recallnest 2.2.0 → 2.3.0

package/scripts/mlp-health-check.ts

@@ -1,382 +0,0 @@
-/**
- * memory-lancedb-pro 通用记忆体检脚本
- * 只读分析 LanceDB 数据，不写入任何内容
- * 兼容 RecallNest (camelCase) 和 memory-lancedb-pro (snake_case) metadata 格式
- *
- * 用法：
- *   bun run scripts/mlp-health-check.ts <lancedb-dir> [bot-name]
- *
- * 示例：
- *   bun run scripts/mlp-health-check.ts /tmp/mlp-healthcheck/antigravity/lancedb-pro AntiBot
- *   bun run scripts/mlp-health-check.ts ./data/lancedb RecallNest
- */
-
-import lancedb from "@lancedb/lancedb";
-
-const DB_PATH = process.argv[2] || "./data/lancedb";
-const BOT_NAME = process.argv[3] || "未命名";
-const TABLE_NAME = "memories";
-
-interface MemoryRow {
-  id: string;
-  text: string;
-  vector: number[];
-  category: string;
-  scope: string;
-  importance: number;
-  timestamp: number;
-  metadata?: string;
-}
-
-/** 兼容 camelCase (RecallNest) 和 snake_case (memory-lancedb-pro) */
-function getAccessCount(meta: Record<string, any>): number {
-  return meta.accessCount ?? meta.access_count ?? 0;
-}
-
-function getLastAccessed(meta: Record<string, any>): number {
-  return meta.lastAccessedAt ?? meta.last_accessed_at ?? 0;
-}
-
-function getTier(meta: Record<string, any>): string {
-  return meta.tier ?? "unknown";
-}
-
-function getSource(meta: Record<string, any>): string {
-  return meta.source ?? meta.source_session ?? "";
-}
-
-function cosineSimilarity(a: number[], b: number[]): number {
-  let dot = 0, normA = 0, normB = 0;
-  for (let i = 0; i < a.length; i++) {
-    dot += a[i] * b[i];
-    normA += a[i] * a[i];
-    normB += b[i] * b[i];
-  }
-  return dot / (Math.sqrt(normA) * Math.sqrt(normB));
-}
-
-function formatDate(ts: number): string {
-  return new Date(ts).toISOString().slice(0, 10);
-}
-
-function percentBar(pct: number, width = 30): string {
-  const filled = Math.round(pct / 100 * width);
-  return "█".repeat(filled) + "░".repeat(width - filled);
-}
-
-async function main() {
-  console.log(`\n🏥 记忆体检报告 — ${BOT_NAME}`);
-  console.log("=".repeat(60));
-  console.log(`📅 检查时间: ${new Date().toISOString().slice(0, 19)}`);
-  console.log(`📂 数据路径: ${DB_PATH}\n`);
-
-  const db = await lancedb.connect(DB_PATH);
-  const table = await db.openTable(TABLE_NAME);
-
-  // Fetch all rows (without vectors first for speed)
-  console.log("⏳ 正在读取全量数据...");
-  const allRows: MemoryRow[] = await table
-    .query()
-    .select(["id", "text", "category", "scope", "importance", "timestamp", "metadata"])
-    .toArray() as any;
-
-  const total = allRows.length;
-  if (total === 0) {
-    console.log("⚠️  记忆库为空，无数据可分析。");
-    return;
-  }
-  console.log(`✅ 读取完成: ${total.toLocaleString()} 条记忆\n`);
-
-  // Parse metadata once
-  const parsedRows = allRows.map(row => ({
-    ...row,
-    meta: row.metadata ? JSON.parse(row.metadata) : {},
-  }));
-
-  // ============================================
-  // 1. Category 分布
-  // ============================================
-  console.log("📊 1. Category 分布");
-  console.log("-".repeat(60));
-  const catMap = new Map<string, number>();
-  for (const row of parsedRows) {
-    catMap.set(row.category, (catMap.get(row.category) || 0) + 1);
-  }
-  const catSorted = [...catMap.entries()].sort((a, b) => b[1] - a[1]);
-  for (const [cat, count] of catSorted) {
-    const pct = (count / total * 100).toFixed(1);
-    console.log(`  ${cat.padEnd(14)} ${count.toString().padStart(6)}  (${pct.padStart(5)}%)  ${percentBar(parseFloat(pct))}`);
-  }
-
-  // ============================================
-  // 2. Tier 分布
-  // ============================================
-  console.log("\n📊 2. Tier 分布");
-  console.log("-".repeat(60));
-  const tierMap = new Map<string, number>();
-  for (const row of parsedRows) {
-    const tier = getTier(row.meta);
-    tierMap.set(tier, (tierMap.get(tier) || 0) + 1);
-  }
-  const tierSorted = [...tierMap.entries()].sort((a, b) => b[1] - a[1]);
-  for (const [tier, count] of tierSorted) {
-    const pct = (count / total * 100).toFixed(1);
-    console.log(`  ${tier.padEnd(14)} ${count.toString().padStart(6)}  (${pct.padStart(5)}%)  ${percentBar(parseFloat(pct))}`);
-  }
-
-  // ============================================
-  // 3. 死记忆分析
-  // ============================================
-  console.log("\n📊 3. 死记忆分析（accessCount = 0 或无记录）");
-  console.log("-".repeat(60));
-  let deadCount = 0;
-  let aliveCount = 0;
-  let totalAccess = 0;
-  let maxAccess = 0;
-  let maxAccessText = "";
-  const accessDistribution = new Map<string, number>();
-
-  for (const row of parsedRows) {
-    const ac = getAccessCount(row.meta);
-    totalAccess += ac;
-
-    if (ac === 0) deadCount++;
-    else aliveCount++;
-
-    if (ac > maxAccess) {
-      maxAccess = ac;
-      maxAccessText = row.text.slice(0, 80);
-    }
-
-    let bucket: string;
-    if (ac === 0) bucket = "0 (死记忆)";
-    else if (ac <= 2) bucket = "1-2";
-    else if (ac <= 5) bucket = "3-5";
-    else if (ac <= 10) bucket = "6-10";
-    else if (ac <= 50) bucket = "11-50";
-    else bucket = "50+";
-    accessDistribution.set(bucket, (accessDistribution.get(bucket) || 0) + 1);
-  }
-
-  const deadPct = (deadCount / total * 100).toFixed(1);
-  const alivePct = (aliveCount / total * 100).toFixed(1);
-  console.log(`  死记忆:   ${deadCount.toLocaleString()} / ${total.toLocaleString()}  (${deadPct}%)`);
-  console.log(`  活记忆:   ${aliveCount.toLocaleString()} / ${total.toLocaleString()}  (${alivePct}%)`);
-  console.log(`  总访问次: ${totalAccess.toLocaleString()}`);
-  console.log(`  平均访问: ${(totalAccess / total).toFixed(2)} 次/条`);
-  if (maxAccess > 0) {
-    console.log(`  最热记忆: [${maxAccess}次] ${maxAccessText}...`);
-  }
-
-  console.log("\n  访问次数分布:");
-  const bucketOrder = ["0 (死记忆)", "1-2", "3-5", "6-10", "11-50", "50+"];
-  for (const bucket of bucketOrder) {
-    const count = accessDistribution.get(bucket) || 0;
-    const pct = (count / total * 100).toFixed(1);
-    console.log(`    ${bucket.padEnd(14)} ${count.toString().padStart(6)}  (${pct.padStart(5)}%)  ${percentBar(parseFloat(pct))}`);
-  }
-
-  // ============================================
-  // 4. Importance 分布
-  // ============================================
-  console.log("\n📊 4. Importance 分布");
-  console.log("-".repeat(60));
-  const impBuckets = new Map<string, number>();
-  let impSum = 0;
-  for (const row of parsedRows) {
-    const imp = row.importance || 0;
-    impSum += imp;
-    let bucket: string;
-    if (imp < 0.2) bucket = "0.0-0.2 (低)";
-    else if (imp < 0.4) bucket = "0.2-0.4";
-    else if (imp < 0.6) bucket = "0.4-0.6 (中)";
-    else if (imp < 0.8) bucket = "0.6-0.8";
-    else bucket = "0.8-1.0 (高)";
-    impBuckets.set(bucket, (impBuckets.get(bucket) || 0) + 1);
-  }
-  console.log(`  平均 importance: ${(impSum / total).toFixed(3)}`);
-  const impOrder = ["0.0-0.2 (低)", "0.2-0.4", "0.4-0.6 (中)", "0.6-0.8", "0.8-1.0 (高)"];
-  for (const bucket of impOrder) {
-    const count = impBuckets.get(bucket) || 0;
-    const pct = (count / total * 100).toFixed(1);
-    console.log(`    ${bucket.padEnd(16)} ${count.toString().padStart(6)}  (${pct.padStart(5)}%)  ${percentBar(parseFloat(pct))}`);
-  }
-
-  // ============================================
-  // 5. Scope 分布
-  // ============================================
-  console.log("\n📊 5. Scope 分布");
-  console.log("-".repeat(60));
-  const scopeMap = new Map<string, number>();
-  for (const row of parsedRows) {
-    const s = row.scope || "(empty)";
-    scopeMap.set(s, (scopeMap.get(s) || 0) + 1);
-  }
-  const scopeSorted = [...scopeMap.entries()].sort((a, b) => b[1] - a[1]);
-  console.log(`  独立 scope 数: ${scopeMap.size}`);
-  console.log("  Top 10:");
-  for (const [scope, count] of scopeSorted.slice(0, 10)) {
-    const pct = (count / total * 100).toFixed(1);
-    const label = scope.length > 30 ? scope.slice(0, 27) + "..." : scope;
-    console.log(`    ${label.padEnd(32)} ${count.toString().padStart(6)}  (${pct.padStart(5)}%)`);
-  }
-
-  // ============================================
-  // 6. 年龄分布
-  // ============================================
-  console.log("\n📊 6. 年龄分布");
-  console.log("-".repeat(60));
-  const now = Date.now();
-  const ageBuckets = new Map<string, number>();
-  let oldest = Infinity, newest = 0;
-  for (const row of parsedRows) {
-    const ts = row.timestamp;
-    if (ts < oldest) oldest = ts;
-    if (ts > newest) newest = ts;
-    const ageHours = (now - ts) / (1000 * 60 * 60);
-    let bucket: string;
-    if (ageHours < 24) bucket = "< 1 天";
-    else if (ageHours < 24 * 7) bucket = "1-7 天";
-    else if (ageHours < 24 * 30) bucket = "7-30 天";
-    else if (ageHours < 24 * 90) bucket = "30-90 天";
-    else bucket = "> 90 天";
-    ageBuckets.set(bucket, (ageBuckets.get(bucket) || 0) + 1);
-  }
-  console.log(`  最早: ${formatDate(oldest)}`);
-  console.log(`  最新: ${formatDate(newest)}`);
-  const ageOrder = ["< 1 天", "1-7 天", "7-30 天", "30-90 天", "> 90 天"];
-  for (const bucket of ageOrder) {
-    const count = ageBuckets.get(bucket) || 0;
-    const pct = (count / total * 100).toFixed(1);
-    console.log(`    ${bucket.padEnd(12)} ${count.toString().padStart(6)}  (${pct.padStart(5)}%)  ${percentBar(parseFloat(pct))}`);
-  }
-
-  // ============================================
-  // 7. 近似重复检测（采样）
-  // ============================================
-  const sampleSize = Math.min(500, total);
-  console.log(`\n📊 7. 近似重复检测（采样 ${sampleSize} 条，cosine > 0.95）`);
-  console.log("-".repeat(60));
-  console.log("  ⏳ 正在读取向量数据（采样）...");
-
-  const sampleRows: MemoryRow[] = await table
-    .query()
-    .select(["id", "text", "vector", "category"])
-    .limit(sampleSize)
-    .toArray() as any;
-
-  let dupPairs = 0;
-  const dupExamples: Array<{sim: number, textA: string, textB: string}> = [];
-
-  for (let i = 0; i < sampleRows.length; i++) {
-    for (let j = i + 1; j < sampleRows.length; j++) {
-      if (sampleRows[i].vector && sampleRows[j].vector) {
-        const sim = cosineSimilarity(sampleRows[i].vector, sampleRows[j].vector);
-        if (sim > 0.95) {
-          dupPairs++;
-          if (dupExamples.length < 3) {
-            dupExamples.push({
-              sim,
-              textA: sampleRows[i].text.slice(0, 60),
-              textB: sampleRows[j].text.slice(0, 60),
-            });
-          }
-        }
-      }
-    }
-  }
-
-  const totalPairs = sampleSize * (sampleSize - 1) / 2;
-  const dupRate = (dupPairs / totalPairs * 100).toFixed(3);
-  console.log(`  采样对数:   ${totalPairs.toLocaleString()}`);
-  console.log(`  高相似对:   ${dupPairs} (${dupRate}%)`);
-  if (total > sampleSize) {
-    console.log(`  推算全库:   ~${Math.round(dupPairs / totalPairs * total * (total-1) / 2).toLocaleString()} 对潜在重复`);
-  }
-
-  if (dupExamples.length > 0) {
-    console.log("\n  示例重复对:");
-    for (const ex of dupExamples) {
-      console.log(`    [sim=${ex.sim.toFixed(3)}]`);
-      console.log(`      A: ${ex.textA}...`);
-      console.log(`      B: ${ex.textB}...`);
-    }
-  }
-
-  // ============================================
-  // 8. 综合诊断
-  // ============================================
-  console.log("\n" + "=".repeat(60));
-  console.log(`🩺 综合诊断 — ${BOT_NAME}`);
-  console.log("=".repeat(60));
-
-  const issues: string[] = [];
-  const healthy: string[] = [];
-
-  // Total count check
-  if (total < 10) {
-    issues.push(`🟡 记忆总量仅 ${total} 条，数据量偏少`);
-  } else {
-    healthy.push(`✅ 记忆总量 ${total.toLocaleString()} 条`);
-  }
-
-  // Dead memory ratio
-  const deadRatio = deadCount / total;
-  if (deadRatio > 0.9) {
-    issues.push(`🔴 ${deadPct}% 记忆从未被访问（死记忆率过高）`);
-  } else if (deadRatio > 0.7) {
-    issues.push(`🟡 ${deadPct}% 记忆从未被访问（死记忆率偏高）`);
-  } else {
-    healthy.push(`✅ 死记忆率 ${deadPct}% 在合理范围`);
-  }
-
-  // Category balance — check if any single legacy category dominates
-  const factCount = catMap.get("fact") || 0;
-  const decisionCount = catMap.get("decision") || 0;
-  const legacyRatio = (factCount + decisionCount) / total;
-  if (legacyRatio > 0.5) {
-    issues.push(`🟡 legacy 类别（fact/decision）占 ${(legacyRatio*100).toFixed(1)}%，建议迁移到 6-category 体系`);
-  }
-  // Check structural category coverage
-  const structuredCount = (catMap.get("entities") || 0) + (catMap.get("patterns") || 0) + (catMap.get("cases") || 0);
-  if (structuredCount === 0 && total > 20) {
-    issues.push(`🟡 无结构化记忆（entities/patterns/cases 均为 0）`);
-  } else if (structuredCount > 0) {
-    healthy.push(`✅ 结构化记忆 ${structuredCount} 条（entities+patterns+cases）`);
-  }
-
-  // Tier distribution
-  const coreCount = tierMap.get("core") || 0;
-  const unknownTier = tierMap.get("unknown") || 0;
-  if (unknownTier / total > 0.5) {
-    issues.push(`🟡 ${(unknownTier/total*100).toFixed(1)}% 记忆无 tier 标记`);
-  }
-  if (coreCount < 5 && total > 50) {
-    issues.push(`🟡 core tier 仅 ${coreCount} 条，高价值记忆太少`);
-  } else if (coreCount >= 5) {
-    healthy.push(`✅ core tier 有 ${coreCount} 条高价值记忆`);
-  }
-
-  // Importance differentiation
-  const modeImpBucket = [...impBuckets.entries()].sort((a, b) => b[1] - a[1])[0];
-  if (modeImpBucket && modeImpBucket[1] / total > 0.9) {
-    issues.push(`🟡 ${(modeImpBucket[1]/total*100).toFixed(1)}% importance 集中在 ${modeImpBucket[0]}，区分度不足`);
-  } else {
-    healthy.push(`✅ importance 分布有区分度`);
-  }
-
-  // Dup rate
-  if (parseFloat(dupRate) > 1) {
-    issues.push(`🟡 采样重复率 ${dupRate}%，全库可能存在大量近似重复`);
-  } else {
-    healthy.push(`✅ 采样重复率 ${dupRate}%，重复可控`);
-  }
-
-  for (const h of healthy) console.log(`  ${h}`);
-  for (const i of issues) console.log(`  ${i}`);
-
-  console.log("\n" + "=".repeat(60));
-  console.log("📋 体检完成\n");
-}
-
-main().catch(console.error);