pi-harness-runtime 0.5.0 → 0.6.3-beta.13

package/packages/evaluation-engine/dist/engine.js

@@ -0,0 +1,344 @@
+/**
+ * Evaluation Engine (RFC-0057)
+ *
+ * Determines whether an agent-produced result is acceptable before advancing.
+ */
+import { randomUUID } from "node:crypto";
+import { DEFAULT_SCORING_WEIGHTS } from "./types.js";
+// ─── Evidence Collection ───────────────────────────────────────────────────────
+function collectEvidence(outputs, testResults, policyFindings) {
+    const evidence = [];
+    // Add test results as evidence
+    for (const test of testResults) {
+        evidence.push({
+            kind: "test",
+            path: test.testFile,
+            description: `Test ${test.id}: ${test.status}`,
+            hash: test.id,
+        });
+    }
+    // Add output files as evidence
+    for (const output of outputs) {
+        if (output.path) {
+            evidence.push({
+                kind: "file",
+                path: output.path,
+                description: output.description,
+                hash: output.hash,
+            });
+        }
+    }
+    // Add policy findings as evidence
+    for (const finding of policyFindings) {
+        evidence.push({
+            kind: "policy",
+            path: finding.filePath,
+            description: `[${finding.severity.toUpperCase()}] ${finding.ruleId}: ${finding.message}`,
+        });
+    }
+    return evidence;
+}
+// ─── Missing Output Detection ──────────────────────────────────────────────────
+function findMissingOutputs(requiredArtifacts, outputs) {
+    const findings = [];
+    const outputPaths = new Set(outputs.filter((o) => o.path).map((o) => o.path));
+    for (const required of requiredArtifacts) {
+        const requiredLower = required.toLowerCase();
+        const found = outputPaths.has(required) ||
+            outputs.some((o) => o.description.toLowerCase().includes(requiredLower));
+        if (!found) {
+            findings.push({
+                dimension: "correctness",
+                severity: "critical",
+                message: `Required artifact '${required}' is missing from outputs`,
+                evidence: `Required: ${required}, Available: ${[...outputPaths].join(", ") || "none"}`,
+            });
+        }
+    }
+    return findings;
+}
+// ─── Test Analysis ────────────────────────────────────────────────────────────
+function analyzeTests(testResults) {
+    const findings = [];
+    let passedCount = 0;
+    let totalCount = 0;
+    let hasFailures = false;
+    for (const test of testResults) {
+        totalCount++;
+        if (test.status === "passed") {
+            passedCount++;
+        }
+        else if (test.status === "failed" || test.status === "error") {
+            hasFailures = true;
+            findings.push({
+                dimension: "test_completeness",
+                severity: "critical",
+                message: `Test '${test.testFile}' ${test.status}: ${test.errorMessage || "Unknown error"}`,
+                evidence: test.errorMessage,
+            });
+        }
+    }
+    if (testResults.length === 0) {
+        findings.push({
+            dimension: "test_completeness",
+            severity: "high",
+            message: "No test results provided",
+        });
+    }
+    return { findings, passedCount, totalCount, hasFailures };
+}
+// ─── Policy Analysis ──────────────────────────────────────────────────────────
+function analyzePolicies(policyFindings) {
+    const findings = [];
+    const criticalFindings = [];
+    for (const finding of policyFindings) {
+        findings.push({
+            dimension: "policy_compliance",
+            severity: finding.severity,
+            message: `${finding.ruleId}: ${finding.message}`,
+            evidence: finding.filePath
+                ? `${finding.filePath}:${finding.line || "?"}`
+                : undefined,
+        });
+        if (finding.severity === "critical") {
+            criticalFindings.push(finding);
+        }
+    }
+    return { findings, criticalFindings };
+}
+// ─── Acceptance Criteria Coverage ─────────────────────────────────────────────
+function analyzeAcceptanceCoverage(acceptanceCriteria, outputs, testResults) {
+    const findings = [];
+    const coveredCriteria = [];
+    const unmetCriteria = [];
+    const outputText = outputs
+        .map((o) => `${o.description} ${o.path || ""}`.toLowerCase())
+        .join(" ");
+    const passingTests = testResults.filter((t) => t.status === "passed");
+    const passingTestText = passingTests
+        .map((t) => `${t.testFile} ${t.errorMessage || ""}`.toLowerCase())
+        .join(" ");
+    for (const criterion of acceptanceCriteria) {
+        const criterionLower = criterion.toLowerCase();
+        const keywords = criterionLower
+            .split(/[^a-z0-9]+/)
+            .filter((word) => word.length >= 4);
+        const outputKeywordMatches = keywords.filter((word) => outputText.includes(word)).length;
+        const coveredByOutput = keywords.length > 0 &&
+            outputKeywordMatches >= Math.ceil(keywords.length / 2);
+        const coveredByPassingTests = passingTests.length > 0 &&
+            (criterionLower.includes("test") ||
+                criterionLower.includes("pass") ||
+                keywords.some((word) => passingTestText.includes(word)));
+        if (coveredByOutput || coveredByPassingTests) {
+            coveredCriteria.push(criterion);
+        }
+        else {
+            unmetCriteria.push(criterion);
+            findings.push({
+                dimension: "acceptance_coverage",
+                severity: "medium",
+                message: `Acceptance criterion not covered: '${criterion}'`,
+                criterion,
+            });
+        }
+    }
+    const coverage = acceptanceCriteria.length > 0
+        ? (coveredCriteria.length / acceptanceCriteria.length) * 100
+        : 100;
+    return { findings, coverage, unmetCriteria };
+}
+// ─── Scoring ─────────────────────────────────────────────────────────────────
+function calculateScore(weights, testAnalysis, acceptanceCoverage, criticalPolicies, securityFindings, regressionFindings) {
+    const breakdown = {};
+    // Correctness (30%) - based on missing outputs
+    const correctnessScore = weights.correctness;
+    breakdown.correctness = correctnessScore;
+    // Test completeness (20%)
+    let testScore = 0;
+    if (testAnalysis.totalCount > 0) {
+        testScore =
+            (testAnalysis.passedCount / testAnalysis.totalCount) * weights.test;
+    }
+    breakdown.tests = testScore;
+    // Acceptance coverage (20%)
+    const acceptanceScore = (acceptanceCoverage / 100) * weights.acceptance;
+    breakdown.acceptance = acceptanceScore;
+    // Code quality (15%) - assume passed if no major findings
+    const qualityScore = weights.quality;
+    breakdown.quality = qualityScore;
+    // Policy penalty (deducted)
+    let policyPenalty = 0;
+    for (const _ of criticalPolicies) {
+        policyPenalty += weights.policyPenalty;
+    }
+    breakdown.policyPenalty = -policyPenalty;
+    // Security penalty (deducted)
+    const securityPenalty = securityFindings.length * weights.securityPenalty;
+    breakdown.securityPenalty = -securityPenalty;
+    // Regression penalty (deducted)
+    const regressionPenalty = regressionFindings.length * weights.regressionPenalty;
+    breakdown.regressionPenalty = -regressionPenalty;
+    const totalScore = Math.max(0, Math.min(100, correctnessScore +
+        testScore +
+        acceptanceScore +
+        qualityScore -
+        policyPenalty -
+        securityPenalty -
+        regressionPenalty));
+    return { score: Math.round(totalScore), breakdown };
+}
+// ─── Failure Classification ───────────────────────────────────────────────────
+function classifyFailure(findings) {
+    const classes = [];
+    for (const finding of findings) {
+        if (finding.dimension === "test_completeness" &&
+            finding.severity === "critical") {
+            if (!classes.includes("test_failure"))
+                classes.push("test_failure");
+        }
+        if (finding.dimension === "correctness" && finding.severity === "high") {
+            if (!classes.includes("missing_output"))
+                classes.push("missing_output");
+        }
+        if (finding.dimension === "acceptance_coverage" &&
+            finding.severity === "medium") {
+            if (!classes.includes("acceptance_gap"))
+                classes.push("acceptance_gap");
+        }
+        if (finding.dimension === "policy_compliance") {
+            if (finding.severity === "critical") {
+                if (!classes.includes("policy_violation"))
+                    classes.push("policy_violation");
+            }
+        }
+        if (finding.dimension === "security_risk") {
+            if (!classes.includes("security_issue"))
+                classes.push("security_issue");
+        }
+        if (finding.dimension === "regression_risk") {
+            if (!classes.includes("regression_risk"))
+                classes.push("regression_risk");
+        }
+    }
+    if (classes.length === 0 && findings.length > 0) {
+        classes.push("insufficient_evidence");
+    }
+    return classes;
+}
+// ─── Main Engine ───────────────────────────────────────────────────────────────
+export class EvaluationEngine {
+    weights;
+    state = "pending";
+    constructor(weights) {
+        this.weights = { ...DEFAULT_SCORING_WEIGHTS, ...weights };
+    }
+    /**
+     * Evaluate a task against its acceptance criteria
+     */
+    evaluate(request) {
+        this.state = "collecting_evidence";
+        const { jobId, task, outputs, testResults, policyFindings } = request;
+        const allFindings = [];
+        this.state = "evaluating";
+        // 1. Check for missing required outputs
+        const missingOutputs = findMissingOutputs(task.requiredArtifacts, outputs);
+        allFindings.push(...missingOutputs);
+        // 2. Analyze tests
+        const testAnalysis = analyzeTests(testResults);
+        allFindings.push(...testAnalysis.findings);
+        // 3. Analyze policies
+        const policyAnalysis = analyzePolicies(policyFindings);
+        allFindings.push(...policyAnalysis.findings);
+        // 4. Check acceptance criteria coverage
+        const acceptanceAnalysis = analyzeAcceptanceCoverage(task.acceptanceCriteria, outputs, testResults);
+        allFindings.push(...acceptanceAnalysis.findings);
+        // 5. Check for security issues
+        const securityFindings = allFindings.filter((f) => f.dimension === "security_risk");
+        // 6. Check for regression risks
+        const regressionFindings = allFindings.filter((f) => f.dimension === "regression_risk");
+        // 7. Calculate score
+        const { score, breakdown: _breakdown } = calculateScore(this.weights, testAnalysis, acceptanceAnalysis.coverage, policyAnalysis.criticalFindings, securityFindings, regressionFindings);
+        // 8. Collect evidence
+        const evidence = collectEvidence(outputs, testResults, policyFindings);
+        // 9. Determine status and recommended action
+        let status;
+        let recommendedAction;
+        // Critical policy violations force failure
+        if (policyAnalysis.criticalFindings.length > 0) {
+            status = "failed";
+            recommendedAction = "repair";
+            this.state = "failed";
+        }
+        // Test failures force failure
+        else if (testAnalysis.hasFailures) {
+            status = "failed";
+            recommendedAction = "repair";
+            this.state = "failed";
+        }
+        // Missing critical outputs
+        else if (missingOutputs.some((f) => f.severity === "critical")) {
+            status = "failed";
+            recommendedAction = "repair";
+            this.state = "failed";
+        }
+        // Low score with significant gaps
+        else if (score < 50) {
+            status = "failed";
+            recommendedAction = "repair";
+            this.state = "failed";
+        }
+        // Ambiguous criteria coverage
+        else if (acceptanceAnalysis.coverage < 80 &&
+            acceptanceAnalysis.unmetCriteria.length > 0) {
+            status = "needs_review";
+            recommendedAction = "human_review";
+            this.state = "needs_review";
+        }
+        // Good score
+        else {
+            status = "passed";
+            recommendedAction = "continue";
+            this.state = "passed";
+        }
+        // 10. Classify failure
+        const failureClasses = classifyFailure(allFindings);
+        // Add failure classification as a finding if failed
+        if (status === "failed" && failureClasses.length > 0) {
+            allFindings.push({
+                dimension: "correctness",
+                severity: "critical",
+                message: `Evaluation failed due to: ${failureClasses.join(", ")}`,
+                evidence: failureClasses.join(", "),
+            });
+        }
+        return {
+            status,
+            score,
+            findings: allFindings,
+            evidence,
+            recommendedAction,
+            evaluationId: randomUUID(),
+            jobId,
+            taskId: task.id,
+            evaluatedAt: new Date().toISOString(),
+        };
+    }
+    /**
+     * Get current evaluation state
+     */
+    getState() {
+        return this.state;
+    }
+    /**
+     * Reset engine state
+     */
+    reset() {
+        this.state = "pending";
+    }
+}
+// ─── Factory ─────────────────────────────────────────────────────────────────
+export function createEvaluationEngine(weights) {
+    return new EvaluationEngine(weights);
+}
+//# sourceMappingURL=engine.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"engine.js","sourceRoot":"","sources":["../src/engine.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAczC,OAAO,EAAE,uBAAuB,EAAE,MAAM,YAAY,CAAC;AAErD,kFAAkF;AAElF,SAAS,eAAe,CACvB,OAA6B,EAC7B,WAAyB,EACzB,cAA+B;IAE/B,MAAM,QAAQ,GAAyB,EAAE,CAAC;IAE1C,+BAA+B;IAC/B,KAAK,MAAM,IAAI,IAAI,WAAW,EAAE,CAAC;QAChC,QAAQ,CAAC,IAAI,CAAC;YACb,IAAI,EAAE,MAAM;YACZ,IAAI,EAAE,IAAI,CAAC,QAAQ;YACnB,WAAW,EAAE,QAAQ,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,MAAM,EAAE;YAC9C,IAAI,EAAE,IAAI,CAAC,EAAE;SACb,CAAC,CAAC;IACJ,CAAC;IAED,+BAA+B;IAC/B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC;YACjB,QAAQ,CAAC,IAAI,CAAC;gBACb,IAAI,EAAE,MAAM;gBACZ,IAAI,EAAE,MAAM,CAAC,IAAI;gBACjB,WAAW,EAAE,MAAM,CAAC,WAAW;gBAC/B,IAAI,EAAE,MAAM,CAAC,IAAI;aACjB,CAAC,CAAC;QACJ,CAAC;IACF,CAAC;IAED,kCAAkC;IAClC,KAAK,MAAM,OAAO,IAAI,cAAc,EAAE,CAAC;QACtC,QAAQ,CAAC,IAAI,CAAC;YACb,IAAI,EAAE,QAAQ;YACd,IAAI,EAAE,OAAO,CAAC,QAAQ;YACtB,WAAW,EAAE,IAAI,OAAO,CAAC,QAAQ,CAAC,WAAW,EAAE,KAAK,OAAO,CAAC,MAAM,KAAK,OAAO,CAAC,OAAO,EAAE;SACxF,CAAC,CAAC;IACJ,CAAC;IAED,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,kFAAkF;AAElF,SAAS,kBAAkB,CAC1B,iBAA2B,EAC3B,OAA6B;IAE7B,MAAM,QAAQ,GAAwB,EAAE,CAAC;IACzC,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IAE9E,KAAK,MAAM,QAAQ,IAAI,iBAAiB,EAAE,CAAC;QAC1C,MAAM,aAAa,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC;QAC7C,MAAM,KAAK,GACV,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC;YACzB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC,CAAC;QAE1E,IAAI,CAAC,KAAK,EAAE,CAAC;YACZ,QAAQ,CAAC,IAAI,CAAC;gBACb,SAAS,EAAE,aAAa;gBACxB,QAAQ,EAAE,UAAU;gBACpB,OAAO,EAAE,sBAAsB,QAAQ,2BAA2B;gBAClE,QAAQ,EAAE,aAAa,QAAQ,gBAAgB,CAAC,GAAG,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,MAAM,EAAE;aACtF,CAAC,CAAC;QACJ,CAAC;IACF,CAAC;IAED,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,iFAAiF;AAEjF,SAAS,YAAY,CAAC,WAAyB;IAM9C,MAAM,QAAQ,GAAwB,EAAE,CAAC;IACzC,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,WAAW,GAAG,KAAK,CAAC;IAExB,KAAK,MAAM,IAAI,IAAI,WAAW,EAAE,CAAC;QAChC,UAAU,EAAE,CAAC;QACb,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC9B,WAAW,EAAE,CAAC;QACf,CAAC;aAAM,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;YAChE,WAAW,GAAG,IAAI,CAAC;YACnB,QAAQ,CAAC,IAAI,CAAC;gBACb,SAAS,EAAE,mBAAmB;gBAC9B,QAAQ,EAAE,UAAU;gBACpB,OAAO,EAAE,SAAS,IAAI,CAAC,QAAQ,KAAK,IAAI,CAAC,MAAM,KAAK,IAAI,CAAC,YAAY,IAAI,eAAe,EAAE;gBAC1F,QAAQ,EAAE,IAAI,CAAC,YAAY;aAC3B,CAAC,CAAC;QACJ,CAAC;IACF,CAAC;IAED,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC9B,QAAQ,CAAC,IAAI,CAAC;YACb,SAAS,EAAE,mBAAmB;YAC9B,QAAQ,EAAE,MAAM;YAChB,OAAO,EAAE,0BAA0B;SACnC,CAAC,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAE,CAAC;AAC3D,CAAC;AAED,iFAAiF;AAEjF,SAAS,eAAe,CAAC,cAA+B;IAIvD,MAAM,QAAQ,GAAwB,EAAE,CAAC;IACzC,MAAM,gBAAgB,GAAoB,EAAE,CAAC;IAE7C,KAAK,MAAM,OAAO,IAAI,cAAc,EAAE,CAAC;QACtC,QAAQ,CAAC,IAAI,CAAC;YACb,SAAS,EAAE,mBAAmB;YAC9B,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,OAAO,EAAE,GAAG,OAAO,CAAC,MAAM,KAAK,OAAO,CAAC,OAAO,EAAE;YAChD,QAAQ,EAAE,OAAO,CAAC,QAAQ;gBACzB,CAAC,CAAC,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,IAAI,IAAI,GAAG,EAAE;gBAC9C,CAAC,CAAC,SAAS;SACZ,CAAC,CAAC;QAEH,IAAI,OAAO,CAAC,QAAQ,KAAK,UAAU,EAAE,CAAC;YACrC,gBAAgB,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAChC,CAAC;IACF,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,gBAAgB,EAAE,CAAC;AACvC,CAAC;AAED,iFAAiF;AAEjF,SAAS,yBAAyB,CACjC,kBAA4B,EAC5B,OAA6B,EAC7B,WAAyB;IAMzB,MAAM,QAAQ,GAAwB,EAAE,CAAC;IACzC,MAAM,eAAe,GAAa,EAAE,CAAC;IACrC,MAAM,aAAa,GAAa,EAAE,CAAC;IACnC,MAAM,UAAU,GAAG,OAAO;SACxB,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,WAAW,IAAI,CAAC,CAAC,IAAI,IAAI,EAAE,EAAE,CAAC,WAAW,EAAE,CAAC;SAC5D,IAAI,CAAC,GAAG,CAAC,CAAC;IACZ,MAAM,YAAY,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC;IACtE,MAAM,eAAe,GAAG,YAAY;SAClC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,YAAY,IAAI,EAAE,EAAE,CAAC,WAAW,EAAE,CAAC;SACjE,IAAI,CAAC,GAAG,CAAC,CAAC;IAEZ,KAAK,MAAM,SAAS,IAAI,kBAAkB,EAAE,CAAC;QAC5C,MAAM,cAAc,GAAG,SAAS,CAAC,WAAW,EAAE,CAAC;QAC/C,MAAM,QAAQ,GAAG,cAAc;aAC7B,KAAK,CAAC,YAAY,CAAC;aACnB,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC;QACrC,MAAM,oBAAoB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CACrD,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,CACzB,CAAC,MAAM,CAAC;QACT,MAAM,eAAe,GACpB,QAAQ,CAAC,MAAM,GAAG,CAAC;YACnB,oBAAoB,IAAI,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACxD,MAAM,qBAAqB,GAC1B,YAAY,CAAC,MAAM,GAAG,CAAC;YACvB,CAAC,cAAc,CAAC,QAAQ,CAAC,MAAM,CAAC;gBAC/B,cAAc,CAAC,QAAQ,CAAC,MAAM,CAAC;gBAC/B,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,eAAe,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAE3D,IAAI,eAAe,IAAI,qBAAqB,EAAE,CAAC;YAC9C,eAAe,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACjC,CAAC;aAAM,CAAC;YACP,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;YAC9B,QAAQ,CAAC,IAAI,CAAC;gBACb,SAAS,EAAE,qBAAqB;gBAChC,QAAQ,EAAE,QAAQ;gBAClB,OAAO,EAAE,sCAAsC,SAAS,GAAG;gBAC3D,SAAS;aACT,CAAC,CAAC;QACJ,CAAC;IACF,CAAC;IAED,MAAM,QAAQ,GACb,kBAAkB,CAAC,MAAM,GAAG,CAAC;QAC5B,CAAC,CAAC,CAAC,eAAe,CAAC,MAAM,GAAG,kBAAkB,CAAC,MAAM,CAAC,GAAG,GAAG;QAC5D,CAAC,CAAC,GAAG,CAAC;IAER,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,aAAa,EAAE,CAAC;AAC9C,CAAC;AAED,gFAAgF;AAEhF,SAAS,cAAc,CACtB,OAAuB,EACvB,YAIC,EACD,kBAA0B,EAC1B,gBAAiC,EACjC,gBAAqC,EACrC,kBAAuC;IAEvC,MAAM,SAAS,GAA2B,EAAE,CAAC;IAE7C,+CAA+C;IAC/C,MAAM,gBAAgB,GAAG,OAAO,CAAC,WAAW,CAAC;IAC7C,SAAS,CAAC,WAAW,GAAG,gBAAgB,CAAC;IAEzC,0BAA0B;IAC1B,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,YAAY,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC;QACjC,SAAS;YACR,CAAC,YAAY,CAAC,WAAW,GAAG,YAAY,CAAC,UAAU,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC;IACtE,CAAC;IACD,SAAS,CAAC,KAAK,GAAG,SAAS,CAAC;IAE5B,4BAA4B;IAC5B,MAAM,eAAe,GAAG,CAAC,kBAAkB,GAAG,GAAG,CAAC,GAAG,OAAO,CAAC,UAAU,CAAC;IACxE,SAAS,CAAC,UAAU,GAAG,eAAe,CAAC;IAEvC,0DAA0D;IAC1D,MAAM,YAAY,GAAG,OAAO,CAAC,OAAO,CAAC;IACrC,SAAS,CAAC,OAAO,GAAG,YAAY,CAAC;IAEjC,4BAA4B;IAC5B,IAAI,aAAa,GAAG,CAAC,CAAC;IACtB,KAAK,MAAM,CAAC,IAAI,gBAAgB,EAAE,CAAC;QAClC,aAAa,IAAI,OAAO,CAAC,aAAa,CAAC;IACxC,CAAC;IACD,SAAS,CAAC,aAAa,GAAG,CAAC,aAAa,CAAC;IAEzC,8BAA8B;IAC9B,MAAM,eAAe,GAAG,gBAAgB,CAAC,MAAM,GAAG,OAAO,CAAC,eAAe,CAAC;IAC1E,SAAS,CAAC,eAAe,GAAG,CAAC,eAAe,CAAC;IAE7C,gCAAgC;IAChC,MAAM,iBAAiB,GACtB,kBAAkB,CAAC,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC;IACvD,SAAS,CAAC,iBAAiB,GAAG,CAAC,iBAAiB,CAAC;IAEjD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAC1B,CAAC,EACD,IAAI,CAAC,GAAG,CACP,GAAG,EACH,gBAAgB;QACf,SAAS;QACT,eAAe;QACf,YAAY;QACZ,aAAa;QACb,eAAe;QACf,iBAAiB,CAClB,CACD,CAAC;IAEF,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,EAAE,SAAS,EAAE,CAAC;AACrD,CAAC;AAED,iFAAiF;AAEjF,SAAS,eAAe,CACvB,QAA6B;IAE7B,MAAM,OAAO,GAA6B,EAAE,CAAC;IAE7C,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAChC,IACC,OAAO,CAAC,SAAS,KAAK,mBAAmB;YACzC,OAAO,CAAC,QAAQ,KAAK,UAAU,EAC9B,CAAC;YACF,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC;gBAAE,OAAO,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;QACrE,CAAC;QACD,IAAI,OAAO,CAAC,SAAS,KAAK,aAAa,IAAI,OAAO,CAAC,QAAQ,KAAK,MAAM,EAAE,CAAC;YACxE,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;gBAAE,OAAO,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;QACzE,CAAC;QACD,IACC,OAAO,CAAC,SAAS,KAAK,qBAAqB;YAC3C,OAAO,CAAC,QAAQ,KAAK,QAAQ,EAC5B,CAAC;YACF,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;gBAAE,OAAO,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;QACzE,CAAC;QACD,IAAI,OAAO,CAAC,SAAS,KAAK,mBAAmB,EAAE,CAAC;YAC/C,IAAI,OAAO,CAAC,QAAQ,KAAK,UAAU,EAAE,CAAC;gBACrC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAC;oBACxC,OAAO,CAAC,IAAI,CAAC,kBAAkB,CAAC,CAAC;YACnC,CAAC;QACF,CAAC;QACD,IAAI,OAAO,CAAC,SAAS,KAAK,eAAe,EAAE,CAAC;YAC3C,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC;gBAAE,OAAO,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;QACzE,CAAC;QACD,IAAI,OAAO,CAAC,SAAS,KAAK,iBAAiB,EAAE,CAAC;YAC7C,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAC;gBAAE,OAAO,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC;QAC3E,CAAC;IACF,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjD,OAAO,CAAC,IAAI,CAAC,uBAAuB,CAAC,CAAC;IACvC,CAAC;IAED,OAAO,OAAO,CAAC;AAChB,CAAC;AAED,kFAAkF;AAElF,MAAM,OAAO,gBAAgB;IACpB,OAAO,CAAiB;IACxB,KAAK,GAAoB,SAAS,CAAC;IAE3C,YAAY,OAAiC;QAC5C,IAAI,CAAC,OAAO,GAAG,EAAE,GAAG,uBAAuB,EAAE,GAAG,OAAO,EAAE,CAAC;IAC3D,CAAC;IAED;;OAEG;IACH,QAAQ,CAAC,OAA0B;QAClC,IAAI,CAAC,KAAK,GAAG,qBAAqB,CAAC;QAEnC,MAAM,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,WAAW,EAAE,cAAc,EAAE,GAAG,OAAO,CAAC;QACtE,MAAM,WAAW,GAAwB,EAAE,CAAC;QAE5C,IAAI,CAAC,KAAK,GAAG,YAAY,CAAC;QAE1B,wCAAwC;QACxC,MAAM,cAAc,GAAG,kBAAkB,CAAC,IAAI,CAAC,iBAAiB,EAAE,OAAO,CAAC,CAAC;QAC3E,WAAW,CAAC,IAAI,CAAC,GAAG,cAAc,CAAC,CAAC;QAEpC,mBAAmB;QACnB,MAAM,YAAY,GAAG,YAAY,CAAC,WAAW,CAAC,CAAC;QAC/C,WAAW,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;QAE3C,sBAAsB;QACtB,MAAM,cAAc,GAAG,eAAe,CAAC,cAAc,CAAC,CAAC;QACvD,WAAW,CAAC,IAAI,CAAC,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;QAE7C,wCAAwC;QACxC,MAAM,kBAAkB,GAAG,yBAAyB,CACnD,IAAI,CAAC,kBAAkB,EACvB,OAAO,EACP,WAAW,CACX,CAAC;QACF,WAAW,CAAC,IAAI,CAAC,GAAG,kBAAkB,CAAC,QAAQ,CAAC,CAAC;QAEjD,+BAA+B;QAC/B,MAAM,gBAAgB,GAAG,WAAW,CAAC,MAAM,CAC1C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,eAAe,CACtC,CAAC;QAEF,gCAAgC;QAChC,MAAM,kBAAkB,GAAG,WAAW,CAAC,MAAM,CAC5C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,iBAAiB,CACxC,CAAC;QAEF,qBAAqB;QACrB,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,UAAU,EAAE,GAAG,cAAc,CACtD,IAAI,CAAC,OAAO,EACZ,YAAY,EACZ,kBAAkB,CAAC,QAAQ,EAC3B,cAAc,CAAC,gBAAgB,EAC/B,gBAAgB,EAChB,kBAAkB,CAClB,CAAC;QAEF,sBAAsB;QACtB,MAAM,QAAQ,GAAG,eAAe,CAAC,OAAO,EAAE,WAAW,EAAE,cAAc,CAAC,CAAC;QAEvE,6CAA6C;QAC7C,IAAI,MAAwB,CAAC;QAC7B,IAAI,iBAAwD,CAAC;QAE7D,2CAA2C;QAC3C,IAAI,cAAc,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChD,MAAM,GAAG,QAAQ,CAAC;YAClB,iBAAiB,GAAG,QAAQ,CAAC;YAC7B,IAAI,CAAC,KAAK,GAAG,QAAQ,CAAC;QACvB,CAAC;QACD,8BAA8B;aACzB,IAAI,YAAY,CAAC,WAAW,EAAE,CAAC;YACnC,MAAM,GAAG,QAAQ,CAAC;YAClB,iBAAiB,GAAG,QAAQ,CAAC;YAC7B,IAAI,CAAC,KAAK,GAAG,QAAQ,CAAC;QACvB,CAAC;QACD,2BAA2B;aACtB,IAAI,cAAc,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,UAAU,CAAC,EAAE,CAAC;YAChE,MAAM,GAAG,QAAQ,CAAC;YAClB,iBAAiB,GAAG,QAAQ,CAAC;YAC7B,IAAI,CAAC,KAAK,GAAG,QAAQ,CAAC;QACvB,CAAC;QACD,kCAAkC;aAC7B,IAAI,KAAK,GAAG,EAAE,EAAE,CAAC;YACrB,MAAM,GAAG,QAAQ,CAAC;YAClB,iBAAiB,GAAG,QAAQ,CAAC;YAC7B,IAAI,CAAC,KAAK,GAAG,QAAQ,CAAC;QACvB,CAAC;QACD,8BAA8B;aACzB,IACJ,kBAAkB,CAAC,QAAQ,GAAG,EAAE;YAChC,kBAAkB,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAC1C,CAAC;YACF,MAAM,GAAG,cAAc,CAAC;YACxB,iBAAiB,GAAG,cAAc,CAAC;YACnC,IAAI,CAAC,KAAK,GAAG,cAAc,CAAC;QAC7B,CAAC;QACD,aAAa;aACR,CAAC;YACL,MAAM,GAAG,QAAQ,CAAC;YAClB,iBAAiB,GAAG,UAAU,CAAC;YAC/B,IAAI,CAAC,KAAK,GAAG,QAAQ,CAAC;QACvB,CAAC;QAED,uBAAuB;QACvB,MAAM,cAAc,GAAG,eAAe,CAAC,WAAW,CAAC,CAAC;QAEpD,oDAAoD;QACpD,IAAI,MAAM,KAAK,QAAQ,IAAI,cAAc,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtD,WAAW,CAAC,IAAI,CAAC;gBAChB,SAAS,EAAE,aAAa;gBACxB,QAAQ,EAAE,UAAU;gBACpB,OAAO,EAAE,6BAA6B,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;gBACjE,QAAQ,EAAE,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC;aACnC,CAAC,CAAC;QACJ,CAAC;QAED,OAAO;YACN,MAAM;YACN,KAAK;YACL,QAAQ,EAAE,WAAW;YACrB,QAAQ;YACR,iBAAiB;YACjB,YAAY,EAAE,UAAU,EAAE;YAC1B,KAAK;YACL,MAAM,EAAE,IAAI,CAAC,EAAE;YACf,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SACrC,CAAC;IACH,CAAC;IAED;;OAEG;IACH,QAAQ;QACP,OAAO,IAAI,CAAC,KAAK,CAAC;IACnB,CAAC;IAED;;OAEG;IACH,KAAK;QACJ,IAAI,CAAC,KAAK,GAAG,SAAS,CAAC;IACxB,CAAC;CACD;AAED,gFAAgF;AAEhF,MAAM,UAAU,sBAAsB,CACrC,OAAiC;IAEjC,OAAO,IAAI,gBAAgB,CAAC,OAAO,CAAC,CAAC;AACtC,CAAC"}

package/packages/evaluation-engine/dist/index.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Evaluation Engine (RFC-0057)
+ *
+ * Re-exports all public types and classes.
+ */
+export { EvaluationEngine, createEvaluationEngine } from "./engine.js";
+export type { EvaluationRequest, EvaluationResult, EvaluationFinding, EvaluationEvidence, EvaluationStatus, EvaluationFailureClass, EvaluationState, CompiledTask, TaskOutputArtifact, TestResult, PolicyFinding, ScoringWeights, EvaluationDimension, } from "./types.js";
+export { DEFAULT_SCORING_WEIGHTS } from "./types.js";
+//# sourceMappingURL=index.d.ts.map

package/packages/evaluation-engine/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,aAAa,CAAC;AACvE,YAAY,EACX,iBAAiB,EACjB,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,EAClB,gBAAgB,EAChB,sBAAsB,EACtB,eAAe,EACf,YAAY,EACZ,kBAAkB,EAClB,UAAU,EACV,aAAa,EACb,cAAc,EACd,mBAAmB,GACnB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,uBAAuB,EAAE,MAAM,YAAY,CAAC"}

package/packages/evaluation-engine/dist/index.js

@@ -0,0 +1,8 @@
+/**
+ * Evaluation Engine (RFC-0057)
+ *
+ * Re-exports all public types and classes.
+ */
+export { EvaluationEngine, createEvaluationEngine } from "./engine.js";
+export { DEFAULT_SCORING_WEIGHTS } from "./types.js";
+//# sourceMappingURL=index.js.map

package/packages/evaluation-engine/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,aAAa,CAAC;AAgBvE,OAAO,EAAE,uBAAuB,EAAE,MAAM,YAAY,CAAC"}

package/packages/evaluation-engine/dist/types.d.ts

@@ -0,0 +1,87 @@
+/**
+ * Evaluation Engine Types (RFC-0057)
+ *
+ * Interfaces for task evaluation and acceptance criteria checking.
+ */
+export interface TaskOutputArtifact {
+    id: string;
+    type: "file" | "directory" | "url" | "test-result" | "screenshot" | "log";
+    path?: string;
+    url?: string;
+    description: string;
+    hash?: string;
+    timestamp?: string;
+}
+export interface TestResult {
+    id: string;
+    testFile: string;
+    status: "passed" | "failed" | "skipped" | "error";
+    durationMs?: number;
+    errorMessage?: string;
+    testCount?: number;
+    passedCount?: number;
+    failedCount?: number;
+}
+export interface PolicyFinding {
+    ruleId: string;
+    severity: "critical" | "high" | "medium" | "low";
+    message: string;
+    filePath?: string;
+    line?: number;
+}
+export interface CompiledTask {
+    id: string;
+    title: string;
+    description: string;
+    acceptanceCriteria: string[];
+    requiredArtifacts: string[];
+    mandatoryStages: string[];
+}
+export interface EvaluationRequest {
+    jobId: string;
+    task: CompiledTask;
+    worktreePath: string;
+    outputs: TaskOutputArtifact[];
+    testResults: TestResult[];
+    policyFindings: PolicyFinding[];
+}
+export type EvaluationStatus = "passed" | "failed" | "needs_review";
+export interface EvaluationFinding {
+    dimension: EvaluationDimension;
+    severity: "critical" | "high" | "medium" | "low";
+    message: string;
+    evidence?: string;
+    criterion?: string;
+}
+export type EvaluationDimension = "correctness" | "test_completeness" | "acceptance_coverage" | "policy_compliance" | "code_quality" | "security_risk" | "regression_risk" | "documentation";
+export interface EvaluationEvidence {
+    kind: "test" | "diff" | "file" | "policy" | "screenshot" | "trace";
+    path?: string;
+    description: string;
+    hash?: string;
+}
+export type EvaluationResultStatus = "passed" | "failed" | "needs_review";
+export interface EvaluationResult {
+    status: EvaluationResultStatus;
+    score: number;
+    findings: EvaluationFinding[];
+    evidence: EvaluationEvidence[];
+    recommendedAction: "continue" | "repair" | "human_review";
+    evaluationId: string;
+    jobId: string;
+    taskId: string;
+    evaluatedAt: string;
+}
+export type EvaluationFailureClass = "test_failure" | "missing_output" | "acceptance_gap" | "policy_violation" | "security_issue" | "regression_risk" | "insufficient_evidence";
+export type EvaluationState = "pending" | "collecting_evidence" | "evaluating" | "passed" | "failed" | "needs_review";
+export interface ScoringWeights {
+    correctness: number;
+    test: number;
+    acceptance: number;
+    quality: number;
+    policyPenalty: number;
+    securityPenalty: number;
+    regressionPenalty: number;
+}
+export declare const DEFAULT_SCORING_WEIGHTS: ScoringWeights;
+//# sourceMappingURL=types.d.ts.map

package/packages/evaluation-engine/dist/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAIH,MAAM,WAAW,kBAAkB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,KAAK,GAAG,aAAa,GAAG,YAAY,GAAG,KAAK,CAAC;IAC1E,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,UAAU;IAC1B,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,CAAC;IAClD,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,aAAa;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAC;IACjD,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,YAAY;IAC5B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,kBAAkB,EAAE,MAAM,EAAE,CAAC;IAC7B,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAC5B,eAAe,EAAE,MAAM,EAAE,CAAC;CAC1B;AAID,MAAM,WAAW,iBAAiB;IACjC,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,YAAY,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,kBAAkB,EAAE,CAAC;IAC9B,WAAW,EAAE,UAAU,EAAE,CAAC;IAC1B,cAAc,EAAE,aAAa,EAAE,CAAC;CAChC;AAED,MAAM,MAAM,gBAAgB,GAAG,QAAQ,GAAG,QAAQ,GAAG,cAAc,CAAC;AAEpE,MAAM,WAAW,iBAAiB;IACjC,SAAS,EAAE,mBAAmB,CAAC;IAC/B,QAAQ,EAAE,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAC;IACjD,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,MAAM,mBAAmB,GAC5B,aAAa,GACb,mBAAmB,GACnB,qBAAqB,GACrB,mBAAmB,GACnB,cAAc,GACd,eAAe,GACf,iBAAiB,GACjB,eAAe,CAAC;AAEnB,MAAM,WAAW,kBAAkB;IAClC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,OAAO,CAAC;IACnE,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED,MAAM,MAAM,sBAAsB,GAAG,QAAQ,GAAG,QAAQ,GAAG,cAAc,CAAC;AAE1E,MAAM,WAAW,gBAAgB;IAChC,MAAM,EAAE,sBAAsB,CAAC;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,iBAAiB,EAAE,CAAC;IAC9B,QAAQ,EAAE,kBAAkB,EAAE,CAAC;IAC/B,iBAAiB,EAAE,UAAU,GAAG,QAAQ,GAAG,cAAc,CAAC;IAC1D,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,sBAAsB,GAC/B,cAAc,GACd,gBAAgB,GAChB,gBAAgB,GAChB,kBAAkB,GAClB,gBAAgB,GAChB,iBAAiB,GACjB,uBAAuB,CAAC;AAE3B,MAAM,MAAM,eAAe,GACxB,SAAS,GACT,qBAAqB,GACrB,YAAY,GACZ,QAAQ,GACR,QAAQ,GACR,cAAc,CAAC;AAIlB,MAAM,WAAW,cAAc;IAC9B,WAAW,EAAE,MAAM,CAAC;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;IACxB,iBAAiB,EAAE,MAAM,CAAC;CAC1B;AAED,eAAO,MAAM,uBAAuB,EAAE,cAQrC,CAAC"}

package/packages/evaluation-engine/dist/types.js

@@ -0,0 +1,15 @@
+/**
+ * Evaluation Engine Types (RFC-0057)
+ *
+ * Interfaces for task evaluation and acceptance criteria checking.
+ */
+export const DEFAULT_SCORING_WEIGHTS = {
+    correctness: 30,
+    test: 20,
+    acceptance: 20,
+    quality: 15,
+    policyPenalty: 20,
+    securityPenalty: 30,
+    regressionPenalty: 15,
+};
+//# sourceMappingURL=types.js.map

package/packages/evaluation-engine/dist/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AA2HH,MAAM,CAAC,MAAM,uBAAuB,GAAmB;IACtD,WAAW,EAAE,EAAE;IACf,IAAI,EAAE,EAAE;IACR,UAAU,EAAE,EAAE;IACd,OAAO,EAAE,EAAE;IACX,aAAa,EAAE,EAAE;IACjB,eAAe,EAAE,EAAE;IACnB,iBAAiB,EAAE,EAAE;CACrB,CAAC"}

package/packages/evaluation-engine/package.json

@@ -0,0 +1,23 @@
+{
+  "name": "@pi-harness/evaluation-engine",
+  "version": "0.6.3-beta.13",
+  "description": "Evaluation engine for pi-harness runtime (RFC-0057)",
+  "type": "module",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": "./dist/index.js",
+      "types": "./dist/index.d.ts"
+    }
+  },
+  "scripts": {
+    "build": "tsc",
+    "test": "node --experimental-vm-modules ../../node_modules/jest/bin/jest.js",
+    "lint": "eslint src --ext .ts"
+  },
+  "devDependencies": {
+    "@types/node": "^22.0.0",
+    "typescript": "^5.5.0"
+  }
+}