;IACvB,WAAW,CAAoB;IAC/B,WAAW,CAAoB;IAC/B,SAAS,CAAS;IAClB,cAAc,CAAS;IAExC,YAAY,OAAiC;QAC3C,IAAI,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;QAC7B,IAAI,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC;QACrC,IAAI,CAAC,WAAW,GAAG,IAAI,iBAAiB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QACzD,IAAI,CAAC,WAAW,GAAG,IAAI,iBAAiB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QACzD,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,SAAS,CAAC;QAChD,IAAI,CAAC,cAAc,GAAG,OAAO,CAAC,cAAc,IAAI,eAAe,CAAC;IAClE,CAAC;IAED;;OAEG;IACI,KAAK,CAAC,UAAU,CACrB,OAAyB,EACzB,SAAiD;QAEjD,MAAM,KAAK,GAAG,YAAY,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;QACrD,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC3C,MAAM,WAAW,GAA2B,EAAE,CAAC;QAE/C,2DAA2D;QAC3D,MAAM,aAAa,GAAG,aAAa,KAAK,EAAE,CAAC;QAC3C,MAAM,aAAa,GAAG,aAAa,KAAK,EAAE,CAAC;QAE3C,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC;YACpB,EAAE,EAAE,aAAa;YACjB,IAAI,EAAE,uBAAuB,KAAK,GAAG;YACrC,QAAQ,EAAE,aAAa,KAAK,EAAE;YAC9B,MAAM,EAAE,QAAQ;YAChB,IAAI,EAAE,CAAC,YAAY,CAAC;YACpB,YAAY,EAAE,SAAS;YACvB,eAAe,EAAE,MAAM;YACvB,oBAAoB,EAAE,SAAS;YAC/B,YAAY,EAAE,MAAM;YACpB,SAAS,EAAE,SAAS;YACpB,YAAY,EAAE,SAAS;YACvB,cAAc,EAAE,SAAS;YACzB,QAAQ,EAAE,EAAE,KAAK,EAAE;SACpB,CAAC,CAAC;QAEH,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC;YACpB,EAAE,EAAE,aAAa;YACjB,SAAS,EAAE,aAAa;YACxB,IAAI,EAAE,uBAAuB,KAAK,GAAG;YACrC,MAAM,EAAE,MAAM;YACd,MAAM,EAAE,QAAQ;YAChB,YAAY,EAAE,SAAS;YACvB,cAAc,EAAE,SAAS;YACzB,IAAI,EAAE,aAAa;YACnB,WAAW,EAAE,EAAE;YACf,SAAS,EAAE,SAAS;YACpB,SAAS,EAAE,SAAS;YACpB,QAAQ,EAAE,EAAE,KAAK,EAAE;SACpB,CAAC,CAAC;QAEH,IAAI,CAAC;YACH,KAAK,MAAM,SAAS,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;gBACtC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,WAAW,CAAC,SAAS,EAAE;oBAC/C,SAAS,EAAE,aAAa;oBACxB,SAAS,EAAE,aAAa;oBACxB,QAAQ,EAAE,SAAS,EAAE,CAAC,SAAS,CAAC,MAAM,CAAC;iBACxC,CAAC,CAAC;gBACH,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YAC3B,CAAC;QACH,CAAC;gBAAS,CAAC;YACT,gDAAgD;YAChD,IAAI,CAAC;gBACH,IAAI,CAAC,WAAW,CAAC,YAAY,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;gBACzD,IAAI,CAAC,WAAW,CAAC,YAAY,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;YAC3D,CAAC;YAAC,MAAM,CAAC;gBACP,kCAAkC;YACpC,CAAC;QACH,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC7C,MAAM,WAAW,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;QAC1E,MAAM,WAAW,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;QAC1E,MAAM,YAAY,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;QAC9E,MAAM,iBAAiB,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,cAAc,CAAC,CAAC,MAAM,CAAC;QACxF,MAAM,UAAU,GAAG,WAAW,CAAC,MAAM,CAAC;QAEtC,MAAM,UAAU,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;QACpE,MAAM,YAAY,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,UAAU,GAAG,UAAU,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QAE5F,OAAO,mBAAmB,CAAC,KAAK,CAAC;YAC/B,EAAE,EAAE,KAAK;YACT,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,cAAc,EAAE,OAAO,CAAC,OAAO;YAC/B,MAAM,EAAE,WAAW,KAAK,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ;YAClD,OAAO,EAAE;gBACP,UAAU;gBACV,WAAW;gBACX,WAAW;gBACX,YAAY;gBACZ,iBAAiB;gBACjB,YAAY;aACb;YACD,OAAO,EAAE,WAAW;YACpB,UAAU,EAAE;gBACV,cAAc,EAAE,IAAI,CAAC,cAAc;gBACnC,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,cAAc,EAAE,OAAO,CAAC,OAAO;gBAC/B,YAAY,EAAE,SAAS;gBACvB,WAAW,EAAE,WAAW;aACzB;YACD,SAAS,EAAE,SAAS;YACpB,WAAW;SACZ,CAAC,CAAC;IACL,CAAC;IAED;;OAEG;IACI,KAAK,CAAC,WAAW,CACtB,SAAwB,EACxB,OAAmF;QAEnF,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC3C,MAAM,iBAAiB,GAAG,IAAI,iBAAiB,EAAE,CAAC;QAElD,gDAAgD;QAChD,MAAM,WAAW,GAAG,IAAI,CAAC,UAAU,CAAC,SAAS,CAC3C,EAAE,SAAS,EAAE,OAAO,CAAC,SAAS,EAAE,EAChC,CAAC,KAAK,EAAE,EAAE;YACR,iBAAiB,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QACvC,CAAC,CACF,CAAC;QAEF,IAAI,qBAAqB,GAAkD,MAAM,CAAC;QAElF,IAAI,CAAC;YACH,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACrB,MAAM,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE;oBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,iBAAiB;oBACjB,MAAM,EAAE,IAAI,CAAC,MAAM;oBACnB,UAAU,EAAE,IAAI,CAAC,UAAU;iBAC5B,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAAC,OAAO,GAAQ,EAAE,CAAC;YAClB,MAAM,GAAG,GAAG,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;YAC9C,IAAI,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC5B,qBAAqB,GAAG,SAAS,CAAC;YACpC,CAAC;iBAAM,IAAI,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC5D,qBAAqB,GAAG,QAAQ,CAAC;YACnC,CAAC;iBAAM,IAAI,GAAG,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;gBACjE,qBAAqB,GAAG,UAAU,CAAC;YACrC,CAAC;iBAAM,CAAC;gBACN,qBAAqB,GAAG,SAAS,CAAC;YACpC,CAAC;QACH,CAAC;gBAAS,CAAC;YACT,WAAW,EAAE,CAAC;QAChB,CAAC;QAED,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QAC1C,MAAM,WAAW,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC7C,MAAM,QAAQ,GAAG,iBAAiB,CAAC,WAAW,EAAE,CAAC;QAEjD,wDAAwD;QACxD,MAAM,gBAAgB,GAAG,SAAS,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CACtD,kBAAkB,CAAC,QAAQ,CAAC,CAAC,EAAE,QAAQ,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC,CAClE,CAAC;QAGF,MAAM,cAAc,GAAG,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,KAAK,UAAU,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAC/F,MAAM,cAAc,GAAG,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,KAAK,UAAU,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAC/F,MAAM,SAAS,GAAG,gBAAgB,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAE1D,IAAI,MAAM,GAAmC,MAAM,CAAC;QACpD,IAAI,KAAK,GAAG,GAAG,CAAC;QAEhB,IAAI,cAAc,EAAE,CAAC;YACnB,MAAM,GAAG,MAAM,CAAC;YAChB,KAAK,GAAG,CAAC,CAAC;YACV,IAAI,qBAAqB,KAAK,MAAM,EAAE,CAAC;gBACrC,qBAAqB,GAAG,WAAW,CAAC;YACtC,CAAC;QACH,CAAC;aAAM,IAAI,cAAc,EAAE,CAAC;YAC1B,MAAM,GAAG,SAAS,CAAC;YACnB,MAAM,WAAW,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;YACpE,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,WAAW,GAAG,gBAAgB,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC;QACpE,CAAC;aAAM,IAAI,CAAC,SAAS,EAAE,CAAC;YACtB,MAAM,GAAG,cAAc,CAAC;YACxB,KAAK,GAAG,EAAE,CAAC;QACb,CAAC;QAED,OAAO,0BAA0B,CAAC,KAAK,CAAC;YACtC,MAAM,EAAE,SAAS,CAAC,MAAM;YACxB,MAAM;YACN,KAAK;YACL,gBAAgB;YAChB,QAAQ,EAAE;gBACR,WAAW,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM;gBACnC,eAAe,EAAE,QAAQ,CAAC,eAAe;gBACzC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,cAAc,CAAC;aAChD;YACD,qBAAqB;YACrB,UAAU;YACV,SAAS;YACT,WAAW;SACZ,CAAC,CAAC;IACL,CAAC;CACF"}

package/dist/evaluation/evaluation-manager.d.ts

@@ -0,0 +1,34 @@
+import { type SqliteEngine } from "../persistence/sqlite-engine.js";
+import { type EventStore } from "../event-state/event-store.js";
+import { EvaluationRepository } from "../persistence/repositories/evaluation-repository.js";
+import { BenchmarkRegistry } from "./benchmark-registry.js";
+import { EvaluationHarness, type BenchmarkCaseExecutor } from "./evaluation-harness.js";
+import { type EvaluationRun, type EvaluationReport } from "../domain/evaluation.js";
+export interface EvaluationManagerOptions {
+    engine: SqliteEngine;
+    eventStore: EventStore;
+    gitCommit?: string;
+    runtimeVersion?: string;
+}
+/**
+ * Central Evaluation Manager.
+ * PRD Part 3 Section 80–120.
+ */
+export declare class EvaluationManager {
+    readonly registry: BenchmarkRegistry;
+    readonly repository: EvaluationRepository;
+    readonly harness: EvaluationHarness;
+    constructor(options: EvaluationManagerOptions);
+    /**
+     * Run a benchmark dataset, persist the results, and return the EvaluationRun.
+     */
+    runEvaluation(datasetId: string, options?: {
+        version?: string;
+        executors?: Record<string, BenchmarkCaseExecutor>;
+    }): Promise<EvaluationRun>;
+    /**
+     * Generate an evaluation report, optionally comparing against a baseline run.
+     */
+    generateReport(runId: string, baselineRunId?: string): EvaluationReport;
+}
+//# sourceMappingURL=evaluation-manager.d.ts.map

package/dist/evaluation/evaluation-manager.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-manager.d.ts","sourceRoot":"","sources":["../../src/evaluation/evaluation-manager.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,YAAY,EAAE,MAAM,iCAAiC,CAAC;AACpE,OAAO,EAAE,KAAK,UAAU,EAAE,MAAM,+BAA+B,CAAC;AAChE,OAAO,EAAE,oBAAoB,EAAE,MAAM,sDAAsD,CAAC;AAC5F,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,iBAAiB,EAAE,KAAK,qBAAqB,EAAE,MAAM,yBAAyB,CAAC;AAExF,OAAO,EACL,KAAK,aAAa,EAClB,KAAK,gBAAgB,EAEtB,MAAM,yBAAyB,CAAC;AAEjC,MAAM,WAAW,wBAAwB;IACvC,MAAM,EAAE,YAAY,CAAC;IACrB,UAAU,EAAE,UAAU,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED;;;GAGG;AACH,qBAAa,iBAAiB;IAC5B,SAAgB,QAAQ,EAAE,iBAAiB,CAAC;IAC5C,SAAgB,UAAU,EAAE,oBAAoB,CAAC;IACjD,SAAgB,OAAO,EAAE,iBAAiB,CAAC;gBAE/B,OAAO,EAAE,wBAAwB;IAW7C;;OAEG;IACU,aAAa,CACxB,SAAS,EAAE,MAAM,EACjB,OAAO,CAAC,EAAE;QACR,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAC;KACnD,GACA,OAAO,CAAC,aAAa,CAAC;IAWzB;;OAEG;IACI,cAAc,CAAC,KAAK,EAAE,MAAM,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,gBAAgB;CAsB/E"}

package/dist/evaluation/evaluation-manager.js

@@ -0,0 +1,61 @@
+import { randomUUID } from "node:crypto";
+import { EvaluationRepository } from "../persistence/repositories/evaluation-repository.js";
+import { BenchmarkRegistry } from "./benchmark-registry.js";
+import { EvaluationHarness } from "./evaluation-harness.js";
+import { RegressionEngine } from "./regression-engine.js";
+import { EvaluationReportSchema, } from "../domain/evaluation.js";
+/**
+ * Central Evaluation Manager.
+ * PRD Part 3 Section 80–120.
+ */
+export class EvaluationManager {
+    registry;
+    repository;
+    harness;
+    constructor(options) {
+        this.registry = new BenchmarkRegistry();
+        this.repository = new EvaluationRepository(options.engine);
+        this.harness = new EvaluationHarness({
+            engine: options.engine,
+            eventStore: options.eventStore,
+            gitCommit: options.gitCommit,
+            runtimeVersion: options.runtimeVersion,
+        });
+    }
+    /**
+     * Run a benchmark dataset, persist the results, and return the EvaluationRun.
+     */
+    async runEvaluation(datasetId, options) {
+        const dataset = this.registry.getDataset(datasetId, options?.version);
+        if (!dataset) {
+            throw new Error(`Benchmark dataset '${datasetId}' not found in registry.`);
+        }
+        const run = await this.harness.executeRun(dataset, options?.executors);
+        this.repository.saveRun(run);
+        return run;
+    }
+    /**
+     * Generate an evaluation report, optionally comparing against a baseline run.
+     */
+    generateReport(runId, baselineRunId) {
+        const run = this.repository.findRunById(runId);
+        if (!run) {
+            throw new Error(`Evaluation run '${runId}' not found.`);
+        }
+        let regression = undefined;
+        if (baselineRunId) {
+            const baseline = this.repository.findRunById(baselineRunId);
+            if (baseline) {
+                regression = RegressionEngine.compare(baseline, run);
+            }
+        }
+        return EvaluationReportSchema.parse({
+            reportId: `eval_rep_${randomUUID().slice(0, 8)}`,
+            runId,
+            generatedAt: new Date().toISOString(),
+            run,
+            regression,
+        });
+    }
+}
+//# sourceMappingURL=evaluation-manager.js.map

package/dist/evaluation/evaluation-manager.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-manager.js","sourceRoot":"","sources":["../../src/evaluation/evaluation-manager.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGzC,OAAO,EAAE,oBAAoB,EAAE,MAAM,sDAAsD,CAAC;AAC5F,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,iBAAiB,EAA8B,MAAM,yBAAyB,CAAC;AACxF,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAGL,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AASjC;;;GAGG;AACH,MAAM,OAAO,iBAAiB;IACZ,QAAQ,CAAoB;IAC5B,UAAU,CAAuB;IACjC,OAAO,CAAoB;IAE3C,YAAY,OAAiC;QAC3C,IAAI,CAAC,QAAQ,GAAG,IAAI,iBAAiB,EAAE,CAAC;QACxC,IAAI,CAAC,UAAU,GAAG,IAAI,oBAAoB,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC3D,IAAI,CAAC,OAAO,GAAG,IAAI,iBAAiB,CAAC;YACnC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,cAAc,EAAE,OAAO,CAAC,cAAc;SACvC,CAAC,CAAC;IACL,CAAC;IAED;;OAEG;IACI,KAAK,CAAC,aAAa,CACxB,SAAiB,EACjB,OAGC;QAED,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;QACtE,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,MAAM,IAAI,KAAK,CAAC,sBAAsB,SAAS,0BAA0B,CAAC,CAAC;QAC7E,CAAC;QAED,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;QACvE,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QAC7B,OAAO,GAAG,CAAC;IACb,CAAC;IAED;;OAEG;IACI,cAAc,CAAC,KAAa,EAAE,aAAsB;QACzD,MAAM,GAAG,GAAG,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAC/C,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,MAAM,IAAI,KAAK,CAAC,mBAAmB,KAAK,cAAc,CAAC,CAAC;QAC1D,CAAC;QAED,IAAI,UAAU,GAAG,SAAS,CAAC;QAC3B,IAAI,aAAa,EAAE,CAAC;YAClB,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,aAAa,CAAC,CAAC;YAC5D,IAAI,QAAQ,EAAE,CAAC;gBACb,UAAU,GAAG,gBAAgB,CAAC,OAAO,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;YACvD,CAAC;QACH,CAAC;QAED,OAAO,sBAAsB,CAAC,KAAK,CAAC;YAClC,QAAQ,EAAE,YAAY,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE;YAChD,KAAK;YACL,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACrC,GAAG;YACH,UAAU;SACX,CAAC,CAAC;IACL,CAAC;CACF"}

package/dist/evaluation/evidence-collector.d.ts

@@ -0,0 +1,35 @@
+import { type HarnessEvent } from "../domain/event.js";
+export interface CollectedEvidence {
+    events: HarnessEvent[];
+    stateSnapshots: Record<string, unknown>;
+    policyDecisions: string[];
+    artifacts: Array<{
+        id: string;
+        path: string;
+        hash?: string;
+    }>;
+    metrics: Record<string, number>;
+}
+/**
+ * Evidence Collector.
+ * Gathers objective, machine-verifiable runtime evidence during benchmark execution.
+ */
+export declare class EvidenceCollector {
+    private readonly events;
+    private readonly stateSnapshots;
+    private readonly policyDecisions;
+    private readonly artifacts;
+    private readonly metrics;
+    recordEvent(event: HarnessEvent): void;
+    recordState(key: string, value: unknown): void;
+    recordPolicyDecision(decision: string): void;
+    recordArtifact(artifact: {
+        id: string;
+        path: string;
+        hash?: string;
+    }): void;
+    recordMetric(name: string, value: number): void;
+    getEvidence(): CollectedEvidence;
+    clear(): void;
+}
+//# sourceMappingURL=evidence-collector.d.ts.map

package/dist/evaluation/evidence-collector.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evidence-collector.d.ts","sourceRoot":"","sources":["../../src/evaluation/evidence-collector.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAEvD,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,YAAY,EAAE,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACxC,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,SAAS,EAAE,KAAK,CAAC;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC9D,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjC;AAED;;;GAGG;AACH,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAsB;IAC7C,OAAO,CAAC,QAAQ,CAAC,cAAc,CAA+B;IAC9D,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAgB;IAChD,OAAO,CAAC,QAAQ,CAAC,SAAS,CAA0D;IACpF,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA8B;IAE/C,WAAW,CAAC,KAAK,EAAE,YAAY,GAAG,IAAI;IAItC,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,GAAG,IAAI;IAI9C,oBAAoB,CAAC,QAAQ,EAAE,MAAM,GAAG,IAAI;IAI5C,cAAc,CAAC,QAAQ,EAAE;QAAE,EAAE,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI;IAI3E,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI;IAI/C,WAAW,IAAI,iBAAiB;IAUhC,KAAK,IAAI,IAAI;CAWrB"}

package/dist/evaluation/evidence-collector.js

@@ -0,0 +1,47 @@
+/**
+ * Evidence Collector.
+ * Gathers objective, machine-verifiable runtime evidence during benchmark execution.
+ */
+export class EvidenceCollector {
+    events = [];
+    stateSnapshots = {};
+    policyDecisions = [];
+    artifacts = [];
+    metrics = {};
+    recordEvent(event) {
+        this.events.push(event);
+    }
+    recordState(key, value) {
+        this.stateSnapshots[key] = value;
+    }
+    recordPolicyDecision(decision) {
+        this.policyDecisions.push(decision);
+    }
+    recordArtifact(artifact) {
+        this.artifacts.push(artifact);
+    }
+    recordMetric(name, value) {
+        this.metrics[name] = value;
+    }
+    getEvidence() {
+        return {
+            events: [...this.events],
+            stateSnapshots: { ...this.stateSnapshots },
+            policyDecisions: [...this.policyDecisions],
+            artifacts: [...this.artifacts],
+            metrics: { ...this.metrics },
+        };
+    }
+    clear() {
+        this.events.length = 0;
+        this.policyDecisions.length = 0;
+        this.artifacts.length = 0;
+        for (const k of Object.keys(this.stateSnapshots)) {
+            delete this.stateSnapshots[k];
+        }
+        for (const k of Object.keys(this.metrics)) {
+            delete this.metrics[k];
+        }
+    }
+}
+//# sourceMappingURL=evidence-collector.js.map

package/dist/evaluation/evidence-collector.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evidence-collector.js","sourceRoot":"","sources":["../../src/evaluation/evidence-collector.ts"],"names":[],"mappings":"AAUA;;;GAGG;AACH,MAAM,OAAO,iBAAiB;IACX,MAAM,GAAmB,EAAE,CAAC;IAC5B,cAAc,GAA4B,EAAE,CAAC;IAC7C,eAAe,GAAa,EAAE,CAAC;IAC/B,SAAS,GAAuD,EAAE,CAAC;IACnE,OAAO,GAA2B,EAAE,CAAC;IAE/C,WAAW,CAAC,KAAmB;QACpC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC1B,CAAC;IAEM,WAAW,CAAC,GAAW,EAAE,KAAc;QAC5C,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC;IACnC,CAAC;IAEM,oBAAoB,CAAC,QAAgB;QAC1C,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACtC,CAAC;IAEM,cAAc,CAAC,QAAqD;QACzE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAChC,CAAC;IAEM,YAAY,CAAC,IAAY,EAAE,KAAa;QAC7C,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC;IAC7B,CAAC;IAEM,WAAW;QAChB,OAAO;YACL,MAAM,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;YACxB,cAAc,EAAE,EAAE,GAAG,IAAI,CAAC,cAAc,EAAE;YAC1C,eAAe,EAAE,CAAC,GAAG,IAAI,CAAC,eAAe,CAAC;YAC1C,SAAS,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC;YAC9B,OAAO,EAAE,EAAE,GAAG,IAAI,CAAC,OAAO,EAAE;SAC7B,CAAC;IACJ,CAAC;IAEM,KAAK;QACV,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;QACvB,IAAI,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,CAAC;QAChC,IAAI,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC;QAC1B,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,cAAc,CAAC,EAAE,CAAC;YACjD,OAAO,IAAI,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC;QAChC,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;YAC1C,OAAO,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,CAAC;IACH,CAAC;CACF"}

package/dist/evaluation/index.d.ts

@@ -0,0 +1,11 @@
+/**
+ * Anantham V2 — Benchmarking, Evaluation & Quality Harness Subsystem
+ * PRD Part 3 Section 80–120 / P9.1
+ */
+export * from "./benchmark-registry.js";
+export * from "./evidence-collector.js";
+export * from "./assertion-evaluator.js";
+export * from "./regression-engine.js";
+export * from "./evaluation-harness.js";
+export * from "./evaluation-manager.js";
+//# sourceMappingURL=index.d.ts.map

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

@@ -0,0 +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"}

package/dist/evaluation/index.js

@@ -0,0 +1,11 @@
+/**
+ * Anantham V2 — Benchmarking, Evaluation & Quality Harness Subsystem
+ * PRD Part 3 Section 80–120 / P9.1
+ */
+export * from "./benchmark-registry.js";
+export * from "./evidence-collector.js";
+export * from "./assertion-evaluator.js";
+export * from "./regression-engine.js";
+export * from "./evaluation-harness.js";
+export * from "./evaluation-manager.js";
+//# sourceMappingURL=index.js.map

package/dist/evaluation/index.js.map

@@ -0,0 +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"}

package/dist/evaluation/regression-engine.d.ts

@@ -0,0 +1,12 @@
+import { type EvaluationRun, type RegressionComparison } from "../domain/evaluation.js";
+/**
+ * Regression Comparison Engine.
+ * PRD Part 3 Section 98.
+ */
+export declare class RegressionEngine {
+    /**
+     * Compare a current evaluation run against a prior baseline run.
+     */
+    static compare(baseline: EvaluationRun, current: EvaluationRun): RegressionComparison;
+}
+//# sourceMappingURL=regression-engine.d.ts.map

package/dist/evaluation/regression-engine.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"regression-engine.d.ts","sourceRoot":"","sources":["../../src/evaluation/regression-engine.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,aAAa,EAClB,KAAK,oBAAoB,EAE1B,MAAM,yBAAyB,CAAC;AAEjC;;;GAGG;AACH,qBAAa,gBAAgB;IAC3B;;OAEG;WACW,OAAO,CACnB,QAAQ,EAAE,aAAa,EACvB,OAAO,EAAE,aAAa,GACrB,oBAAoB;CA0CxB"}

package/dist/evaluation/regression-engine.js

@@ -0,0 +1,48 @@
+import { RegressionComparisonSchema, } from "../domain/evaluation.js";
+/**
+ * Regression Comparison Engine.
+ * PRD Part 3 Section 98.
+ */
+export class RegressionEngine {
+    /**
+     * Compare a current evaluation run against a prior baseline run.
+     */
+    static compare(baseline, current) {
+        if (baseline.datasetId !== current.datasetId) {
+            throw new Error(`Cannot compare runs across different datasets: '${baseline.datasetId}' vs '${current.datasetId}'`);
+        }
+        const baselineCaseMap = new Map(baseline.results.map((r) => [r.caseId, r]));
+        const newFailures = [];
+        const fixedFailures = [];
+        const unchangedFailures = [];
+        for (const currRes of current.results) {
+            const baseRes = baselineCaseMap.get(currRes.caseId);
+            if (!baseRes)
+                continue;
+            const basePassed = baseRes.status === "PASS";
+            const currPassed = currRes.status === "PASS";
+            if (basePassed && !currPassed) {
+                newFailures.push(currRes.caseId);
+            }
+            else if (!basePassed && currPassed) {
+                fixedFailures.push(currRes.caseId);
+            }
+            else if (!basePassed && !currPassed) {
+                unchangedFailures.push(currRes.caseId);
+            }
+        }
+        const scoreDelta = Math.round((current.summary.overallScore - baseline.summary.overallScore) * 100) / 100;
+        const regressionDetected = newFailures.length > 0 || scoreDelta < 0;
+        return RegressionComparisonSchema.parse({
+            baselineRunId: baseline.id,
+            currentRunId: current.id,
+            datasetId: current.datasetId,
+            scoreDelta,
+            newFailures,
+            fixedFailures,
+            unchangedFailures,
+            regressionDetected,
+        });
+    }
+}
+//# sourceMappingURL=regression-engine.js.map

package/dist/evaluation/regression-engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"regression-engine.js","sourceRoot":"","sources":["../../src/evaluation/regression-engine.ts"],"names":[],"mappings":"AAAA,OAAO,EAGL,0BAA0B,GAC3B,MAAM,yBAAyB,CAAC;AAEjC;;;GAGG;AACH,MAAM,OAAO,gBAAgB;IAC3B;;OAEG;IACI,MAAM,CAAC,OAAO,CACnB,QAAuB,EACvB,OAAsB;QAEtB,IAAI,QAAQ,CAAC,SAAS,KAAK,OAAO,CAAC,SAAS,EAAE,CAAC;YAC7C,MAAM,IAAI,KAAK,CACb,mDAAmD,QAAQ,CAAC,SAAS,SAAS,OAAO,CAAC,SAAS,GAAG,CACnG,CAAC;QACJ,CAAC;QAED,MAAM,eAAe,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAC5E,MAAM,WAAW,GAAa,EAAE,CAAC;QACjC,MAAM,aAAa,GAAa,EAAE,CAAC;QACnC,MAAM,iBAAiB,GAAa,EAAE,CAAC;QAEvC,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;YACtC,MAAM,OAAO,GAAG,eAAe,CAAC,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACpD,IAAI,CAAC,OAAO;gBAAE,SAAS;YAEvB,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,KAAK,MAAM,CAAC;YAC7C,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,KAAK,MAAM,CAAC;YAE7C,IAAI,UAAU,IAAI,CAAC,UAAU,EAAE,CAAC;gBAC9B,WAAW,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACnC,CAAC;iBAAM,IAAI,CAAC,UAAU,IAAI,UAAU,EAAE,CAAC;gBACrC,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACrC,CAAC;iBAAM,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,EAAE,CAAC;gBACtC,iBAAiB,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACzC,CAAC;QACH,CAAC;QAED,MAAM,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,YAAY,GAAG,QAAQ,CAAC,OAAO,CAAC,YAAY,CAAC,GAAG,GAAG,CAAC,GAAG,GAAG,CAAC;QAC1G,MAAM,kBAAkB,GAAG,WAAW,CAAC,MAAM,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,CAAC;QAEpE,OAAO,0BAA0B,CAAC,KAAK,CAAC;YACtC,aAAa,EAAE,QAAQ,CAAC,EAAE;YAC1B,YAAY,EAAE,OAAO,CAAC,EAAE;YACxB,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,UAAU;YACV,WAAW;YACX,aAAa;YACb,iBAAiB;YACjB,kBAAkB;SACnB,CAAC,CAAC;IACL,CAAC;CACF"}

package/dist/event-state/event-store.d.ts

@@ -0,0 +1,63 @@
+import type { HarnessEvent } from "../domain/event.js";
+import type { SqliteEngine } from "../persistence/sqlite-engine.js";
+export interface StreamQueryOptions {
+    type?: string;
+    fromTimestamp?: string;
+    limit?: number;
+    offset?: number;
+}
+export interface EventSubscriptionFilter {
+    sessionId?: string;
+    projectId?: string;
+    type?: string;
+}
+export type EventListener = (event: Readonly<HarnessEvent>) => void | Promise<void>;
+/**
+ * EventStore manages the append-only authoritative event stream,
+ * provides query operations, and distributes ephemeral notifications to subscribers.
+ * PRD Part 1 Section 37-40 / Tech Stack Section 5.
+ */
+export declare class EventStore {
+    private readonly eventRepo;
+    private readonly engine;
+    private subscriptions;
+    private nextSubscriptionId;
+    constructor(engine: SqliteEngine);
+    get sqliteEngine(): SqliteEngine;
+    /**
+     * Appends an immutable event within a durable transaction and notifies active subscribers.
+     * Section 40: "Once committed, an authoritative event cannot be edited in place."
+     */
+    append(event: HarnessEvent): Readonly<HarnessEvent>;
+    /**
+     * Appends an immutable event directly within an existing caller-managed SQLite transaction.
+     * Guarantees atomic RPO-0 consistency between relational table mutations and EventStore.
+     */
+    appendWithinTransaction(event: HarnessEvent): Readonly<HarnessEvent>;
+    /**
+     * Safely dispatches ephemeral notifications for events committed in an outer transaction.
+     */
+    notifyCommitted(events: Readonly<HarnessEvent>[]): void;
+    /**
+     * Reads all events for a given session in strictly chronological order.
+     */
+    getEventsBySession(sessionId: string, options?: StreamQueryOptions): Readonly<HarnessEvent>[];
+    /**
+     * Reads all events for a given project in strictly chronological order.
+     */
+    getEventsByProject(projectId: string, options?: StreamQueryOptions): Readonly<HarnessEvent>[];
+    /**
+     * Retrieves a single event by ID.
+     */
+    getEventById(id: string): Readonly<HarnessEvent> | null;
+    /**
+     * Subscribes to new committed events matching an optional filter.
+     * Returns an unsubscribe function.
+     */
+    subscribe(filter: EventSubscriptionFilter, listener: EventListener): () => void;
+    /**
+     * Safely notifies subscribers. Any subscriber error is trapped and does not propagate.
+     */
+    private notifySubscribers;
+}
+//# sourceMappingURL=event-store.d.ts.map

package/dist/event-state/event-store.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"event-store.d.ts","sourceRoot":"","sources":["../../src/event-state/event-store.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAEvD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iCAAiC,CAAC;AAEpE,MAAM,WAAW,kBAAkB;IACjC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,uBAAuB;IACtC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,MAAM,MAAM,aAAa,GAAG,CAAC,KAAK,EAAE,QAAQ,CAAC,YAAY,CAAC,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;AAQpF;;;;GAIG;AACH,qBAAa,UAAU;IACrB,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAkB;IAC5C,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAe;IACtC,OAAO,CAAC,aAAa,CAA4B;IACjD,OAAO,CAAC,kBAAkB,CAAK;gBAEnB,MAAM,EAAE,YAAY;IAKhC,IAAW,YAAY,IAAI,YAAY,CAEtC;IAED;;;OAGG;IACI,MAAM,CAAC,KAAK,EAAE,YAAY,GAAG,QAAQ,CAAC,YAAY,CAAC;IAY1D;;;OAGG;IACI,uBAAuB,CAAC,KAAK,EAAE,YAAY,GAAG,QAAQ,CAAC,YAAY,CAAC;IAI3E;;OAEG;IACI,eAAe,CAAC,MAAM,EAAE,QAAQ,CAAC,YAAY,CAAC,EAAE,GAAG,IAAI;IAO9D;;OAEG;IACI,kBAAkB,CACvB,SAAS,EAAE,MAAM,EACjB,OAAO,CAAC,EAAE,kBAAkB,GAC3B,QAAQ,CAAC,YAAY,CAAC,EAAE;IAI3B;;OAEG;IACI,kBAAkB,CACvB,SAAS,EAAE,MAAM,EACjB,OAAO,CAAC,EAAE,kBAAkB,GAC3B,QAAQ,CAAC,YAAY,CAAC,EAAE;IA4D3B;;OAEG;IACI,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,QAAQ,CAAC,YAAY,CAAC,GAAG,IAAI;IAI9D;;;OAGG;IACI,SAAS,CAAC,MAAM,EAAE,uBAAuB,EAAE,QAAQ,EAAE,aAAa,GAAG,MAAM,IAAI;IAStF;;OAEG;IACH,OAAO,CAAC,iBAAiB;CAwB1B"}