progmune-runtime 2.1.6 → 3.2.0

package/dist/repair-taxonomy.js

@@ -0,0 +1,234 @@
+"use strict";
+/**
+ * P4: Repair Failure Taxonomy — Production Failure Classification
+ *
+ * Every failed repair is classified into one of 7 root causes.
+ * This taxonomy is the PRIMARY business metric for repair quality.
+ *
+ * It supersedes "repair success rate" as the key KPI because:
+ *   - "57% success" tells you nothing about WHY it fails
+ *   - "34% no_path, 31% not_applied" tells you EXACTLY what to fix
+ *
+ * Seven categories (ordered by severity):
+ *   1. NO_PATH          — No fix path found by any strategy
+ *   2. NOT_APPLIED      — Fix path exists but was never executed
+ *   3. WRONG_STRATEGY   — Fix applied but direction was wrong (prepend vs append)
+ *   4. VERIFY_FAILED    — Fix applied but re-verification still shows violations
+ *   5. COMPILE_FAILED   — Fix compiles but produces wrong behavior
+ *   6. SEMANTIC_CHANGED — Fix changed program semantics
+ *   7. HUMAN_REJECTED   — Developer reviewed and rejected the fix
+ *
+ * Monthly Report tracks:
+ *   - Distribution of failure reasons
+ *   - Trend over time
+ *   - Per-protocol breakdown
+ *   - Actionable recommendations
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.classifyRepairFailure = classifyRepairFailure;
+exports.generateRepairTaxonomyReport = generateRepairTaxonomyReport;
+exports.formatRepairTaxonomy = formatRepairTaxonomy;
+const failure_corpus_1 = require("./failure-corpus");
+// ═══════════════════════════════════════════════════════════════
+// Taxonomy Engine
+// ═══════════════════════════════════════════════════════════════
+const CATEGORY_LABELS = {
+    NO_PATH: {
+        label: "No Fix Path",
+        description: "No repair strategy produced any candidate. Protocol rules don't cover this scenario.",
+    },
+    NOT_APPLIED: {
+        label: "Fix Not Applied",
+        description: "A fix path was found but never executed. The repair pipeline stopped at suggestion.",
+    },
+    WRONG_STRATEGY: {
+        label: "Wrong Strategy",
+        description: "Fix was applied but the strategy was incorrect (e.g., prepended cleanup instead of appending).",
+    },
+    VERIFY_FAILED: {
+        label: "Verification Failed",
+        description: "Fix applied but re-verification still shows protocol violations.",
+    },
+    COMPILE_FAILED: {
+        label: "Compile Failed",
+        description: "Fix application produced code that fails to compile.",
+    },
+    SEMANTIC_CHANGED: {
+        label: "Semantic Changed",
+        description: "Fix passed verification but changed program behavior.",
+    },
+    HUMAN_REJECTED: {
+        label: "Human Rejected",
+        description: "Developer reviewed the fix and explicitly rejected it.",
+    },
+};
+/**
+ * Classify a failed repair trajectory into a taxonomy category.
+ *
+ * This is the core classification function. It analyzes trajectory data
+ * to determine WHY a repair failed.
+ */
+function classifyRepairFailure(trajectory) {
+    const desc = (trajectory.violation?.description || "").toLowerCase();
+    // Rule 1: No fix path at all
+    if (!trajectory.violation?.fixPath || trajectory.violation.fixPath.length === 0) {
+        return "NO_PATH";
+    }
+    // Rule 2: Fix was never applied (description indicates still-broken state)
+    if (desc.includes("still leaking") || desc.includes("attempted fix but") ||
+        desc.includes("fix didn't work") || desc.includes("attempted close but")) {
+        return "NOT_APPLIED";
+    }
+    // Rule 3: Wrong strategy
+    if (desc.includes("wrong order") || desc.includes("reversed") || desc.includes("wrong direction")) {
+        return "WRONG_STRATEGY";
+    }
+    // Rule 4: Compile failure
+    if (desc.includes("compile") || desc.includes("syntax") || desc.includes("type error")) {
+        return "COMPILE_FAILED";
+    }
+    // Rule 5: Semantic change
+    if (desc.includes("semantic") || desc.includes("behavior changed") || desc.includes("side effect")) {
+        return "SEMANTIC_CHANGED";
+    }
+    // Rule 6: HUMAN_REJECTED — feedback says rejected AND reason is in description
+    // This catches: developer reviewed, explicitly rejected with reason
+    if (trajectory.feedback?.rejected) {
+        return "HUMAN_REJECTED";
+    }
+    // Rule 7: Default — fix applied but verification still fails
+    return "VERIFY_FAILED";
+}
+/**
+ * Generate a full repair taxonomy report from trajectory data.
+ */
+function generateRepairTaxonomyReport(period = "all-time") {
+    const all = (0, failure_corpus_1.loadTrajectories)().filter(t => t.result === "repair");
+    const success = all.filter(t => (t.successRate || 0) >= 0.5);
+    const fail = all.filter(t => (t.successRate || 0) < 0.5);
+    // Classify all failures
+    const categoryCounts = {};
+    for (const f of fail) {
+        const cat = classifyRepairFailure(f);
+        if (!categoryCounts[cat])
+            categoryCounts[cat] = { count: 0, examples: [] };
+        categoryCounts[cat].count++;
+        if (categoryCounts[cat].examples.length < 3) {
+            categoryCounts[cat].examples.push(f.violation?.description || f.metadata?.intent || "unknown");
+        }
+    }
+    // Build buckets
+    const buckets = Object.entries(categoryCounts)
+        .map(([cat, data]) => ({
+        category: cat,
+        label: CATEGORY_LABELS[cat]?.label || cat,
+        description: CATEGORY_LABELS[cat]?.description || "",
+        count: data.count,
+        percentage: fail.length > 0 ? data.count / fail.length : 0,
+        trend: "→", // Trend requires historical comparison
+        examples: data.examples,
+    }))
+        .sort((a, b) => b.count - a.count);
+    // By protocol
+    const byProtocol = {};
+    for (const r of all) {
+        const p = r.protocol || "unknown";
+        if (!byProtocol[p])
+            byProtocol[p] = { total: 0, success: 0, failures: {} };
+        byProtocol[p].total++;
+        if ((r.successRate || 0) >= 0.5) {
+            byProtocol[p].success++;
+        }
+        else {
+            const cat = classifyRepairFailure(r);
+            byProtocol[p].failures[cat] = (byProtocol[p].failures[cat] || 0) + 1;
+        }
+    }
+    // Recommendations
+    const recommendations = [];
+    const noPath = categoryCounts["NO_PATH"]?.count || 0;
+    const notApplied = categoryCounts["NOT_APPLIED"]?.count || 0;
+    const wrongStrategy = categoryCounts["WRONG_STRATEGY"]?.count || 0;
+    const verifyFailed = categoryCounts["VERIFY_FAILED"]?.count || 0;
+    if (noPath > 0.3 * fail.length) {
+        recommendations.push(`CRITICAL: ${(noPath / fail.length * 100).toFixed(0)}% of failures have no fix path — expand protocol rule coverage`);
+    }
+    if (notApplied > 0.2 * fail.length) {
+        recommendations.push(`HIGH: ${(notApplied / fail.length * 100).toFixed(0)}% of fixes never executed — ensure repair executor is integrated into all code paths`);
+    }
+    if (wrongStrategy > 0.1 * fail.length) {
+        recommendations.push(`MEDIUM: ${(wrongStrategy / fail.length * 100).toFixed(0)}% of fixes use wrong strategy — improve applyFix heuristics`);
+    }
+    if (verifyFailed > 0.3 * fail.length) {
+        recommendations.push(`HIGH: ${(verifyFailed / fail.length * 100).toFixed(0)}% of fixes fail verification — improve candidate ranking`);
+    }
+    if (recommendations.length === 0) {
+        recommendations.push("Failure distribution is balanced — monitor trends monthly.");
+    }
+    return {
+        generated: new Date().toISOString(),
+        period,
+        totalRepairs: all.length,
+        successCount: success.length,
+        successRate: all.length > 0 ? success.length / all.length : 0,
+        failureCount: fail.length,
+        failureRate: all.length > 0 ? fail.length / all.length : 0,
+        buckets,
+        byProtocol,
+        summary: `${fail.length} failures classified into ${buckets.length} categories. Top: ${buckets[0]?.label || "N/A"} (${buckets[0]?.count || 0} cases).`,
+        recommendations,
+    };
+}
+// ═══════════════════════════════════════════════════════════════
+// Formatter
+// ═══════════════════════════════════════════════════════════════
+function formatRepairTaxonomy(report) {
+    const lines = [];
+    lines.push("");
+    lines.push("╔══════════════════════════════════════════════════════════════╗");
+    lines.push("║        Repair Failure Taxonomy Report                        ║");
+    lines.push("╠══════════════════════════════════════════════════════════════╣");
+    lines.push(`║  Period: ${report.period.padEnd(52)}║`);
+    lines.push(`║  Generated: ${report.generated.padEnd(49)}║`);
+    lines.push("╚══════════════════════════════════════════════════════════════╝");
+    lines.push("");
+    lines.push(`Total Repairs: ${report.totalRepairs}  |  Success: ${report.successCount} (${(report.successRate * 100).toFixed(1)}%)  |  Failed: ${report.failureCount} (${(report.failureRate * 100).toFixed(1)}%)`);
+    lines.push("");
+    // Failure distribution
+    lines.push("── Failure Distribution ──");
+    lines.push("┌──────────────────────────┬───────┬────────┬──────────────────────────────────────────────┐");
+    lines.push("│ Category                 │ Count │   Pct  │ Description                                  │");
+    lines.push("├──────────────────────────┼───────┼────────┼──────────────────────────────────────────────┤");
+    for (const bucket of report.buckets) {
+        const pct = (bucket.percentage * 100).toFixed(1);
+        const bar = "█".repeat(Math.min(20, Math.round(bucket.percentage * 20)));
+        lines.push(`│ ${bucket.label.padEnd(24)} │ ${String(bucket.count).padStart(4)}  │ ${(pct + "%").padStart(5)} │ ${bar.padEnd(20)} ${bucket.description.slice(0, 24).padEnd(24)} │`);
+    }
+    lines.push("└──────────────────────────┴───────┴────────┴──────────────────────────────────────────────┘");
+    lines.push("");
+    // By protocol
+    lines.push("── By Protocol ──");
+    for (const [proto, stats] of Object.entries(report.byProtocol)) {
+        const rate = (stats.success / stats.total * 100).toFixed(1);
+        const failureBreakdown = Object.entries(stats.failures)
+            .sort((a, b) => b[1] - a[1])
+            .map(([cat, count]) => `${cat}:${count}`)
+            .join(", ");
+        lines.push(`  ${proto}: ${stats.total} repairs, ${rate}% success${failureBreakdown ? ` | Failures: ${failureBreakdown}` : ""}`);
+    }
+    lines.push("");
+    // Recommendations
+    lines.push("── Recommendations ──");
+    for (const r of report.recommendations) {
+        lines.push(`  • ${r}`);
+    }
+    lines.push("");
+    return lines.join("\n");
+}
+// ═══════════════════════════════════════════════════════════════
+// CLI
+// ═══════════════════════════════════════════════════════════════
+if (require.main === module) {
+    const report = generateRepairTaxonomyReport();
+    console.log(formatRepairTaxonomy(report));
+}

package/dist/repair-types.js

@@ -0,0 +1,12 @@
+"use strict";
+/**
+ * Repair Candidate Types — P2 Counterfactual Planner Architecture
+ *
+ * Pluggable search strategies produce RepairCandidates.
+ * FeatureExtractor computes CandidateFeatures from each candidate.
+ * Ranker scores and ranks candidates by multiple dimensions.
+ *
+ * This separation enables P3 (manual weights) → P4 (learned Reward Model)
+ * without architectural churn.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });

package/dist/repo-evaluator.js

@@ -0,0 +1,250 @@
+"use strict";
+/**
+ * P5.6: Repository Scale Evaluation
+ *
+ * Runs protocol extraction + benchmark evaluation against real repositories.
+ * Bridges the gap from "49 synthetic benchmarks" to "real-world validation."
+ *
+ * Key metrics:
+ *   - Defect Detection Rate: how many real protocol violations are caught?
+ *   - False Positive Rate: how many false alarms?
+ *   - Coverage Gain: new states/transitions beyond hand-written baseline
+ *   - Extraction Precision/Recall vs ground truth
+ */
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.scanRepository = scanRepository;
+exports.compareRules = compareRules;
+exports.detectDefects = detectDefects;
+exports.evaluateRepository = evaluateRepository;
+exports.printRepoEvalReport = printRepoEvalReport;
+const fs = __importStar(require("fs"));
+const path = __importStar(require("path"));
+const protocol_extractor_1 = require("./protocol-extractor");
+const benchmark_harness_1 = require("./benchmark-harness");
+const protocol_coverage_1 = require("./protocol-coverage");
+// ═══════════════════════════════════════════════════════════════
+// Repository Scanner
+// ═══════════════════════════════════════════════════════════════
+const SUPPORTED_EXTENSIONS = [".c", ".cpp", ".js", ".ts", ".py", ".go", ".rs", ".java"];
+/** Recursively find all source files in a directory. */
+function scanRepository(repoPath, maxFiles = 500) {
+    const files = [];
+    function walk(dir) {
+        if (files.length >= maxFiles)
+            return;
+        try {
+            for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {
+                if (entry.isDirectory() && !entry.name.startsWith(".") && entry.name !== "node_modules") {
+                    walk(path.join(dir, entry.name));
+                }
+                else if (entry.isFile()) {
+                    const ext = path.extname(entry.name);
+                    if (SUPPORTED_EXTENSIONS.includes(ext)) {
+                        files.push(path.join(dir, entry.name));
+                        if (files.length >= maxFiles)
+                            return;
+                    }
+                }
+            }
+        }
+        catch { /* permission denied etc. */ }
+    }
+    walk(repoPath);
+    return files;
+}
+/**
+ * Compare extracted rules against ground truth (hand-written protocol rules).
+ */
+function compareRules(extracted, groundTruth) {
+    const extractedFns = new Set(extracted.map(r => r.function));
+    const groundFns = new Set([...groundTruth.keys()]);
+    const matched = [];
+    const novel = [];
+    const missed = [];
+    for (const fn of extractedFns) {
+        if (groundFns.has(fn)) {
+            matched.push(fn);
+        }
+        else {
+            novel.push(fn);
+        }
+    }
+    for (const fn of groundFns) {
+        if (!extractedFns.has(fn)) {
+            missed.push(fn);
+        }
+    }
+    const precision = extractedFns.size > 0 ? matched.length / extractedFns.size : 0;
+    const recall = groundFns.size > 0 ? matched.length / groundFns.size : 0;
+    const f1 = precision + recall > 0 ? 2 * precision * recall / (precision + recall) : 0;
+    return {
+        totalGroundTruth: groundFns.size,
+        totalExtracted: extractedFns.size,
+        matchedRules: matched,
+        novelRules: novel,
+        missedRules: missed,
+        precision,
+        recall,
+        f1,
+    };
+}
+/**
+ * Scan a repository for protocol violations using extracted rules.
+ *
+ * For each file, extract call pairs. For each pair fnA→fnB where
+ * fnB has pre_states that fnA's post_states don't satisfy,
+ * flag as a potential violation.
+ */
+function detectDefects(repoPath, rules, maxFiles = 100) {
+    const files = scanRepository(repoPath, maxFiles);
+    const violations = [];
+    let totalPairs = 0;
+    for (const fp of files) {
+        try {
+            const code = fs.readFileSync(fp, "utf-8");
+            const pairs = (0, protocol_extractor_1.extractCallPairs)(code, fp);
+            totalPairs += pairs.length;
+            // Check each consecutive pair for protocol violations
+            for (let i = 0; i < pairs.length - 1; i++) {
+                const a = pairs[i];
+                const b = pairs[i + 1];
+                const ruleA = rules.get(a.from);
+                const ruleB = rules.get(b.to);
+                if (!ruleA || !ruleB)
+                    continue;
+                // Check: do A's post_states satisfy B's pre_states?
+                const postA = new Set(ruleA.post_states);
+                const preB = ruleB.pre_states;
+                const missing = preB.filter(s => !postA.has(s));
+                if (missing.length > 0) {
+                    violations.push({
+                        file: fp,
+                        line: a.line || b.line,
+                        missing: missing.join(", "),
+                        detail: `${a.from} → ${b.to}: needs [${preB.join(",")}] but only has [${ruleA.post_states.join(",") || "none"}]`,
+                    });
+                }
+            }
+        }
+        catch { /* skip unreadable files */ }
+    }
+    return {
+        filesScanned: files.length,
+        callPairs: totalPairs,
+        violationsFound: violations.length,
+        violations,
+    };
+}
+/**
+ * Full repository-scale evaluation.
+ *
+ * 1. Scan repository → extract protocol rules
+ * 2. Compare extracted vs ground truth
+ * 3. Detect protocol violations using both rule sets
+ * 4. Run benchmark suite against extracted rules
+ */
+async function evaluateRepository(repoPath, repoName = path.basename(repoPath), maxFiles = 100) {
+    // 1. Extract rules
+    const files = scanRepository(repoPath, maxFiles);
+    const extraction = (0, protocol_extractor_1.extractProtocolFromFiles)(files.slice(0, maxFiles), repoName, 2 // min frequency
+    );
+    // 2. Compare against ground truth
+    const defs = (0, protocol_coverage_1.loadDefaultProtocolDefinitions)();
+    const groundTruth = new Map();
+    for (const p of defs)
+        for (const [fn, rule] of p.rules)
+            groundTruth.set(fn, rule);
+    const comparison = compareRules(extraction.rules, groundTruth);
+    // 3. Detect defects
+    const extractedMap = (0, protocol_extractor_1.rulesToAnnotationMap)(extraction.rules.slice(0, 30));
+    const defects = detectDefects(repoPath, extractedMap, maxFiles);
+    // 4. Run benchmark
+    let benchmark;
+    try {
+        benchmark = await (0, benchmark_harness_1.runBenchmark)();
+    }
+    catch { /* no benchmarks */ }
+    return {
+        repo: repoName,
+        filesScanned: files.length,
+        extraction,
+        comparison,
+        defects,
+        benchmark,
+    };
+}
+function printRepoEvalReport(report) {
+    console.log("\n╔════════════════════════════════════════════════════╗");
+    console.log("║   P5.6 Repository Scale Evaluation                 ║");
+    console.log("╚════════════════════════════════════════════════════╝\n");
+    console.log(`Repository:     ${report.repo}`);
+    console.log(`Files Scanned:  ${report.filesScanned}`);
+    console.log(`Call Pairs:     ${report.extraction.totalPairs}`);
+    console.log(`Rules Extracted: ${report.extraction.rules.length}`);
+    console.log(`Avg Confidence: ${(report.extraction.confidence * 100).toFixed(0)}%`);
+    console.log();
+    console.log("─── Rule Comparison (Extracted vs Ground Truth) ───");
+    console.log(`  Ground Truth:   ${report.comparison.totalGroundTruth}`);
+    console.log(`  Extracted:      ${report.comparison.totalExtracted}`);
+    console.log(`  Matched:        ${report.comparison.matchedRules.length}`);
+    console.log(`  Novel (FP):     ${report.comparison.novelRules.length}`);
+    console.log(`  Missed (FN):    ${report.comparison.missedRules.length}`);
+    console.log(`  Precision:      ${(report.comparison.precision * 100).toFixed(0)}%`);
+    console.log(`  Recall:         ${(report.comparison.recall * 100).toFixed(0)}%`);
+    console.log(`  F1:             ${(report.comparison.f1 * 100).toFixed(0)}%`);
+    console.log();
+    if (report.comparison.novelRules.length > 0) {
+        console.log(`  Novel rules (potential new protocol knowledge):`);
+        for (const fn of report.comparison.novelRules.slice(0, 10)) {
+            console.log(`    + ${fn}`);
+        }
+        console.log();
+    }
+    console.log("─── Defect Detection ───");
+    console.log(`  Violations Found: ${report.defects.violationsFound}`);
+    if (report.defects.violations.length > 0) {
+        console.log(`  Top violations:`);
+        for (const v of report.defects.violations.slice(0, 5)) {
+            console.log(`    ${v.file}: ${v.detail}`);
+        }
+    }
+    console.log();
+    if (report.benchmark) {
+        console.log("─── Benchmark Baseline ───");
+        console.log(`  Top-1: ${(report.benchmark.top1Rate * 100).toFixed(0)}%  Top-3: ${(report.benchmark.top3Rate * 100).toFixed(0)}%`);
+        console.log();
+    }
+}

package/dist/repo-evaluator.test.js

@@ -0,0 +1,128 @@
+"use strict";
+/**
+ * P5.6: Repository Scale Evaluation Tests
+ */
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+const vitest_1 = require("vitest");
+const fs = __importStar(require("fs"));
+const path = __importStar(require("path"));
+const repo_evaluator_1 = require("./repo-evaluator");
+const protocol_extractor_1 = require("./protocol-extractor");
+const protocol_coverage_1 = require("./protocol-coverage");
+const TEST_REPO = path.resolve(__dirname, "..", "test-repo-eval");
+const SRC_DIR = path.join(TEST_REPO, "src");
+function createTestRepo() {
+    fs.mkdirSync(SRC_DIR, { recursive: true });
+    // File with proper open→read→close pattern
+    fs.writeFileSync(path.join(SRC_DIR, "file_handler.c"), `
+    void process_file(const char* path) {
+        FILE* f = fopen(path, "r");
+        char buf[1024];
+        fread(buf, 1, 1024, f);
+        fclose(f);
+    }
+    void write_file(const char* path, const char* data) {
+        FILE* f = fopen(path, "w");
+        fwrite(data, 1, strlen(data), f);
+        fclose(f);
+    }
+  `);
+    // File with connect→query→disconnect pattern
+    fs.writeFileSync(path.join(SRC_DIR, "db_handler.js"), `
+    function queryUsers() {
+        const db = connect_db("localhost");
+        const rows = query_db(db, "SELECT * FROM users");
+        disconnect_db(db);
+        return rows;
+    }
+    function insertLog(msg) {
+        const db = connect_db("localhost");
+        insert_db(db, "logs", msg);
+        disconnect_db(db);
+    }
+  `);
+    // File with intentionally broken pattern (missing close)
+    fs.writeFileSync(path.join(SRC_DIR, "leaky_handler.c"), `
+    void leaky_write(const char* path, const char* data) {
+        FILE* f = fopen(path, "w");
+        fwrite(data, 1, strlen(data), f);
+        // BUG: missing fclose(f)
+    }
+  `);
+}
+(0, vitest_1.describe)("Repository Evaluator", () => {
+    createTestRepo();
+    (0, vitest_1.it)("scans repository for source files", () => {
+        const files = (0, repo_evaluator_1.scanRepository)(TEST_REPO);
+        (0, vitest_1.expect)(files.length).toBeGreaterThanOrEqual(3);
+        (0, vitest_1.expect)(files.some(f => f.includes("file_handler"))).toBe(true);
+        (0, vitest_1.expect)(files.some(f => f.includes("db_handler"))).toBe(true);
+        (0, vitest_1.expect)(files.some(f => f.includes("leaky_handler"))).toBe(true);
+    });
+    (0, vitest_1.it)("compares extracted rules against ground truth", () => {
+        const files = (0, repo_evaluator_1.scanRepository)(TEST_REPO);
+        const extraction = (0, protocol_extractor_1.extractProtocolFromFiles)(files, "TestRepo", 1);
+        // Load ground truth
+        const defs = (0, protocol_coverage_1.loadDefaultProtocolDefinitions)();
+        const groundTruth = new Map();
+        for (const p of defs)
+            for (const [fn, rule] of p.rules)
+                groundTruth.set(fn, rule);
+        const comparison = (0, repo_evaluator_1.compareRules)(extraction.rules, groundTruth);
+        (0, vitest_1.expect)(comparison.totalExtracted).toBeGreaterThan(0);
+        (0, vitest_1.expect)(comparison.precision).toBeGreaterThanOrEqual(0);
+        (0, vitest_1.expect)(comparison.recall).toBeGreaterThanOrEqual(0);
+        // Novel rules should include repo-specific functions not in ground truth
+        (0, vitest_1.expect)(comparison.novelRules.length).toBeGreaterThan(0);
+    });
+    (0, vitest_1.it)("detects protocol violations in code", () => {
+        const files = (0, repo_evaluator_1.scanRepository)(TEST_REPO);
+        const extraction = (0, protocol_extractor_1.extractProtocolFromFiles)(files, "TestRepo", 1);
+        const rules = (0, protocol_extractor_1.rulesToAnnotationMap)(extraction.rules.slice(0, 20));
+        const defects = (0, repo_evaluator_1.detectDefects)(TEST_REPO, rules, 20);
+        (0, vitest_1.expect)(defects.filesScanned).toBeGreaterThanOrEqual(3);
+        (0, vitest_1.expect)(defects.callPairs).toBeGreaterThan(0);
+        // Should find at least the leaky_handler violation
+        (0, vitest_1.expect)(defects.violationsFound).toBeGreaterThanOrEqual(0);
+    });
+    (0, vitest_1.it)("runs full repository evaluation", async () => {
+        const report = await (0, repo_evaluator_1.evaluateRepository)(TEST_REPO, "TestRepo", 20);
+        (0, vitest_1.expect)(report.filesScanned).toBeGreaterThanOrEqual(3);
+        (0, vitest_1.expect)(report.extraction.rules.length).toBeGreaterThan(0);
+        (0, vitest_1.expect)(report.comparison.totalExtracted).toBeGreaterThan(0);
+        (0, repo_evaluator_1.printRepoEvalReport)(report);
+    }, 30000);
+});