@tekmidian/pai 0.37.0 → 0.39.0

package/dist/{indexer-backend-Cox9BCo-.mjs → indexer-backend-TG64CCQC.mjs}

@@ -296,4 +296,4 @@ async function indexAllWithBackend(backend, registryDb) {
 
 //#endregion
 export { embedChunksWithBackend, indexAllWithBackend };
-//# sourceMappingURL=indexer-backend-Cox9BCo-.mjs.map
+//# sourceMappingURL=indexer-backend-TG64CCQC.mjs.map

package/dist/{indexer-backend-Cox9BCo-.mjs.map → indexer-backend-TG64CCQC.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"indexer-backend-Cox9BCo-.mjs","names":[],"sources":["../src/memory/indexer/async.ts"],"sourcesContent":["/**\n * Backend-aware async indexer for PAI federation memory.\n *\n * Provides the same functionality as sync.ts but writes through the\n * StorageBackend interface instead of directly to better-sqlite3.\n * Used when the daemon is configured with the Postgres backend.\n *\n * The SQLite path still uses sync.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult } from \"./types.js\";\n\nexport type { IndexResult };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file through the StorageBackend interface.\n * Returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport async function indexFileWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): Promise<boolean> {\n  const absPath = join(rootPath, relativePath);\n\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Change detection\n  const existingHash = await backend.getFileHash(projectId, relativePath);\n  if (existingHash === hash) return false;\n\n  // Delete old chunks\n  await backend.deleteChunksForFile(projectId, relativePath);\n\n  // Chunk the content\n  const rawChunks = chunkMarkdown(content);\n  const updatedAt = Date.now();\n\n  const chunks: ChunkRow[] = rawChunks.map((c, i) => ({\n    id: chunkId(projectId, relativePath, i, c.startLine, c.endLine),\n    projectId,\n    source,\n    tier,\n    path: relativePath,\n    startLine: c.startLine,\n    endLine: c.endLine,\n    hash: c.hash,\n    text: c.text,\n    updatedAt,\n    embedding: null,\n  }));\n\n  // Insert chunks + update file record\n  await backend.insertChunks(chunks);\n  await backend.upsertFile({ projectId, path: relativePath, source, tier, hash, mtime, size });\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexProjectWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files\n  {\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        try {\n          await backend.insertChunks([titleChunk]);\n        } catch {\n          // Skip title chunks that cause backend errors\n        }\n      }\n    }\n\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({ absPath: claudeMemoryMd, rootBase: claudeProjectDir, source: \"memory\", tier: \"evergreen\" });\n      }\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    try {\n      const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n      if (changed) {\n        const ids = await backend.getChunkIds(projectId, relPath);\n        result.filesProcessed++;\n        result.chunksCreated += ids.length;\n      } else {\n        result.filesSkipped++;\n      }\n    } catch {\n      // Skip files that cause backend errors (e.g. null bytes in Postgres)\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);\n\n  const stalePaths: string[] = [];\n  for (const p of dbChunkPaths) {\n    const basePath = p.endsWith(\"::title\") ? p.slice(0, -\"::title\".length) : p;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(p);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    await backend.deletePaths(projectId, stalePaths);\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 1;\n\n/** Default ceiling on one pass. See EmbedPassOptions. */\nconst DEFAULT_MAX_CHUNKS_PER_PASS = 5_000;\n/** Default wall-clock ceiling on one pass, in milliseconds. */\nconst DEFAULT_MAX_MILLIS_PER_PASS = 120_000;\n\nexport interface EmbedPassOptions {\n  /**\n   * Stop the pass after this many chunks. Unbounded passes are the reason a\n   * six-figure backlog starves the indexer: the daemon serialises indexing and\n   * embedding against each other, so a pass that runs for hours means nothing\n   * new gets indexed for hours. Bounding the pass costs nothing — every\n   * embedding is written as it is produced, so the next pass resumes exactly\n   * where this one stopped.\n   */\n  maxChunks?: number;\n  /** Stop the pass after this much wall-clock time, whichever bound hits first. */\n  maxMillis?: number;\n}\n\n/**\n * Generate and store embeddings for unembedded chunks via the StorageBackend.\n *\n * The pass is deliberately bounded (see EmbedPassOptions) and resumable: it\n * takes a slice of the backlog, embeds it, and returns so the scheduler can run\n * an index pass before coming back. Draining the whole backlog in one call\n * starves indexing for as long as the call runs.\n *\n * The optional `shouldStop` callback is checked between every batch. When it\n * returns true the embed loop exits early so the caller (e.g. the daemon\n * shutdown handler) can close the pool without racing against active queries.\n *\n * Returns the number of newly embedded chunks.\n */\nexport async function embedChunksWithBackend(\n  backend: StorageBackend,\n  shouldStop?: () => boolean,\n  projectNames?: Map<number, string>,\n  options?: EmbedPassOptions,\n): Promise<number> {\n  const { generateEmbeddings, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const maxChunks = options?.maxChunks ?? DEFAULT_MAX_CHUNKS_PER_PASS;\n  const maxMillis = options?.maxMillis ?? DEFAULT_MAX_MILLIS_PER_PASS;\n  const deadline = Date.now() + maxMillis;\n\n  const rows = await backend.getUnembeddedChunkIds(undefined, maxChunks);\n  if (rows.length === 0) return 0;\n\n  const total = rows.length;\n  let embedded = 0;\n\n  // Build a summary of what needs embedding: count chunks per project_id\n  const projectChunkCounts = new Map<number, { count: number; samplePath: string }>();\n  for (const row of rows) {\n    const entry = projectChunkCounts.get(row.project_id);\n    if (entry) {\n      entry.count++;\n    } else {\n      projectChunkCounts.set(row.project_id, { count: 1, samplePath: row.path });\n    }\n  }\n  const pName = (pid: number) => projectNames?.get(pid) ?? `project ${pid}`;\n  const projectSummary = Array.from(projectChunkCounts.entries())\n    .map(([pid, { count, samplePath }]) => `  ${pName(pid)}: ${count} chunks (e.g. ${samplePath})`)\n    .join(\"\\n\");\n  process.stderr.write(\n    `[pai-daemon] Embed pass: ${total} unembedded chunks across ${projectChunkCounts.size} project(s)\\n${projectSummary}\\n`\n  );\n\n  // Track current project for transition logging\n  let currentProjectId = -1;\n  let projectEmbedded = 0;\n\n  for (let i = 0; i < rows.length; i += EMBED_BATCH_SIZE) {\n    // Check cancellation between every batch before touching the pool again\n    if (shouldStop?.()) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass cancelled after ${embedded}/${total} chunks (shutdown requested)\\n`\n      );\n      break;\n    }\n\n    // Yield the daemon back to the indexer rather than run past the deadline.\n    // The remaining rows are simply picked up by the next scheduled pass.\n    if (Date.now() >= deadline) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass yielding after ${embedded}/${total} chunks (${maxMillis}ms budget spent); resuming next pass\\n`\n      );\n      break;\n    }\n\n    const batch = rows.slice(i, i + EMBED_BATCH_SIZE);\n\n    // Keep IPC responsive: the forward pass below is synchronous inside the\n    // model, so yield before entering it rather than between chunks.\n    await yieldToEventLoop();\n\n    const vecs = await generateEmbeddings(batch.map((r) => r.text));\n    // Issue the writes together rather than one round-trip at a time. Measured\n    // on this machine the model embeds 40-67 chunks/s in isolation while the\n    // daemon managed ~5/s: the difference was one sequential UPDATE per chunk,\n    // so the pass spent most of its budget waiting on the network, not working.\n    // Concurrency is bounded by the batch size, which the pool handles; the\n    // SQLite backend is synchronous and simply ignores the difference.\n    await Promise.all(\n      batch.map((row, j) => backend.updateEmbedding(row.id, serializeEmbedding(vecs[j])))\n    );\n\n    // Attribute the batch to projects only after it is durably stored, and walk\n    // it in order so a batch straddling a project boundary credits each side\n    // correctly. Rows arrive ordered by project, so a boundary is a real\n    // transition, not the per-chunk flapping this logging used to produce.\n    for (const { project_id, path } of batch) {\n      if (project_id !== currentProjectId) {\n        if (currentProjectId !== -1) {\n          process.stderr.write(\n            `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n          );\n        }\n        const info = projectChunkCounts.get(project_id);\n        process.stderr.write(\n          `[pai-daemon] Embedding ${pName(project_id)} (${info?.count ?? \"?\"} chunks, starting at ${path})\\n`\n        );\n        currentProjectId = project_id;\n        projectEmbedded = 0;\n      }\n      projectEmbedded++;\n    }\n\n    embedded += batch.length;\n\n    // Log progress with current file path for context\n    const lastChunk = batch[batch.length - 1];\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks (${pName(lastChunk.project_id)}: ${lastChunk.path})\\n`\n    );\n  }\n\n  // Log final project completion\n  if (currentProjectId !== -1) {\n    process.stderr.write(\n      `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n    );\n  }\n\n  return embedded;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexAllWithBackend(\n  backend: StorageBackend,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProjectWithBackend(backend, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAuCA,eAAsB,qBACpB,SACA,WACA,UACA,cACA,QACA,MACkB;CAClB,MAAM,UAAU,KAAK,UAAU,aAAa;CAE5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AACN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AAIlB,KADqB,MAAM,QAAQ,YAAY,WAAW,aAAa,KAClD,KAAM,QAAO;AAGlC,OAAM,QAAQ,oBAAoB,WAAW,aAAa;CAG1D,MAAM,YAAY,cAAc,QAAQ;CACxC,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAAqB,UAAU,KAAK,GAAG,OAAO;EAClD,IAAI,QAAQ,WAAW,cAAc,GAAG,EAAE,WAAW,EAAE,QAAQ;EAC/D;EACA;EACA;EACA,MAAM;EACN,WAAW,EAAE;EACb,SAAS,EAAE;EACX,MAAM,EAAE;EACR,MAAM,EAAE;EACR;EACA,WAAW;EACZ,EAAE;AAGH,OAAM,QAAQ,aAAa,OAAO;AAClC,OAAM,QAAQ,WAAW;EAAE;EAAW,MAAM;EAAc;EAAQ;EAAM;EAAM;EAAO;EAAM,CAAC;AAE5F,QAAO;;AAOT,eAAsB,wBACpB,SACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;CAEpF,MAAM,eAA2F,EAAE;CAEnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAGvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAG5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAItF;EACE,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GAGjC,MAAM,aAAuB;IAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IAG/C;IAAW,QAAQ;IAAS,MAAM;IACtC,MAAM;IAAe,WAAW;IAAG,SAAS;IAC5C,MAJW,WAAW,KAAK;IAIrB;IAAM;IAAW,WAAW;IACnC;AACD,OAAI;AACF,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;WAClC;;;AAMZ,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAIxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;AAC5B,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IAGjC,MAAM,aAAuB;KAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;KAG/C;KAAW,QAAQ;KAAS,MAAM;KACtC,MAAM;KAAe,WAAW;KAAG,SAAS;KAC5C,MAJW,WAAW,KAAK;KAIrB;KAAM;KAAW,WAAW;KACnC;AACD,QAAI;AACF,WAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;YAClC;;;AAMZ,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAAE,SAAS;IAAgB,UAAU;IAAkB,QAAQ;IAAU,MAAM;IAAa,CAAC;GAEjH,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;AACxD,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAKxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAC3C,MAAI;AAGF,OAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;IACX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,MAAM,QAAQ,sBAAsB,UAAU;CAEnE,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,KAAK,cAAc;EAC5B,MAAM,WAAW,EAAE,SAAS,UAAU,GAAG,EAAE,MAAM,GAAG,GAAkB,GAAG;AACzE,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,EAAE;;AAItB,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,YAAY,WAAW,WAAW;AAGlD,QAAO;;AAOT,MAAM,mBAAmB;;AAIzB,MAAM,8BAA8B;;AAEpC,MAAM,8BAA8B;;;;;;;;;;;;;;;AA8BpC,eAAsB,uBACpB,SACA,YACA,cACA,SACiB;CACjB,MAAM,EAAE,oBAAoB,uBAAuB,MAAM,OAAO;CAEhE,MAAM,YAAY,SAAS,aAAa;CACxC,MAAM,YAAY,SAAS,aAAa;CACxC,MAAM,WAAW,KAAK,KAAK,GAAG;CAE9B,MAAM,OAAO,MAAM,QAAQ,sBAAsB,QAAW,UAAU;AACtE,KAAI,KAAK,WAAW,EAAG,QAAO;CAE9B,MAAM,QAAQ,KAAK;CACnB,IAAI,WAAW;CAGf,MAAM,qCAAqB,IAAI,KAAoD;AACnF,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,QAAQ,mBAAmB,IAAI,IAAI,WAAW;AACpD,MAAI,MACF,OAAM;MAEN,oBAAmB,IAAI,IAAI,YAAY;GAAE,OAAO;GAAG,YAAY,IAAI;GAAM,CAAC;;CAG9E,MAAM,SAAS,QAAgB,cAAc,IAAI,IAAI,IAAI,WAAW;CACpE,MAAM,iBAAiB,MAAM,KAAK,mBAAmB,SAAS,CAAC,CAC5D,KAAK,CAAC,KAAK,EAAE,OAAO,kBAAkB,KAAK,MAAM,IAAI,CAAC,IAAI,MAAM,gBAAgB,WAAW,GAAG,CAC9F,KAAK,KAAK;AACb,SAAQ,OAAO,MACb,4BAA4B,MAAM,4BAA4B,mBAAmB,KAAK,eAAe,eAAe,IACrH;CAGD,IAAI,mBAAmB;CACvB,IAAI,kBAAkB;AAEtB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,kBAAkB;AAEtD,MAAI,cAAc,EAAE;AAClB,WAAQ,OAAO,MACb,2CAA2C,SAAS,GAAG,MAAM,gCAC9D;AACD;;AAKF,MAAI,KAAK,KAAK,IAAI,UAAU;AAC1B,WAAQ,OAAO,MACb,0CAA0C,SAAS,GAAG,MAAM,WAAW,UAAU,wCAClF;AACD;;EAGF,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,iBAAiB;AAIjD,QAAM,kBAAkB;EAExB,MAAM,OAAO,MAAM,mBAAmB,MAAM,KAAK,MAAM,EAAE,KAAK,CAAC;AAO/D,QAAM,QAAQ,IACZ,MAAM,KAAK,KAAK,MAAM,QAAQ,gBAAgB,IAAI,IAAI,mBAAmB,KAAK,GAAG,CAAC,CAAC,CACpF;AAMD,OAAK,MAAM,EAAE,YAAY,UAAU,OAAO;AACxC,OAAI,eAAe,kBAAkB;AACnC,QAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;IAEH,MAAM,OAAO,mBAAmB,IAAI,WAAW;AAC/C,YAAQ,OAAO,MACb,0BAA0B,MAAM,WAAW,CAAC,IAAI,MAAM,SAAS,IAAI,uBAAuB,KAAK,KAChG;AACD,uBAAmB;AACnB,sBAAkB;;AAEpB;;AAGF,cAAY,MAAM;EAGlB,MAAM,YAAY,MAAM,MAAM,SAAS;AACvC,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAAW,MAAM,UAAU,WAAW,CAAC,IAAI,UAAU,KAAK,KACtG;;AAIH,KAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;AAGH,QAAO;;AAOT,eAAsB,oBACpB,SACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;AAEpF,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,wBAAwB,SAAS,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzG,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ"}
+{"version":3,"file":"indexer-backend-TG64CCQC.mjs","names":[],"sources":["../src/memory/indexer/async.ts"],"sourcesContent":["/**\n * Backend-aware async indexer for PAI federation memory.\n *\n * Provides the same functionality as sync.ts but writes through the\n * StorageBackend interface instead of directly to better-sqlite3.\n * Used when the daemon is configured with the Postgres backend.\n *\n * The SQLite path still uses sync.ts directly (which is faster for SQLite\n * due to synchronous transactions).\n */\n\nimport { readFileSync, statSync, existsSync } from \"node:fs\";\nimport { join, relative, basename } from \"node:path\";\nimport type { Database } from \"better-sqlite3\";\nimport type { StorageBackend, ChunkRow } from \"../../storage/interface.js\";\nimport { chunkMarkdown } from \"../chunker.js\";\nimport {\n  sha256File,\n  chunkId,\n  detectTier,\n  walkMdFiles,\n  walkContentFiles,\n  isPathTooBroadForContentScan,\n  parseSessionTitleChunk,\n  yieldToEventLoop,\n  INDEX_YIELD_EVERY,\n} from \"./helpers.js\";\nimport type { IndexResult } from \"./types.js\";\n\nexport type { IndexResult };\n\n// ---------------------------------------------------------------------------\n// Single-file indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\n/**\n * Index a single file through the StorageBackend interface.\n * Returns true if the file was re-indexed (changed or new), false if skipped.\n */\nexport async function indexFileWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  relativePath: string,\n  source: string,\n  tier: string,\n): Promise<boolean> {\n  const absPath = join(rootPath, relativePath);\n\n  let content: string;\n  let stat: ReturnType<typeof statSync>;\n  try {\n    content = readFileSync(absPath, \"utf8\");\n    stat = statSync(absPath);\n  } catch {\n    return false;\n  }\n\n  const hash = sha256File(content);\n  const mtime = Math.floor(stat.mtimeMs);\n  const size = stat.size;\n\n  // Change detection\n  const existingHash = await backend.getFileHash(projectId, relativePath);\n  if (existingHash === hash) return false;\n\n  // Delete old chunks\n  await backend.deleteChunksForFile(projectId, relativePath);\n\n  // Chunk the content\n  const rawChunks = chunkMarkdown(content);\n  const updatedAt = Date.now();\n\n  const chunks: ChunkRow[] = rawChunks.map((c, i) => ({\n    id: chunkId(projectId, relativePath, i, c.startLine, c.endLine),\n    projectId,\n    source,\n    tier,\n    path: relativePath,\n    startLine: c.startLine,\n    endLine: c.endLine,\n    hash: c.hash,\n    text: c.text,\n    updatedAt,\n    embedding: null,\n  }));\n\n  // Insert chunks + update file record\n  await backend.insertChunks(chunks);\n  await backend.upsertFile({ projectId, path: relativePath, source, tier, hash, mtime, size });\n\n  return true;\n}\n\n// ---------------------------------------------------------------------------\n// Project-level indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexProjectWithBackend(\n  backend: StorageBackend,\n  projectId: number,\n  rootPath: string,\n  claudeNotesDir?: string | null,\n): Promise<IndexResult> {\n  const result: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  const filesToIndex: Array<{ absPath: string; rootBase: string; source: string; tier: string }> = [];\n\n  const rootMemoryMd = join(rootPath, \"MEMORY.md\");\n  if (existsSync(rootMemoryMd)) {\n    filesToIndex.push({ absPath: rootMemoryMd, rootBase: rootPath, source: \"memory\", tier: \"evergreen\" });\n  }\n\n  const memoryDir = join(rootPath, \"memory\");\n  for (const absPath of walkMdFiles(memoryDir)) {\n    const relPath = relative(rootPath, absPath);\n    const tier = detectTier(relPath);\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"memory\", tier });\n  }\n\n  const notesDir = join(rootPath, \"Notes\");\n  for (const absPath of walkMdFiles(notesDir)) {\n    filesToIndex.push({ absPath, rootBase: rootPath, source: \"notes\", tier: \"session\" });\n  }\n\n  // Synthetic session-title chunks for Notes files\n  {\n    const updatedAt = Date.now();\n    for (const absPath of walkMdFiles(notesDir)) {\n      const fileName = basename(absPath);\n      const text = parseSessionTitleChunk(fileName);\n      if (!text) continue;\n      const relPath = relative(rootPath, absPath);\n      const syntheticPath = `${relPath}::title`;\n      const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n      const hash = sha256File(text);\n      const titleChunk: ChunkRow = {\n        id, projectId, source: \"notes\", tier: \"session\",\n        path: syntheticPath, startLine: 0, endLine: 0,\n        hash, text, updatedAt, embedding: null,\n      };\n      try {\n        await backend.insertChunks([titleChunk]);\n      } catch {\n        // Skip title chunks that cause backend errors\n      }\n    }\n  }\n\n  if (!isPathTooBroadForContentScan(rootPath)) {\n    for (const absPath of walkContentFiles(rootPath)) {\n      filesToIndex.push({ absPath, rootBase: rootPath, source: \"content\", tier: \"topic\" });\n    }\n  }\n\n  if (claudeNotesDir && claudeNotesDir !== notesDir) {\n    for (const absPath of walkMdFiles(claudeNotesDir)) {\n      filesToIndex.push({ absPath, rootBase: claudeNotesDir, source: \"notes\", tier: \"session\" });\n    }\n\n    // Synthetic title chunks for claude notes dir\n    {\n      const updatedAt = Date.now();\n      for (const absPath of walkMdFiles(claudeNotesDir)) {\n        const fileName = basename(absPath);\n        const text = parseSessionTitleChunk(fileName);\n        if (!text) continue;\n        const relPath = relative(claudeNotesDir, absPath);\n        const syntheticPath = `${relPath}::title`;\n        const id = chunkId(projectId, syntheticPath, 0, 0, 0);\n        const hash = sha256File(text);\n        const titleChunk: ChunkRow = {\n          id, projectId, source: \"notes\", tier: \"session\",\n          path: syntheticPath, startLine: 0, endLine: 0,\n          hash, text, updatedAt, embedding: null,\n        };\n        try {\n          await backend.insertChunks([titleChunk]);\n        } catch {\n          // Skip title chunks that cause backend errors\n        }\n      }\n    }\n\n    if (claudeNotesDir.endsWith(\"/Notes\")) {\n      const claudeProjectDir = claudeNotesDir.slice(0, -\"/Notes\".length);\n      const claudeMemoryMd = join(claudeProjectDir, \"MEMORY.md\");\n      if (existsSync(claudeMemoryMd)) {\n        filesToIndex.push({ absPath: claudeMemoryMd, rootBase: claudeProjectDir, source: \"memory\", tier: \"evergreen\" });\n      }\n      const claudeMemoryDir = join(claudeProjectDir, \"memory\");\n      for (const absPath of walkMdFiles(claudeMemoryDir)) {\n        const relPath = relative(claudeProjectDir, absPath);\n        const tier = detectTier(relPath);\n        filesToIndex.push({ absPath, rootBase: claudeProjectDir, source: \"memory\", tier });\n      }\n    }\n  }\n\n  await yieldToEventLoop();\n\n  let filesSinceYield = 0;\n\n  for (const { absPath, rootBase, source, tier } of filesToIndex) {\n    if (filesSinceYield >= INDEX_YIELD_EVERY) {\n      await yieldToEventLoop();\n      filesSinceYield = 0;\n    }\n    filesSinceYield++;\n\n    const relPath = relative(rootBase, absPath);\n    try {\n      const changed = await indexFileWithBackend(backend, projectId, rootBase, relPath, source, tier);\n\n      if (changed) {\n        const ids = await backend.getChunkIds(projectId, relPath);\n        result.filesProcessed++;\n        result.chunksCreated += ids.length;\n      } else {\n        result.filesSkipped++;\n      }\n    } catch {\n      // Skip files that cause backend errors (e.g. null bytes in Postgres)\n      result.filesSkipped++;\n    }\n  }\n\n  // Prune stale paths\n  const livePaths = new Set<string>();\n  for (const { absPath, rootBase } of filesToIndex) {\n    livePaths.add(relative(rootBase, absPath));\n  }\n\n  const dbChunkPaths = await backend.getDistinctChunkPaths(projectId);\n\n  const stalePaths: string[] = [];\n  for (const p of dbChunkPaths) {\n    const basePath = p.endsWith(\"::title\") ? p.slice(0, -\"::title\".length) : p;\n    if (!livePaths.has(basePath)) {\n      stalePaths.push(p);\n    }\n  }\n\n  if (stalePaths.length > 0) {\n    await backend.deletePaths(projectId, stalePaths);\n  }\n\n  return result;\n}\n\n// ---------------------------------------------------------------------------\n// Embedding generation via StorageBackend\n// ---------------------------------------------------------------------------\n\nconst EMBED_BATCH_SIZE = 50;\nconst EMBED_YIELD_EVERY = 1;\n\n/** Default ceiling on one pass. See EmbedPassOptions. */\nconst DEFAULT_MAX_CHUNKS_PER_PASS = 5_000;\n/** Default wall-clock ceiling on one pass, in milliseconds. */\nconst DEFAULT_MAX_MILLIS_PER_PASS = 120_000;\n\nexport interface EmbedPassOptions {\n  /**\n   * Stop the pass after this many chunks. Unbounded passes are the reason a\n   * six-figure backlog starves the indexer: the daemon serialises indexing and\n   * embedding against each other, so a pass that runs for hours means nothing\n   * new gets indexed for hours. Bounding the pass costs nothing — every\n   * embedding is written as it is produced, so the next pass resumes exactly\n   * where this one stopped.\n   */\n  maxChunks?: number;\n  /** Stop the pass after this much wall-clock time, whichever bound hits first. */\n  maxMillis?: number;\n}\n\n/**\n * Generate and store embeddings for unembedded chunks via the StorageBackend.\n *\n * The pass is deliberately bounded (see EmbedPassOptions) and resumable: it\n * takes a slice of the backlog, embeds it, and returns so the scheduler can run\n * an index pass before coming back. Draining the whole backlog in one call\n * starves indexing for as long as the call runs.\n *\n * The optional `shouldStop` callback is checked between every batch. When it\n * returns true the embed loop exits early so the caller (e.g. the daemon\n * shutdown handler) can close the pool without racing against active queries.\n *\n * Returns the number of newly embedded chunks.\n */\nexport async function embedChunksWithBackend(\n  backend: StorageBackend,\n  shouldStop?: () => boolean,\n  projectNames?: Map<number, string>,\n  options?: EmbedPassOptions,\n): Promise<number> {\n  const { generateEmbeddings, serializeEmbedding } = await import(\"../embeddings.js\");\n\n  const maxChunks = options?.maxChunks ?? DEFAULT_MAX_CHUNKS_PER_PASS;\n  const maxMillis = options?.maxMillis ?? DEFAULT_MAX_MILLIS_PER_PASS;\n  const deadline = Date.now() + maxMillis;\n\n  const rows = await backend.getUnembeddedChunkIds(undefined, maxChunks);\n  if (rows.length === 0) return 0;\n\n  const total = rows.length;\n  let embedded = 0;\n\n  // Build a summary of what needs embedding: count chunks per project_id\n  const projectChunkCounts = new Map<number, { count: number; samplePath: string }>();\n  for (const row of rows) {\n    const entry = projectChunkCounts.get(row.project_id);\n    if (entry) {\n      entry.count++;\n    } else {\n      projectChunkCounts.set(row.project_id, { count: 1, samplePath: row.path });\n    }\n  }\n  const pName = (pid: number) => projectNames?.get(pid) ?? `project ${pid}`;\n  const projectSummary = Array.from(projectChunkCounts.entries())\n    .map(([pid, { count, samplePath }]) => `  ${pName(pid)}: ${count} chunks (e.g. ${samplePath})`)\n    .join(\"\\n\");\n  process.stderr.write(\n    `[pai-daemon] Embed pass: ${total} unembedded chunks across ${projectChunkCounts.size} project(s)\\n${projectSummary}\\n`\n  );\n\n  // Track current project for transition logging\n  let currentProjectId = -1;\n  let projectEmbedded = 0;\n\n  for (let i = 0; i < rows.length; i += EMBED_BATCH_SIZE) {\n    // Check cancellation between every batch before touching the pool again\n    if (shouldStop?.()) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass cancelled after ${embedded}/${total} chunks (shutdown requested)\\n`\n      );\n      break;\n    }\n\n    // Yield the daemon back to the indexer rather than run past the deadline.\n    // The remaining rows are simply picked up by the next scheduled pass.\n    if (Date.now() >= deadline) {\n      process.stderr.write(\n        `[pai-daemon] Embed pass yielding after ${embedded}/${total} chunks (${maxMillis}ms budget spent); resuming next pass\\n`\n      );\n      break;\n    }\n\n    const batch = rows.slice(i, i + EMBED_BATCH_SIZE);\n\n    // Keep IPC responsive: the forward pass below is synchronous inside the\n    // model, so yield before entering it rather than between chunks.\n    await yieldToEventLoop();\n\n    const vecs = await generateEmbeddings(batch.map((r) => r.text));\n    // Issue the writes together rather than one round-trip at a time. Measured\n    // on this machine the model embeds 40-67 chunks/s in isolation while the\n    // daemon managed ~5/s: the difference was one sequential UPDATE per chunk,\n    // so the pass spent most of its budget waiting on the network, not working.\n    // Concurrency is bounded by the batch size, which the pool handles; the\n    // SQLite backend is synchronous and simply ignores the difference.\n    await Promise.all(\n      batch.map((row, j) => backend.updateEmbedding(row.id, serializeEmbedding(vecs[j])))\n    );\n\n    // Attribute the batch to projects only after it is durably stored, and walk\n    // it in order so a batch straddling a project boundary credits each side\n    // correctly. Rows arrive ordered by project, so a boundary is a real\n    // transition, not the per-chunk flapping this logging used to produce.\n    for (const { project_id, path } of batch) {\n      if (project_id !== currentProjectId) {\n        if (currentProjectId !== -1) {\n          process.stderr.write(\n            `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n          );\n        }\n        const info = projectChunkCounts.get(project_id);\n        process.stderr.write(\n          `[pai-daemon] Embedding ${pName(project_id)} (${info?.count ?? \"?\"} chunks, starting at ${path})\\n`\n        );\n        currentProjectId = project_id;\n        projectEmbedded = 0;\n      }\n      projectEmbedded++;\n    }\n\n    embedded += batch.length;\n\n    // Log progress with current file path for context\n    const lastChunk = batch[batch.length - 1];\n    process.stderr.write(\n      `[pai-daemon] Embedded ${embedded}/${total} chunks (${pName(lastChunk.project_id)}: ${lastChunk.path})\\n`\n    );\n  }\n\n  // Log final project completion\n  if (currentProjectId !== -1) {\n    process.stderr.write(\n      `[pai-daemon] Finished ${pName(currentProjectId)}: ${projectEmbedded} chunks embedded\\n`\n    );\n  }\n\n  return embedded;\n}\n\n// ---------------------------------------------------------------------------\n// Global indexing via StorageBackend\n// ---------------------------------------------------------------------------\n\nexport async function indexAllWithBackend(\n  backend: StorageBackend,\n  registryDb: Database,\n): Promise<{ projects: number; result: IndexResult }> {\n  const projects = registryDb\n    .prepare(\"SELECT id, root_path, claude_notes_dir FROM projects WHERE status = 'active'\")\n    .all() as Array<{ id: number; root_path: string; claude_notes_dir: string | null }>;\n\n  const totals: IndexResult = { filesProcessed: 0, chunksCreated: 0, filesSkipped: 0 };\n\n  for (const project of projects) {\n    await yieldToEventLoop();\n    const r = await indexProjectWithBackend(backend, project.id, project.root_path, project.claude_notes_dir);\n    totals.filesProcessed += r.filesProcessed;\n    totals.chunksCreated += r.chunksCreated;\n    totals.filesSkipped += r.filesSkipped;\n  }\n\n  return { projects: projects.length, result: totals };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;AAuCA,eAAsB,qBACpB,SACA,WACA,UACA,cACA,QACA,MACkB;CAClB,MAAM,UAAU,KAAK,UAAU,aAAa;CAE5C,IAAI;CACJ,IAAI;AACJ,KAAI;AACF,YAAU,aAAa,SAAS,OAAO;AACvC,SAAO,SAAS,QAAQ;SAClB;AACN,SAAO;;CAGT,MAAM,OAAO,WAAW,QAAQ;CAChC,MAAM,QAAQ,KAAK,MAAM,KAAK,QAAQ;CACtC,MAAM,OAAO,KAAK;AAIlB,KADqB,MAAM,QAAQ,YAAY,WAAW,aAAa,KAClD,KAAM,QAAO;AAGlC,OAAM,QAAQ,oBAAoB,WAAW,aAAa;CAG1D,MAAM,YAAY,cAAc,QAAQ;CACxC,MAAM,YAAY,KAAK,KAAK;CAE5B,MAAM,SAAqB,UAAU,KAAK,GAAG,OAAO;EAClD,IAAI,QAAQ,WAAW,cAAc,GAAG,EAAE,WAAW,EAAE,QAAQ;EAC/D;EACA;EACA;EACA,MAAM;EACN,WAAW,EAAE;EACb,SAAS,EAAE;EACX,MAAM,EAAE;EACR,MAAM,EAAE;EACR;EACA,WAAW;EACZ,EAAE;AAGH,OAAM,QAAQ,aAAa,OAAO;AAClC,OAAM,QAAQ,WAAW;EAAE;EAAW,MAAM;EAAc;EAAQ;EAAM;EAAM;EAAO;EAAM,CAAC;AAE5F,QAAO;;AAOT,eAAsB,wBACpB,SACA,WACA,UACA,gBACsB;CACtB,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;CAEpF,MAAM,eAA2F,EAAE;CAEnG,MAAM,eAAe,KAAK,UAAU,YAAY;AAChD,KAAI,WAAW,aAAa,CAC1B,cAAa,KAAK;EAAE,SAAS;EAAc,UAAU;EAAU,QAAQ;EAAU,MAAM;EAAa,CAAC;CAGvG,MAAM,YAAY,KAAK,UAAU,SAAS;AAC1C,MAAK,MAAM,WAAW,YAAY,UAAU,EAAE;EAE5C,MAAM,OAAO,WADG,SAAS,UAAU,QAAQ,CACX;AAChC,eAAa,KAAK;GAAE;GAAS,UAAU;GAAU,QAAQ;GAAU;GAAM,CAAC;;CAG5E,MAAM,WAAW,KAAK,UAAU,QAAQ;AACxC,MAAK,MAAM,WAAW,YAAY,SAAS,CACzC,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAS,MAAM;EAAW,CAAC;CAItF;EACE,MAAM,YAAY,KAAK,KAAK;AAC5B,OAAK,MAAM,WAAW,YAAY,SAAS,EAAE;GAE3C,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,OAAI,CAAC,KAAM;GAEX,MAAM,gBAAgB,GADN,SAAS,UAAU,QAAQ,CACV;GAGjC,MAAM,aAAuB;IAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;IAG/C;IAAW,QAAQ;IAAS,MAAM;IACtC,MAAM;IAAe,WAAW;IAAG,SAAS;IAC5C,MAJW,WAAW,KAAK;IAIrB;IAAM;IAAW,WAAW;IACnC;AACD,OAAI;AACF,UAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;WAClC;;;AAMZ,KAAI,CAAC,6BAA6B,SAAS,CACzC,MAAK,MAAM,WAAW,iBAAiB,SAAS,CAC9C,cAAa,KAAK;EAAE;EAAS,UAAU;EAAU,QAAQ;EAAW,MAAM;EAAS,CAAC;AAIxF,KAAI,kBAAkB,mBAAmB,UAAU;AACjD,OAAK,MAAM,WAAW,YAAY,eAAe,CAC/C,cAAa,KAAK;GAAE;GAAS,UAAU;GAAgB,QAAQ;GAAS,MAAM;GAAW,CAAC;EAI5F;GACE,MAAM,YAAY,KAAK,KAAK;AAC5B,QAAK,MAAM,WAAW,YAAY,eAAe,EAAE;IAEjD,MAAM,OAAO,uBADI,SAAS,QAAQ,CACW;AAC7C,QAAI,CAAC,KAAM;IAEX,MAAM,gBAAgB,GADN,SAAS,gBAAgB,QAAQ,CAChB;IAGjC,MAAM,aAAuB;KAC3B,IAHS,QAAQ,WAAW,eAAe,GAAG,GAAG,EAAE;KAG/C;KAAW,QAAQ;KAAS,MAAM;KACtC,MAAM;KAAe,WAAW;KAAG,SAAS;KAC5C,MAJW,WAAW,KAAK;KAIrB;KAAM;KAAW,WAAW;KACnC;AACD,QAAI;AACF,WAAM,QAAQ,aAAa,CAAC,WAAW,CAAC;YAClC;;;AAMZ,MAAI,eAAe,SAAS,SAAS,EAAE;GACrC,MAAM,mBAAmB,eAAe,MAAM,GAAG,GAAiB;GAClE,MAAM,iBAAiB,KAAK,kBAAkB,YAAY;AAC1D,OAAI,WAAW,eAAe,CAC5B,cAAa,KAAK;IAAE,SAAS;IAAgB,UAAU;IAAkB,QAAQ;IAAU,MAAM;IAAa,CAAC;GAEjH,MAAM,kBAAkB,KAAK,kBAAkB,SAAS;AACxD,QAAK,MAAM,WAAW,YAAY,gBAAgB,EAAE;IAElD,MAAM,OAAO,WADG,SAAS,kBAAkB,QAAQ,CACnB;AAChC,iBAAa,KAAK;KAAE;KAAS,UAAU;KAAkB,QAAQ;KAAU;KAAM,CAAC;;;;AAKxF,OAAM,kBAAkB;CAExB,IAAI,kBAAkB;AAEtB,MAAK,MAAM,EAAE,SAAS,UAAU,QAAQ,UAAU,cAAc;AAC9D,MAAI,mBAAmB,mBAAmB;AACxC,SAAM,kBAAkB;AACxB,qBAAkB;;AAEpB;EAEA,MAAM,UAAU,SAAS,UAAU,QAAQ;AAC3C,MAAI;AAGF,OAFgB,MAAM,qBAAqB,SAAS,WAAW,UAAU,SAAS,QAAQ,KAAK,EAElF;IACX,MAAM,MAAM,MAAM,QAAQ,YAAY,WAAW,QAAQ;AACzD,WAAO;AACP,WAAO,iBAAiB,IAAI;SAE5B,QAAO;UAEH;AAEN,UAAO;;;CAKX,MAAM,4BAAY,IAAI,KAAa;AACnC,MAAK,MAAM,EAAE,SAAS,cAAc,aAClC,WAAU,IAAI,SAAS,UAAU,QAAQ,CAAC;CAG5C,MAAM,eAAe,MAAM,QAAQ,sBAAsB,UAAU;CAEnE,MAAM,aAAuB,EAAE;AAC/B,MAAK,MAAM,KAAK,cAAc;EAC5B,MAAM,WAAW,EAAE,SAAS,UAAU,GAAG,EAAE,MAAM,GAAG,GAAkB,GAAG;AACzE,MAAI,CAAC,UAAU,IAAI,SAAS,CAC1B,YAAW,KAAK,EAAE;;AAItB,KAAI,WAAW,SAAS,EACtB,OAAM,QAAQ,YAAY,WAAW,WAAW;AAGlD,QAAO;;AAOT,MAAM,mBAAmB;;AAIzB,MAAM,8BAA8B;;AAEpC,MAAM,8BAA8B;;;;;;;;;;;;;;;AA8BpC,eAAsB,uBACpB,SACA,YACA,cACA,SACiB;CACjB,MAAM,EAAE,oBAAoB,uBAAuB,MAAM,OAAO;CAEhE,MAAM,YAAY,SAAS,aAAa;CACxC,MAAM,YAAY,SAAS,aAAa;CACxC,MAAM,WAAW,KAAK,KAAK,GAAG;CAE9B,MAAM,OAAO,MAAM,QAAQ,sBAAsB,QAAW,UAAU;AACtE,KAAI,KAAK,WAAW,EAAG,QAAO;CAE9B,MAAM,QAAQ,KAAK;CACnB,IAAI,WAAW;CAGf,MAAM,qCAAqB,IAAI,KAAoD;AACnF,MAAK,MAAM,OAAO,MAAM;EACtB,MAAM,QAAQ,mBAAmB,IAAI,IAAI,WAAW;AACpD,MAAI,MACF,OAAM;MAEN,oBAAmB,IAAI,IAAI,YAAY;GAAE,OAAO;GAAG,YAAY,IAAI;GAAM,CAAC;;CAG9E,MAAM,SAAS,QAAgB,cAAc,IAAI,IAAI,IAAI,WAAW;CACpE,MAAM,iBAAiB,MAAM,KAAK,mBAAmB,SAAS,CAAC,CAC5D,KAAK,CAAC,KAAK,EAAE,OAAO,kBAAkB,KAAK,MAAM,IAAI,CAAC,IAAI,MAAM,gBAAgB,WAAW,GAAG,CAC9F,KAAK,KAAK;AACb,SAAQ,OAAO,MACb,4BAA4B,MAAM,4BAA4B,mBAAmB,KAAK,eAAe,eAAe,IACrH;CAGD,IAAI,mBAAmB;CACvB,IAAI,kBAAkB;AAEtB,MAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,kBAAkB;AAEtD,MAAI,cAAc,EAAE;AAClB,WAAQ,OAAO,MACb,2CAA2C,SAAS,GAAG,MAAM,gCAC9D;AACD;;AAKF,MAAI,KAAK,KAAK,IAAI,UAAU;AAC1B,WAAQ,OAAO,MACb,0CAA0C,SAAS,GAAG,MAAM,WAAW,UAAU,wCAClF;AACD;;EAGF,MAAM,QAAQ,KAAK,MAAM,GAAG,IAAI,iBAAiB;AAIjD,QAAM,kBAAkB;EAExB,MAAM,OAAO,MAAM,mBAAmB,MAAM,KAAK,MAAM,EAAE,KAAK,CAAC;AAO/D,QAAM,QAAQ,IACZ,MAAM,KAAK,KAAK,MAAM,QAAQ,gBAAgB,IAAI,IAAI,mBAAmB,KAAK,GAAG,CAAC,CAAC,CACpF;AAMD,OAAK,MAAM,EAAE,YAAY,UAAU,OAAO;AACxC,OAAI,eAAe,kBAAkB;AACnC,QAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;IAEH,MAAM,OAAO,mBAAmB,IAAI,WAAW;AAC/C,YAAQ,OAAO,MACb,0BAA0B,MAAM,WAAW,CAAC,IAAI,MAAM,SAAS,IAAI,uBAAuB,KAAK,KAChG;AACD,uBAAmB;AACnB,sBAAkB;;AAEpB;;AAGF,cAAY,MAAM;EAGlB,MAAM,YAAY,MAAM,MAAM,SAAS;AACvC,UAAQ,OAAO,MACb,yBAAyB,SAAS,GAAG,MAAM,WAAW,MAAM,UAAU,WAAW,CAAC,IAAI,UAAU,KAAK,KACtG;;AAIH,KAAI,qBAAqB,GACvB,SAAQ,OAAO,MACb,yBAAyB,MAAM,iBAAiB,CAAC,IAAI,gBAAgB,oBACtE;AAGH,QAAO;;AAOT,eAAsB,oBACpB,SACA,YACoD;CACpD,MAAM,WAAW,WACd,QAAQ,+EAA+E,CACvF,KAAK;CAER,MAAM,SAAsB;EAAE,gBAAgB;EAAG,eAAe;EAAG,cAAc;EAAG;AAEpF,MAAK,MAAM,WAAW,UAAU;AAC9B,QAAM,kBAAkB;EACxB,MAAM,IAAI,MAAM,wBAAwB,SAAS,QAAQ,IAAI,QAAQ,WAAW,QAAQ,iBAAiB;AACzG,SAAO,kBAAkB,EAAE;AAC3B,SAAO,iBAAiB,EAAE;AAC1B,SAAO,gBAAgB,EAAE;;AAG3B,QAAO;EAAE,UAAU,SAAS;EAAQ,QAAQ;EAAQ"}

package/dist/{latent-ideas-BC1oINZ-.mjs → latent-ideas-B7wq75Pt.mjs}

@@ -1,6 +1,6 @@
 import "./embeddings-DOLZnT1X.mjs";
 import { n as TITLE_STOP_WORDS } from "./stop-words-Hfu8u22w.mjs";
-import { t as zettelThemes } from "./themes-BN0a2duq.mjs";
+import { t as zettelThemes } from "./themes-BI4GMRP9.mjs";
 import { mkdirSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 
@@ -188,4 +188,4 @@ function handleIdeaMaterialize(params, vaultPath) {
 
 //#endregion
 export { handleGraphLatentIdeas, handleIdeaMaterialize };
-//# sourceMappingURL=latent-ideas-BC1oINZ-.mjs.map
+//# sourceMappingURL=latent-ideas-B7wq75Pt.mjs.map

package/dist/{latent-ideas-BC1oINZ-.mjs.map → latent-ideas-B7wq75Pt.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"latent-ideas-BC1oINZ-.mjs","names":[],"sources":["../src/graph/latent-ideas.ts"],"sourcesContent":["/**\n * latent-ideas.ts — graph_latent_ideas and idea_materialize endpoint handlers\n *\n * \"Latent ideas\" are recurring themes in the vault that exist as embedding\n * clusters but have NO dedicated note written about them yet.  PAI surfaces\n * these by running the same agglomerative clustering used by graph_clusters /\n * zettelThemes and then filtering OUT any cluster whose label is well-matched\n * by an existing note title.\n *\n * The materialize endpoint writes a new Markdown note to the vault filesystem\n * and returns its content so the plugin can open it immediately.\n */\n\nimport { mkdirSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { TITLE_STOP_WORDS } from \"../utils/stop-words.js\";\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport { zettelThemes } from \"../zettelkasten/themes.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphLatentIdeasParams {\n  project_id: number;\n  /** Minimum notes in a cluster (default: 3) */\n  min_cluster_size?: number;\n  /** Cap on returned ideas (default: 15) */\n  max_ideas?: number;\n  /** How far back to look in days (default: 180) */\n  lookback_days?: number;\n  /** Cosine similarity clustering threshold (default: 0.65) */\n  similarity_threshold?: number;\n}\n\nexport interface LatentIdeaSourceNote {\n  vault_path: string;\n  title: string;\n  /** How strongly this note relates to the theme (0-1) */\n  relevance: number;\n}\n\nexport interface LatentIdea {\n  id: number;\n  /** Auto-generated cluster label from zettelThemes */\n  label: string;\n  /** Number of notes touching this theme */\n  size: number;\n  /** 0-1, how likely this is a real coherent idea */\n  confidence: number;\n  /** Notes that contribute to this theme */\n  source_notes: LatentIdeaSourceNote[];\n  /** Cleaned-up version of label for a potential note title */\n  suggested_title: string;\n  /** Most common folder among source notes */\n  suggested_folder: string;\n  /** Number of distinct session date-folders (e.g. \"2026/03\") touching this theme */\n  sessions_count: number;\n}\n\nexport interface GraphLatentIdeasResult {\n  ideas: LatentIdea[];\n  total_clusters_analyzed: number;\n  /** How many clusters already have a matching note (excluded from results) */\n  materialized_count: number;\n}\n\n// ---------------------------------------------------------------------------\n// Materialize params / result\n// ---------------------------------------------------------------------------\n\nexport interface IdeaMaterializeParams {\n  idea_label: string;\n  /** User-chosen title for the new note */\n  title: string;\n  /** Vault-relative folder path where the note should be created */\n  folder: string;\n  /** Vault-relative paths of the source notes to link from the new note */\n  source_paths: string[];\n  project_id: number;\n}\n\nexport interface IdeaMaterializeResult {\n  /** Vault-relative path of the created note */\n  vault_path: string;\n  /** Generated markdown content */\n  content: string;\n  /** Number of wikilinks inserted */\n  links_created: number;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: check if a cluster already has a matching note\n// ---------------------------------------------------------------------------\n\n/**\n * Returns true when any existing vault note title closely matches the cluster\n * label — meaning a dedicated note already exists for this topic.\n *\n * Matching strategy (simple, fast, no embeddings needed):\n *   1. Lowercase both sides and split into words.\n *   2. Remove stop words from the label words.\n *   3. If ≥ 60% of the significant label words appear in a note title → match.\n */\n// TITLE_STOP_WORDS imported from utils/stop-words.ts\n\nfunction labelMatchesTitle(label: string, title: string): boolean {\n  const labelWords = label\n    .toLowerCase()\n    .split(/[\\s\\-_/]+/)\n    .filter((w) => w.length > 2 && !TITLE_STOP_WORDS.has(w));\n\n  if (labelWords.length === 0) return false;\n\n  const titleLower = title.toLowerCase();\n  const matchCount = labelWords.filter((w) => titleLower.includes(w)).length;\n  return matchCount / labelWords.length >= 0.6;\n}\n\n/**\n * Check whether any note indexed in the vault has a title matching the label.\n * Fetches all vault file rows via StorageBackend for efficiency.\n */\nasync function clusterHasMatchingNote(\n  backend: StorageBackend,\n  label: string,\n  notePaths: string[]\n): Promise<boolean> {\n  // First check the notes already in the cluster themselves — if any cluster\n  // member's title matches the label it IS the index note → materialized.\n  const pathSet = new Set(notePaths);\n\n  // Fetch all vault files (bounded — vault rarely > 50k notes)\n  const rows = await backend.getAllVaultFiles();\n\n  for (const row of rows) {\n    if (!row.title) continue;\n    // Skip notes already counted inside the cluster — they don't count as\n    // \"dedicated notes\"; we only skip a cluster if a SEPARATE note exists.\n    if (pathSet.has(row.vaultPath)) continue;\n    if (labelMatchesTitle(label, row.title)) return true;\n  }\n  return false;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: generate a clean suggested title\n// ---------------------------------------------------------------------------\n\nfunction toSuggestedTitle(label: string): string {\n  // Remove leading/trailing whitespace, capitalize each word, remove stop words\n  // that are all-lowercase at the start of the title.\n  const words = label\n    .trim()\n    .split(/\\s+/)\n    .map((w, i) => {\n      const lower = w.toLowerCase();\n      // Drop leading stop words (but keep if they're the only word)\n      if (i === 0 && TITLE_STOP_WORDS.has(lower) && label.trim().split(/\\s+/).length > 1) {\n        return \"\";\n      }\n      return w.charAt(0).toUpperCase() + w.slice(1);\n    })\n    .filter(Boolean);\n\n  return words.join(\" \") || label;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: find most common folder\n// ---------------------------------------------------------------------------\n\nfunction mostCommonFolder(vaultPaths: string[]): string {\n  const counts = new Map<string, number>();\n  for (const p of vaultPaths) {\n    const parts = p.split(\"/\");\n    const folder = parts.length > 1 ? parts.slice(0, -1).join(\"/\") : \"\";\n    counts.set(folder, (counts.get(folder) ?? 0) + 1);\n  }\n\n  let best = \"\";\n  let bestCount = 0;\n  for (const [folder, count] of counts) {\n    if (count > bestCount) {\n      bestCount = count;\n      best = folder;\n    }\n  }\n  return best;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: count distinct session date-folders\n// ---------------------------------------------------------------------------\n\n/**\n * Heuristic: vault notes are often stored in date-based folders like\n * \"2026/03/15\" or \"Daily/2026-03\".  We extract the first numeric path\n * segment that looks like a year (2020-2030) and group by year+month.\n *\n * Falls back to counting distinct top-level folders.\n */\nfunction countDistinctSessions(vaultPaths: string[]): number {\n  const sessions = new Set<string>();\n  const yearMonthRe = /\\b(202\\d)\\D?(0[1-9]|1[0-2])\\b/;\n\n  for (const p of vaultPaths) {\n    const m = yearMonthRe.exec(p);\n    if (m) {\n      sessions.add(`${m[1]}-${m[2]}`);\n    } else {\n      // Fallback: use top-level folder as a proxy for \"session bucket\"\n      const topFolder = p.split(\"/\")[0];\n      sessions.add(topFolder);\n    }\n  }\n  return sessions.size;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: calculate confidence score\n// ---------------------------------------------------------------------------\n\n/**\n * Confidence combines:\n *  - Cluster size (normalized, capped at 20 for max contribution)\n *  - Folder diversity (0-1 already)\n *  - Sessions count (normalized, capped at 5)\n *\n * Formula: 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore\n */\nfunction calcConfidence(\n  size: number,\n  folderDiversity: number,\n  sessionsCount: number\n): number {\n  const sizeScore = Math.min(size / 20, 1.0);\n  const sessionScore = Math.min(sessionsCount / 5, 1.0);\n  const raw = 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore;\n  return Math.round(raw * 100) / 100;\n}\n\n// ---------------------------------------------------------------------------\n// Main handler: graph_latent_ideas\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphLatentIdeas(\n  backend: StorageBackend,\n  params: GraphLatentIdeasParams\n): Promise<GraphLatentIdeasResult> {\n  const minClusterSize = params.min_cluster_size ?? 3;\n  const maxIdeas = params.max_ideas ?? 15;\n  const lookbackDays = params.lookback_days ?? 180;\n  const similarityThreshold = params.similarity_threshold ?? 0.65;\n\n  const { project_id: vaultProjectId } = params;\n  if (!vaultProjectId) {\n    throw new Error(\n      \"graph_latent_ideas: project_id is required (pass the vault project's numeric ID)\"\n    );\n  }\n\n  // Run the same clustering algorithm used by graph_clusters\n  const themeResult = await zettelThemes(backend, {\n    vaultProjectId,\n    lookbackDays,\n    minClusterSize,\n    maxThemes: maxIdeas * 3, // Over-fetch — many will be filtered as materialized\n    similarityThreshold,\n  });\n\n  const ideas: LatentIdea[] = [];\n  let materializedCount = 0;\n\n  for (const theme of themeResult.themes) {\n    const notePaths = theme.notes.map((n) => n.path);\n\n    // Check if a dedicated note already exists for this theme\n    if (await clusterHasMatchingNote(backend, theme.label, notePaths)) {\n      materializedCount++;\n      continue;\n    }\n\n    // This is a latent idea — no dedicated note exists yet\n    const suggestedFolder = mostCommonFolder(notePaths);\n    const sessionsCount = countDistinctSessions(notePaths);\n    const confidence = calcConfidence(theme.size, theme.folderDiversity, sessionsCount);\n\n    // Build source notes with relevance scores\n    // Relevance is approximated by position in cluster (centroid-closest first)\n    // zettelThemes returns notes in no guaranteed order; assign uniform relevance\n    // decreasing from 1.0 to 0.5 across the list.\n    const sourceNotes: LatentIdeaSourceNote[] = theme.notes.map((n, idx) => ({\n      vault_path: n.path,\n      title: n.title ?? n.path.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? n.path,\n      relevance: Math.round((1.0 - (idx / Math.max(theme.notes.length - 1, 1)) * 0.5) * 100) / 100,\n    }));\n\n    ideas.push({\n      id: theme.id,\n      label: theme.label,\n      size: theme.size,\n      confidence,\n      source_notes: sourceNotes,\n      suggested_title: toSuggestedTitle(theme.label),\n      suggested_folder: suggestedFolder,\n      sessions_count: sessionsCount,\n    });\n\n    if (ideas.length >= maxIdeas) break;\n  }\n\n  // Sort by confidence descending\n  ideas.sort((a, b) => b.confidence - a.confidence);\n\n  return {\n    ideas,\n    total_clusters_analyzed: themeResult.themes.length + materializedCount,\n    materialized_count: materializedCount,\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Materialize handler: idea_materialize\n// ---------------------------------------------------------------------------\n\nexport function handleIdeaMaterialize(\n  params: IdeaMaterializeParams,\n  vaultPath: string\n): IdeaMaterializeResult {\n  const { idea_label, title, folder, source_paths } = params;\n\n  // Sanitize filename: replace characters illegal in filenames\n  const safeTitle = title.replace(/[/\\\\:*?\"<>|]/g, \"-\");\n  const fileName = `${safeTitle}.md`;\n\n  // Vault-relative path (forward slashes, no leading slash)\n  const relFolder = folder.replace(/^\\/+|\\/+$/g, \"\");\n  const vault_path = relFolder ? `${relFolder}/${fileName}` : fileName;\n\n  // Absolute filesystem path\n  const absPath = join(vaultPath, vault_path);\n  const absDir = dirname(absPath);\n\n  // Build wikilinks from source_paths\n  const wikilinks = source_paths\n    .map((p) => {\n      // Derive a display name: filename without extension\n      const name = p.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? p;\n      // Relative wikilink — use just the filename (Obsidian resolves by title)\n      return `- [[${name}]]`;\n    })\n    .join(\"\\n\");\n\n  const links_created = source_paths.length;\n\n  const content = [\n    `# ${title}`,\n    \"\",\n    `*Materialized from latent idea: \"${idea_label}\"*`,\n    `*Sources: ${links_created} notes*`,\n    \"\",\n    \"## Related Notes\",\n    \"\",\n    wikilinks || \"*(no source notes)*\",\n    \"\",\n    \"## Notes\",\n    \"\",\n    \"<!-- Add your thoughts about this idea here -->\",\n    \"\",\n  ].join(\"\\n\");\n\n  // Write the file (create parent directories as needed)\n  mkdirSync(absDir, { recursive: true });\n  writeFileSync(absPath, content, \"utf-8\");\n\n  return {\n    vault_path,\n    content,\n    links_created,\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA0GA,SAAS,kBAAkB,OAAe,OAAwB;CAChE,MAAM,aAAa,MAChB,aAAa,CACb,MAAM,YAAY,CAClB,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,iBAAiB,IAAI,EAAE,CAAC;AAE1D,KAAI,WAAW,WAAW,EAAG,QAAO;CAEpC,MAAM,aAAa,MAAM,aAAa;AAEtC,QADmB,WAAW,QAAQ,MAAM,WAAW,SAAS,EAAE,CAAC,CAAC,SAChD,WAAW,UAAU;;;;;;AAO3C,eAAe,uBACb,SACA,OACA,WACkB;CAGlB,MAAM,UAAU,IAAI,IAAI,UAAU;CAGlC,MAAM,OAAO,MAAM,QAAQ,kBAAkB;AAE7C,MAAK,MAAM,OAAO,MAAM;AACtB,MAAI,CAAC,IAAI,MAAO;AAGhB,MAAI,QAAQ,IAAI,IAAI,UAAU,CAAE;AAChC,MAAI,kBAAkB,OAAO,IAAI,MAAM,CAAE,QAAO;;AAElD,QAAO;;AAOT,SAAS,iBAAiB,OAAuB;AAgB/C,QAbc,MACX,MAAM,CACN,MAAM,MAAM,CACZ,KAAK,GAAG,MAAM;EACb,MAAM,QAAQ,EAAE,aAAa;AAE7B,MAAI,MAAM,KAAK,iBAAiB,IAAI,MAAM,IAAI,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,SAAS,EAC/E,QAAO;AAET,SAAO,EAAE,OAAO,EAAE,CAAC,aAAa,GAAG,EAAE,MAAM,EAAE;GAC7C,CACD,OAAO,QAAQ,CAEL,KAAK,IAAI,IAAI;;AAO5B,SAAS,iBAAiB,YAA8B;CACtD,MAAM,yBAAS,IAAI,KAAqB;AACxC,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,QAAQ,EAAE,MAAM,IAAI;EAC1B,MAAM,SAAS,MAAM,SAAS,IAAI,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,IAAI,GAAG;AACjE,SAAO,IAAI,SAAS,OAAO,IAAI,OAAO,IAAI,KAAK,EAAE;;CAGnD,IAAI,OAAO;CACX,IAAI,YAAY;AAChB,MAAK,MAAM,CAAC,QAAQ,UAAU,OAC5B,KAAI,QAAQ,WAAW;AACrB,cAAY;AACZ,SAAO;;AAGX,QAAO;;;;;;;;;AAcT,SAAS,sBAAsB,YAA8B;CAC3D,MAAM,2BAAW,IAAI,KAAa;CAClC,MAAM,cAAc;AAEpB,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,IAAI,YAAY,KAAK,EAAE;AAC7B,MAAI,EACF,UAAS,IAAI,GAAG,EAAE,GAAG,GAAG,EAAE,KAAK;OAC1B;GAEL,MAAM,YAAY,EAAE,MAAM,IAAI,CAAC;AAC/B,YAAS,IAAI,UAAU;;;AAG3B,QAAO,SAAS;;;;;;;;;;AAelB,SAAS,eACP,MACA,iBACA,eACQ;CACR,MAAM,YAAY,KAAK,IAAI,OAAO,IAAI,EAAI;CAC1C,MAAM,eAAe,KAAK,IAAI,gBAAgB,GAAG,EAAI;CACrD,MAAM,MAAM,KAAM,YAAY,MAAO,kBAAkB,MAAO;AAC9D,QAAO,KAAK,MAAM,MAAM,IAAI,GAAG;;AAOjC,eAAsB,uBACpB,SACA,QACiC;CACjC,MAAM,iBAAiB,OAAO,oBAAoB;CAClD,MAAM,WAAW,OAAO,aAAa;CACrC,MAAM,eAAe,OAAO,iBAAiB;CAC7C,MAAM,sBAAsB,OAAO,wBAAwB;CAE3D,MAAM,EAAE,YAAY,mBAAmB;AACvC,KAAI,CAAC,eACH,OAAM,IAAI,MACR,mFACD;CAIH,MAAM,cAAc,MAAM,aAAa,SAAS;EAC9C;EACA;EACA;EACA,WAAW,WAAW;EACtB;EACD,CAAC;CAEF,MAAM,QAAsB,EAAE;CAC9B,IAAI,oBAAoB;AAExB,MAAK,MAAM,SAAS,YAAY,QAAQ;EACtC,MAAM,YAAY,MAAM,MAAM,KAAK,MAAM,EAAE,KAAK;AAGhD,MAAI,MAAM,uBAAuB,SAAS,MAAM,OAAO,UAAU,EAAE;AACjE;AACA;;EAIF,MAAM,kBAAkB,iBAAiB,UAAU;EACnD,MAAM,gBAAgB,sBAAsB,UAAU;EACtD,MAAM,aAAa,eAAe,MAAM,MAAM,MAAM,iBAAiB,cAAc;EAMnF,MAAM,cAAsC,MAAM,MAAM,KAAK,GAAG,SAAS;GACvE,YAAY,EAAE;GACd,OAAO,EAAE,SAAS,EAAE,KAAK,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAAE;GACtE,WAAW,KAAK,OAAO,IAAO,MAAM,KAAK,IAAI,MAAM,MAAM,SAAS,GAAG,EAAE,GAAI,MAAO,IAAI,GAAG;GAC1F,EAAE;AAEH,QAAM,KAAK;GACT,IAAI,MAAM;GACV,OAAO,MAAM;GACb,MAAM,MAAM;GACZ;GACA,cAAc;GACd,iBAAiB,iBAAiB,MAAM,MAAM;GAC9C,kBAAkB;GAClB,gBAAgB;GACjB,CAAC;AAEF,MAAI,MAAM,UAAU,SAAU;;AAIhC,OAAM,MAAM,GAAG,MAAM,EAAE,aAAa,EAAE,WAAW;AAEjD,QAAO;EACL;EACA,yBAAyB,YAAY,OAAO,SAAS;EACrD,oBAAoB;EACrB;;AAOH,SAAgB,sBACd,QACA,WACuB;CACvB,MAAM,EAAE,YAAY,OAAO,QAAQ,iBAAiB;CAIpD,MAAM,WAAW,GADC,MAAM,QAAQ,iBAAiB,IAAI,CACvB;CAG9B,MAAM,YAAY,OAAO,QAAQ,cAAc,GAAG;CAClD,MAAM,aAAa,YAAY,GAAG,UAAU,GAAG,aAAa;CAG5D,MAAM,UAAU,KAAK,WAAW,WAAW;CAC3C,MAAM,SAAS,QAAQ,QAAQ;CAG/B,MAAM,YAAY,aACf,KAAK,MAAM;AAIV,SAAO,OAFM,EAAE,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAEvC;GACnB,CACD,KAAK,KAAK;CAEb,MAAM,gBAAgB,aAAa;CAEnC,MAAM,UAAU;EACd,KAAK;EACL;EACA,oCAAoC,WAAW;EAC/C,aAAa,cAAc;EAC3B;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK;AAGZ,WAAU,QAAQ,EAAE,WAAW,MAAM,CAAC;AACtC,eAAc,SAAS,SAAS,QAAQ;AAExC,QAAO;EACL;EACA;EACA;EACD"}
+{"version":3,"file":"latent-ideas-B7wq75Pt.mjs","names":[],"sources":["../src/graph/latent-ideas.ts"],"sourcesContent":["/**\n * latent-ideas.ts — graph_latent_ideas and idea_materialize endpoint handlers\n *\n * \"Latent ideas\" are recurring themes in the vault that exist as embedding\n * clusters but have NO dedicated note written about them yet.  PAI surfaces\n * these by running the same agglomerative clustering used by graph_clusters /\n * zettelThemes and then filtering OUT any cluster whose label is well-matched\n * by an existing note title.\n *\n * The materialize endpoint writes a new Markdown note to the vault filesystem\n * and returns its content so the plugin can open it immediately.\n */\n\nimport { mkdirSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { TITLE_STOP_WORDS } from \"../utils/stop-words.js\";\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport { zettelThemes } from \"../zettelkasten/themes.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphLatentIdeasParams {\n  project_id: number;\n  /** Minimum notes in a cluster (default: 3) */\n  min_cluster_size?: number;\n  /** Cap on returned ideas (default: 15) */\n  max_ideas?: number;\n  /** How far back to look in days (default: 180) */\n  lookback_days?: number;\n  /** Cosine similarity clustering threshold (default: 0.65) */\n  similarity_threshold?: number;\n}\n\nexport interface LatentIdeaSourceNote {\n  vault_path: string;\n  title: string;\n  /** How strongly this note relates to the theme (0-1) */\n  relevance: number;\n}\n\nexport interface LatentIdea {\n  id: number;\n  /** Auto-generated cluster label from zettelThemes */\n  label: string;\n  /** Number of notes touching this theme */\n  size: number;\n  /** 0-1, how likely this is a real coherent idea */\n  confidence: number;\n  /** Notes that contribute to this theme */\n  source_notes: LatentIdeaSourceNote[];\n  /** Cleaned-up version of label for a potential note title */\n  suggested_title: string;\n  /** Most common folder among source notes */\n  suggested_folder: string;\n  /** Number of distinct session date-folders (e.g. \"2026/03\") touching this theme */\n  sessions_count: number;\n}\n\nexport interface GraphLatentIdeasResult {\n  ideas: LatentIdea[];\n  total_clusters_analyzed: number;\n  /** How many clusters already have a matching note (excluded from results) */\n  materialized_count: number;\n}\n\n// ---------------------------------------------------------------------------\n// Materialize params / result\n// ---------------------------------------------------------------------------\n\nexport interface IdeaMaterializeParams {\n  idea_label: string;\n  /** User-chosen title for the new note */\n  title: string;\n  /** Vault-relative folder path where the note should be created */\n  folder: string;\n  /** Vault-relative paths of the source notes to link from the new note */\n  source_paths: string[];\n  project_id: number;\n}\n\nexport interface IdeaMaterializeResult {\n  /** Vault-relative path of the created note */\n  vault_path: string;\n  /** Generated markdown content */\n  content: string;\n  /** Number of wikilinks inserted */\n  links_created: number;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: check if a cluster already has a matching note\n// ---------------------------------------------------------------------------\n\n/**\n * Returns true when any existing vault note title closely matches the cluster\n * label — meaning a dedicated note already exists for this topic.\n *\n * Matching strategy (simple, fast, no embeddings needed):\n *   1. Lowercase both sides and split into words.\n *   2. Remove stop words from the label words.\n *   3. If ≥ 60% of the significant label words appear in a note title → match.\n */\n// TITLE_STOP_WORDS imported from utils/stop-words.ts\n\nfunction labelMatchesTitle(label: string, title: string): boolean {\n  const labelWords = label\n    .toLowerCase()\n    .split(/[\\s\\-_/]+/)\n    .filter((w) => w.length > 2 && !TITLE_STOP_WORDS.has(w));\n\n  if (labelWords.length === 0) return false;\n\n  const titleLower = title.toLowerCase();\n  const matchCount = labelWords.filter((w) => titleLower.includes(w)).length;\n  return matchCount / labelWords.length >= 0.6;\n}\n\n/**\n * Check whether any note indexed in the vault has a title matching the label.\n * Fetches all vault file rows via StorageBackend for efficiency.\n */\nasync function clusterHasMatchingNote(\n  backend: StorageBackend,\n  label: string,\n  notePaths: string[]\n): Promise<boolean> {\n  // First check the notes already in the cluster themselves — if any cluster\n  // member's title matches the label it IS the index note → materialized.\n  const pathSet = new Set(notePaths);\n\n  // Fetch all vault files (bounded — vault rarely > 50k notes)\n  const rows = await backend.getAllVaultFiles();\n\n  for (const row of rows) {\n    if (!row.title) continue;\n    // Skip notes already counted inside the cluster — they don't count as\n    // \"dedicated notes\"; we only skip a cluster if a SEPARATE note exists.\n    if (pathSet.has(row.vaultPath)) continue;\n    if (labelMatchesTitle(label, row.title)) return true;\n  }\n  return false;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: generate a clean suggested title\n// ---------------------------------------------------------------------------\n\nfunction toSuggestedTitle(label: string): string {\n  // Remove leading/trailing whitespace, capitalize each word, remove stop words\n  // that are all-lowercase at the start of the title.\n  const words = label\n    .trim()\n    .split(/\\s+/)\n    .map((w, i) => {\n      const lower = w.toLowerCase();\n      // Drop leading stop words (but keep if they're the only word)\n      if (i === 0 && TITLE_STOP_WORDS.has(lower) && label.trim().split(/\\s+/).length > 1) {\n        return \"\";\n      }\n      return w.charAt(0).toUpperCase() + w.slice(1);\n    })\n    .filter(Boolean);\n\n  return words.join(\" \") || label;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: find most common folder\n// ---------------------------------------------------------------------------\n\nfunction mostCommonFolder(vaultPaths: string[]): string {\n  const counts = new Map<string, number>();\n  for (const p of vaultPaths) {\n    const parts = p.split(\"/\");\n    const folder = parts.length > 1 ? parts.slice(0, -1).join(\"/\") : \"\";\n    counts.set(folder, (counts.get(folder) ?? 0) + 1);\n  }\n\n  let best = \"\";\n  let bestCount = 0;\n  for (const [folder, count] of counts) {\n    if (count > bestCount) {\n      bestCount = count;\n      best = folder;\n    }\n  }\n  return best;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: count distinct session date-folders\n// ---------------------------------------------------------------------------\n\n/**\n * Heuristic: vault notes are often stored in date-based folders like\n * \"2026/03/15\" or \"Daily/2026-03\".  We extract the first numeric path\n * segment that looks like a year (2020-2030) and group by year+month.\n *\n * Falls back to counting distinct top-level folders.\n */\nfunction countDistinctSessions(vaultPaths: string[]): number {\n  const sessions = new Set<string>();\n  const yearMonthRe = /\\b(202\\d)\\D?(0[1-9]|1[0-2])\\b/;\n\n  for (const p of vaultPaths) {\n    const m = yearMonthRe.exec(p);\n    if (m) {\n      sessions.add(`${m[1]}-${m[2]}`);\n    } else {\n      // Fallback: use top-level folder as a proxy for \"session bucket\"\n      const topFolder = p.split(\"/\")[0];\n      sessions.add(topFolder);\n    }\n  }\n  return sessions.size;\n}\n\n// ---------------------------------------------------------------------------\n// Helper: calculate confidence score\n// ---------------------------------------------------------------------------\n\n/**\n * Confidence combines:\n *  - Cluster size (normalized, capped at 20 for max contribution)\n *  - Folder diversity (0-1 already)\n *  - Sessions count (normalized, capped at 5)\n *\n * Formula: 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore\n */\nfunction calcConfidence(\n  size: number,\n  folderDiversity: number,\n  sessionsCount: number\n): number {\n  const sizeScore = Math.min(size / 20, 1.0);\n  const sessionScore = Math.min(sessionsCount / 5, 1.0);\n  const raw = 0.4 * sizeScore + 0.35 * folderDiversity + 0.25 * sessionScore;\n  return Math.round(raw * 100) / 100;\n}\n\n// ---------------------------------------------------------------------------\n// Main handler: graph_latent_ideas\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphLatentIdeas(\n  backend: StorageBackend,\n  params: GraphLatentIdeasParams\n): Promise<GraphLatentIdeasResult> {\n  const minClusterSize = params.min_cluster_size ?? 3;\n  const maxIdeas = params.max_ideas ?? 15;\n  const lookbackDays = params.lookback_days ?? 180;\n  const similarityThreshold = params.similarity_threshold ?? 0.65;\n\n  const { project_id: vaultProjectId } = params;\n  if (!vaultProjectId) {\n    throw new Error(\n      \"graph_latent_ideas: project_id is required (pass the vault project's numeric ID)\"\n    );\n  }\n\n  // Run the same clustering algorithm used by graph_clusters\n  const themeResult = await zettelThemes(backend, {\n    vaultProjectId,\n    lookbackDays,\n    minClusterSize,\n    maxThemes: maxIdeas * 3, // Over-fetch — many will be filtered as materialized\n    similarityThreshold,\n  });\n\n  const ideas: LatentIdea[] = [];\n  let materializedCount = 0;\n\n  for (const theme of themeResult.themes) {\n    const notePaths = theme.notes.map((n) => n.path);\n\n    // Check if a dedicated note already exists for this theme\n    if (await clusterHasMatchingNote(backend, theme.label, notePaths)) {\n      materializedCount++;\n      continue;\n    }\n\n    // This is a latent idea — no dedicated note exists yet\n    const suggestedFolder = mostCommonFolder(notePaths);\n    const sessionsCount = countDistinctSessions(notePaths);\n    const confidence = calcConfidence(theme.size, theme.folderDiversity, sessionsCount);\n\n    // Build source notes with relevance scores\n    // Relevance is approximated by position in cluster (centroid-closest first)\n    // zettelThemes returns notes in no guaranteed order; assign uniform relevance\n    // decreasing from 1.0 to 0.5 across the list.\n    const sourceNotes: LatentIdeaSourceNote[] = theme.notes.map((n, idx) => ({\n      vault_path: n.path,\n      title: n.title ?? n.path.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? n.path,\n      relevance: Math.round((1.0 - (idx / Math.max(theme.notes.length - 1, 1)) * 0.5) * 100) / 100,\n    }));\n\n    ideas.push({\n      id: theme.id,\n      label: theme.label,\n      size: theme.size,\n      confidence,\n      source_notes: sourceNotes,\n      suggested_title: toSuggestedTitle(theme.label),\n      suggested_folder: suggestedFolder,\n      sessions_count: sessionsCount,\n    });\n\n    if (ideas.length >= maxIdeas) break;\n  }\n\n  // Sort by confidence descending\n  ideas.sort((a, b) => b.confidence - a.confidence);\n\n  return {\n    ideas,\n    total_clusters_analyzed: themeResult.themes.length + materializedCount,\n    materialized_count: materializedCount,\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Materialize handler: idea_materialize\n// ---------------------------------------------------------------------------\n\nexport function handleIdeaMaterialize(\n  params: IdeaMaterializeParams,\n  vaultPath: string\n): IdeaMaterializeResult {\n  const { idea_label, title, folder, source_paths } = params;\n\n  // Sanitize filename: replace characters illegal in filenames\n  const safeTitle = title.replace(/[/\\\\:*?\"<>|]/g, \"-\");\n  const fileName = `${safeTitle}.md`;\n\n  // Vault-relative path (forward slashes, no leading slash)\n  const relFolder = folder.replace(/^\\/+|\\/+$/g, \"\");\n  const vault_path = relFolder ? `${relFolder}/${fileName}` : fileName;\n\n  // Absolute filesystem path\n  const absPath = join(vaultPath, vault_path);\n  const absDir = dirname(absPath);\n\n  // Build wikilinks from source_paths\n  const wikilinks = source_paths\n    .map((p) => {\n      // Derive a display name: filename without extension\n      const name = p.split(\"/\").pop()?.replace(/\\.md$/i, \"\") ?? p;\n      // Relative wikilink — use just the filename (Obsidian resolves by title)\n      return `- [[${name}]]`;\n    })\n    .join(\"\\n\");\n\n  const links_created = source_paths.length;\n\n  const content = [\n    `# ${title}`,\n    \"\",\n    `*Materialized from latent idea: \"${idea_label}\"*`,\n    `*Sources: ${links_created} notes*`,\n    \"\",\n    \"## Related Notes\",\n    \"\",\n    wikilinks || \"*(no source notes)*\",\n    \"\",\n    \"## Notes\",\n    \"\",\n    \"<!-- Add your thoughts about this idea here -->\",\n    \"\",\n  ].join(\"\\n\");\n\n  // Write the file (create parent directories as needed)\n  mkdirSync(absDir, { recursive: true });\n  writeFileSync(absPath, content, \"utf-8\");\n\n  return {\n    vault_path,\n    content,\n    links_created,\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AA0GA,SAAS,kBAAkB,OAAe,OAAwB;CAChE,MAAM,aAAa,MAChB,aAAa,CACb,MAAM,YAAY,CAClB,QAAQ,MAAM,EAAE,SAAS,KAAK,CAAC,iBAAiB,IAAI,EAAE,CAAC;AAE1D,KAAI,WAAW,WAAW,EAAG,QAAO;CAEpC,MAAM,aAAa,MAAM,aAAa;AAEtC,QADmB,WAAW,QAAQ,MAAM,WAAW,SAAS,EAAE,CAAC,CAAC,SAChD,WAAW,UAAU;;;;;;AAO3C,eAAe,uBACb,SACA,OACA,WACkB;CAGlB,MAAM,UAAU,IAAI,IAAI,UAAU;CAGlC,MAAM,OAAO,MAAM,QAAQ,kBAAkB;AAE7C,MAAK,MAAM,OAAO,MAAM;AACtB,MAAI,CAAC,IAAI,MAAO;AAGhB,MAAI,QAAQ,IAAI,IAAI,UAAU,CAAE;AAChC,MAAI,kBAAkB,OAAO,IAAI,MAAM,CAAE,QAAO;;AAElD,QAAO;;AAOT,SAAS,iBAAiB,OAAuB;AAgB/C,QAbc,MACX,MAAM,CACN,MAAM,MAAM,CACZ,KAAK,GAAG,MAAM;EACb,MAAM,QAAQ,EAAE,aAAa;AAE7B,MAAI,MAAM,KAAK,iBAAiB,IAAI,MAAM,IAAI,MAAM,MAAM,CAAC,MAAM,MAAM,CAAC,SAAS,EAC/E,QAAO;AAET,SAAO,EAAE,OAAO,EAAE,CAAC,aAAa,GAAG,EAAE,MAAM,EAAE;GAC7C,CACD,OAAO,QAAQ,CAEL,KAAK,IAAI,IAAI;;AAO5B,SAAS,iBAAiB,YAA8B;CACtD,MAAM,yBAAS,IAAI,KAAqB;AACxC,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,QAAQ,EAAE,MAAM,IAAI;EAC1B,MAAM,SAAS,MAAM,SAAS,IAAI,MAAM,MAAM,GAAG,GAAG,CAAC,KAAK,IAAI,GAAG;AACjE,SAAO,IAAI,SAAS,OAAO,IAAI,OAAO,IAAI,KAAK,EAAE;;CAGnD,IAAI,OAAO;CACX,IAAI,YAAY;AAChB,MAAK,MAAM,CAAC,QAAQ,UAAU,OAC5B,KAAI,QAAQ,WAAW;AACrB,cAAY;AACZ,SAAO;;AAGX,QAAO;;;;;;;;;AAcT,SAAS,sBAAsB,YAA8B;CAC3D,MAAM,2BAAW,IAAI,KAAa;CAClC,MAAM,cAAc;AAEpB,MAAK,MAAM,KAAK,YAAY;EAC1B,MAAM,IAAI,YAAY,KAAK,EAAE;AAC7B,MAAI,EACF,UAAS,IAAI,GAAG,EAAE,GAAG,GAAG,EAAE,KAAK;OAC1B;GAEL,MAAM,YAAY,EAAE,MAAM,IAAI,CAAC;AAC/B,YAAS,IAAI,UAAU;;;AAG3B,QAAO,SAAS;;;;;;;;;;AAelB,SAAS,eACP,MACA,iBACA,eACQ;CACR,MAAM,YAAY,KAAK,IAAI,OAAO,IAAI,EAAI;CAC1C,MAAM,eAAe,KAAK,IAAI,gBAAgB,GAAG,EAAI;CACrD,MAAM,MAAM,KAAM,YAAY,MAAO,kBAAkB,MAAO;AAC9D,QAAO,KAAK,MAAM,MAAM,IAAI,GAAG;;AAOjC,eAAsB,uBACpB,SACA,QACiC;CACjC,MAAM,iBAAiB,OAAO,oBAAoB;CAClD,MAAM,WAAW,OAAO,aAAa;CACrC,MAAM,eAAe,OAAO,iBAAiB;CAC7C,MAAM,sBAAsB,OAAO,wBAAwB;CAE3D,MAAM,EAAE,YAAY,mBAAmB;AACvC,KAAI,CAAC,eACH,OAAM,IAAI,MACR,mFACD;CAIH,MAAM,cAAc,MAAM,aAAa,SAAS;EAC9C;EACA;EACA;EACA,WAAW,WAAW;EACtB;EACD,CAAC;CAEF,MAAM,QAAsB,EAAE;CAC9B,IAAI,oBAAoB;AAExB,MAAK,MAAM,SAAS,YAAY,QAAQ;EACtC,MAAM,YAAY,MAAM,MAAM,KAAK,MAAM,EAAE,KAAK;AAGhD,MAAI,MAAM,uBAAuB,SAAS,MAAM,OAAO,UAAU,EAAE;AACjE;AACA;;EAIF,MAAM,kBAAkB,iBAAiB,UAAU;EACnD,MAAM,gBAAgB,sBAAsB,UAAU;EACtD,MAAM,aAAa,eAAe,MAAM,MAAM,MAAM,iBAAiB,cAAc;EAMnF,MAAM,cAAsC,MAAM,MAAM,KAAK,GAAG,SAAS;GACvE,YAAY,EAAE;GACd,OAAO,EAAE,SAAS,EAAE,KAAK,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAAE;GACtE,WAAW,KAAK,OAAO,IAAO,MAAM,KAAK,IAAI,MAAM,MAAM,SAAS,GAAG,EAAE,GAAI,MAAO,IAAI,GAAG;GAC1F,EAAE;AAEH,QAAM,KAAK;GACT,IAAI,MAAM;GACV,OAAO,MAAM;GACb,MAAM,MAAM;GACZ;GACA,cAAc;GACd,iBAAiB,iBAAiB,MAAM,MAAM;GAC9C,kBAAkB;GAClB,gBAAgB;GACjB,CAAC;AAEF,MAAI,MAAM,UAAU,SAAU;;AAIhC,OAAM,MAAM,GAAG,MAAM,EAAE,aAAa,EAAE,WAAW;AAEjD,QAAO;EACL;EACA,yBAAyB,YAAY,OAAO,SAAS;EACrD,oBAAoB;EACrB;;AAOH,SAAgB,sBACd,QACA,WACuB;CACvB,MAAM,EAAE,YAAY,OAAO,QAAQ,iBAAiB;CAIpD,MAAM,WAAW,GADC,MAAM,QAAQ,iBAAiB,IAAI,CACvB;CAG9B,MAAM,YAAY,OAAO,QAAQ,cAAc,GAAG;CAClD,MAAM,aAAa,YAAY,GAAG,UAAU,GAAG,aAAa;CAG5D,MAAM,UAAU,KAAK,WAAW,WAAW;CAC3C,MAAM,SAAS,QAAQ,QAAQ;CAG/B,MAAM,YAAY,aACf,KAAK,MAAM;AAIV,SAAO,OAFM,EAAE,MAAM,IAAI,CAAC,KAAK,EAAE,QAAQ,UAAU,GAAG,IAAI,EAEvC;GACnB,CACD,KAAK,KAAK;CAEb,MAAM,gBAAgB,aAAa;CAEnC,MAAM,UAAU;EACd,KAAK;EACL;EACA,oCAAoC,WAAW;EAC/C,aAAa,cAAc;EAC3B;EACA;EACA;EACA,aAAa;EACb;EACA;EACA;EACA;EACA;EACD,CAAC,KAAK,KAAK;AAGZ,WAAU,QAAQ,EAAE,WAAW,MAAM,CAAC;AACtC,eAAc,SAAS,SAAS,QAAQ;AAExC,QAAO;EACL;EACA;EACA;EACD"}

package/dist/{link-boost-QFLrJwD6.mjs → link-boost-HkG7JWZR.mjs}

@@ -31,4 +31,4 @@ function applyLinkBoost(results, edges, opts) {
 
 //#endregion
 export { applyLinkBoost };
-//# sourceMappingURL=link-boost-QFLrJwD6.mjs.map
+//# sourceMappingURL=link-boost-HkG7JWZR.mjs.map

package/dist/{link-boost-QFLrJwD6.mjs.map → link-boost-HkG7JWZR.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"link-boost-QFLrJwD6.mjs","names":[],"sources":["../src/memory/link-boost.ts"],"sourcesContent":["/**\n * Rank search results by how the corpus links to them, not only by similarity.\n *\n * Why this exists: the store holds 33,709 wikilinks that are *facts* — one note\n * pointing at another, written by a person — alongside 2.4M chunks whose only\n * ranking signal is embedding similarity. Similarity answers \"what reads like\n * the query\". It cannot answer \"which of these is the one the others refer\n * back to\", which is usually the note worth reading first.\n *\n * The boost is deliberately query-local: it counts links *between the results\n * themselves*, not global popularity. A note linked by many other notes that\n * also match the query is a hub for that question. A note linked by half the\n * vault is merely popular, which is not the same thing and would flatten every\n * ranking toward the same few index pages.\n *\n * Links cost nothing to maintain — no embedding pass, no model call — so this\n * signal stays correct while the embedding backlog drains, and works for chunks\n * that have no embedding at all.\n */\n\nimport type { SearchResult } from \"./search.js\";\n\n/** A directed link between two note paths, as stored in vault_links. */\nexport interface LinkEdge {\n  sourcePath: string;\n  targetPath: string;\n}\n\nexport interface LinkBoostOptions {\n  /**\n   * How much the boost may move a result, as a fraction of its current score.\n   * 0.25 means the most-linked result gains 25%. Kept modest by default: the\n   * link graph is a supporting signal, and a note nobody links to can still be\n   * the right answer.\n   */\n  weight?: number;\n}\n\n/**\n * Re-rank results by inbound links *from other results in the same set*.\n *\n * Returns a new array, sorted by the adjusted score. Input is not mutated.\n * Results whose paths carry no inbound links are unchanged, so a corpus with\n * no links at all is a no-op rather than a distortion.\n */\nexport function applyLinkBoost(\n  results: SearchResult[],\n  edges: LinkEdge[],\n  opts?: LinkBoostOptions,\n): SearchResult[] {\n  const weight = opts?.weight ?? 0.25;\n  if (results.length === 0 || edges.length === 0 || weight === 0) {\n    return [...results];\n  }\n\n  // Only links whose BOTH ends are in the result set count. An edge pointing\n  // out of the set says nothing about the relative rank of results inside it.\n  const paths = new Set(results.map((r) => r.path));\n  const inbound = new Map<string, number>();\n  for (const e of edges) {\n    if (e.sourcePath === e.targetPath) continue;      // self-links are noise\n    if (!paths.has(e.sourcePath) || !paths.has(e.targetPath)) continue;\n    inbound.set(e.targetPath, (inbound.get(e.targetPath) ?? 0) + 1);\n  }\n  if (inbound.size === 0) return [...results];\n\n  // Normalise against the most-linked result so the boost is bounded by\n  // `weight` regardless of corpus size. Without this, a densely linked project\n  // would swamp similarity entirely while a sparse one would see no effect.\n  const maxInbound = Math.max(...inbound.values());\n\n  return results\n    .map((r) => {\n      const links = inbound.get(r.path) ?? 0;\n      if (links === 0) return { ...r };\n      const factor = 1 + weight * (links / maxInbound);\n      return { ...r, score: r.score * factor };\n    })\n    .sort((a, b) => b.score - a.score);\n}\n"],"mappings":";;;;;;;;AA6CA,SAAgB,eACd,SACA,OACA,MACgB;CAChB,MAAM,SAAS,MAAM,UAAU;AAC/B,KAAI,QAAQ,WAAW,KAAK,MAAM,WAAW,KAAK,WAAW,EAC3D,QAAO,CAAC,GAAG,QAAQ;CAKrB,MAAM,QAAQ,IAAI,IAAI,QAAQ,KAAK,MAAM,EAAE,KAAK,CAAC;CACjD,MAAM,0BAAU,IAAI,KAAqB;AACzC,MAAK,MAAM,KAAK,OAAO;AACrB,MAAI,EAAE,eAAe,EAAE,WAAY;AACnC,MAAI,CAAC,MAAM,IAAI,EAAE,WAAW,IAAI,CAAC,MAAM,IAAI,EAAE,WAAW,CAAE;AAC1D,UAAQ,IAAI,EAAE,aAAa,QAAQ,IAAI,EAAE,WAAW,IAAI,KAAK,EAAE;;AAEjE,KAAI,QAAQ,SAAS,EAAG,QAAO,CAAC,GAAG,QAAQ;CAK3C,MAAM,aAAa,KAAK,IAAI,GAAG,QAAQ,QAAQ,CAAC;AAEhD,QAAO,QACJ,KAAK,MAAM;EACV,MAAM,QAAQ,QAAQ,IAAI,EAAE,KAAK,IAAI;AACrC,MAAI,UAAU,EAAG,QAAO,EAAE,GAAG,GAAG;EAChC,MAAM,SAAS,IAAI,UAAU,QAAQ;AACrC,SAAO;GAAE,GAAG;GAAG,OAAO,EAAE,QAAQ;GAAQ;GACxC,CACD,MAAM,GAAG,MAAM,EAAE,QAAQ,EAAE,MAAM"}
+{"version":3,"file":"link-boost-HkG7JWZR.mjs","names":[],"sources":["../src/memory/link-boost.ts"],"sourcesContent":["/**\n * Rank search results by how the corpus links to them, not only by similarity.\n *\n * Why this exists: the store holds 33,709 wikilinks that are *facts* — one note\n * pointing at another, written by a person — alongside 2.4M chunks whose only\n * ranking signal is embedding similarity. Similarity answers \"what reads like\n * the query\". It cannot answer \"which of these is the one the others refer\n * back to\", which is usually the note worth reading first.\n *\n * The boost is deliberately query-local: it counts links *between the results\n * themselves*, not global popularity. A note linked by many other notes that\n * also match the query is a hub for that question. A note linked by half the\n * vault is merely popular, which is not the same thing and would flatten every\n * ranking toward the same few index pages.\n *\n * Links cost nothing to maintain — no embedding pass, no model call — so this\n * signal stays correct while the embedding backlog drains, and works for chunks\n * that have no embedding at all.\n */\n\nimport type { SearchResult } from \"./search.js\";\n\n/** A directed link between two note paths, as stored in vault_links. */\nexport interface LinkEdge {\n  sourcePath: string;\n  targetPath: string;\n}\n\nexport interface LinkBoostOptions {\n  /**\n   * How much the boost may move a result, as a fraction of its current score.\n   * 0.25 means the most-linked result gains 25%. Kept modest by default: the\n   * link graph is a supporting signal, and a note nobody links to can still be\n   * the right answer.\n   */\n  weight?: number;\n}\n\n/**\n * Re-rank results by inbound links *from other results in the same set*.\n *\n * Returns a new array, sorted by the adjusted score. Input is not mutated.\n * Results whose paths carry no inbound links are unchanged, so a corpus with\n * no links at all is a no-op rather than a distortion.\n */\nexport function applyLinkBoost(\n  results: SearchResult[],\n  edges: LinkEdge[],\n  opts?: LinkBoostOptions,\n): SearchResult[] {\n  const weight = opts?.weight ?? 0.25;\n  if (results.length === 0 || edges.length === 0 || weight === 0) {\n    return [...results];\n  }\n\n  // Only links whose BOTH ends are in the result set count. An edge pointing\n  // out of the set says nothing about the relative rank of results inside it.\n  const paths = new Set(results.map((r) => r.path));\n  const inbound = new Map<string, number>();\n  for (const e of edges) {\n    if (e.sourcePath === e.targetPath) continue;      // self-links are noise\n    if (!paths.has(e.sourcePath) || !paths.has(e.targetPath)) continue;\n    inbound.set(e.targetPath, (inbound.get(e.targetPath) ?? 0) + 1);\n  }\n  if (inbound.size === 0) return [...results];\n\n  // Normalise against the most-linked result so the boost is bounded by\n  // `weight` regardless of corpus size. Without this, a densely linked project\n  // would swamp similarity entirely while a sparse one would see no effect.\n  const maxInbound = Math.max(...inbound.values());\n\n  return results\n    .map((r) => {\n      const links = inbound.get(r.path) ?? 0;\n      if (links === 0) return { ...r };\n      const factor = 1 + weight * (links / maxInbound);\n      return { ...r, score: r.score * factor };\n    })\n    .sort((a, b) => b.score - a.score);\n}\n"],"mappings":";;;;;;;;AA6CA,SAAgB,eACd,SACA,OACA,MACgB;CAChB,MAAM,SAAS,MAAM,UAAU;AAC/B,KAAI,QAAQ,WAAW,KAAK,MAAM,WAAW,KAAK,WAAW,EAC3D,QAAO,CAAC,GAAG,QAAQ;CAKrB,MAAM,QAAQ,IAAI,IAAI,QAAQ,KAAK,MAAM,EAAE,KAAK,CAAC;CACjD,MAAM,0BAAU,IAAI,KAAqB;AACzC,MAAK,MAAM,KAAK,OAAO;AACrB,MAAI,EAAE,eAAe,EAAE,WAAY;AACnC,MAAI,CAAC,MAAM,IAAI,EAAE,WAAW,IAAI,CAAC,MAAM,IAAI,EAAE,WAAW,CAAE;AAC1D,UAAQ,IAAI,EAAE,aAAa,QAAQ,IAAI,EAAE,WAAW,IAAI,KAAK,EAAE;;AAEjE,KAAI,QAAQ,SAAS,EAAG,QAAO,CAAC,GAAG,QAAQ;CAK3C,MAAM,aAAa,KAAK,IAAI,GAAG,QAAQ,QAAQ,CAAC;AAEhD,QAAO,QACJ,KAAK,MAAM;EACV,MAAM,QAAQ,QAAQ,IAAI,EAAE,KAAK,IAAI;AACrC,MAAI,UAAU,EAAG,QAAO,EAAE,GAAG,GAAG;EAChC,MAAM,SAAS,IAAI,UAAU,QAAQ;AACrC,SAAO;GAAE,GAAG;GAAG,OAAO,EAAE,QAAQ;GAAQ;GACxC,CACD,MAAM,GAAG,MAAM,EAAE,QAAQ,EAAE,MAAM"}

package/dist/{neighborhood-D9MJ1c8f.mjs → neighborhood-lThN-MaQ.mjs}

@@ -132,4 +132,4 @@ async function handleGraphNeighborhood(pool, backend, params) {
 
 //#endregion
 export { handleGraphNeighborhood };
-//# sourceMappingURL=neighborhood-D9MJ1c8f.mjs.map
+//# sourceMappingURL=neighborhood-lThN-MaQ.mjs.map

package/dist/{neighborhood-D9MJ1c8f.mjs.map → neighborhood-lThN-MaQ.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"neighborhood-D9MJ1c8f.mjs","names":[],"sources":["../src/graph/neighborhood.ts"],"sourcesContent":["/**\n * neighborhood.ts — graph_neighborhood endpoint handler\n *\n * Given a set of vault note paths (typically the notes inside a cluster),\n * returns the individual note nodes and the wikilink edges between them.\n *\n * Optionally enriches with semantic edges computed from cosine similarity\n * between chunk embeddings stored in the federation database.\n */\n\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport type { Pool } from \"pg\";\nimport { deserializeEmbedding } from \"../memory/embeddings.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphNeighborhoodParams {\n  /** Vault-relative paths of notes in the cluster */\n  vault_paths: string[];\n  /** Numeric PAI project ID */\n  project_id: number;\n  /** Whether to compute semantic similarity edges (default: false) */\n  include_semantic_edges?: boolean;\n  /** Cosine similarity threshold for semantic edges (default: 0.7) */\n  semantic_threshold?: number;\n}\n\nexport interface NoteNode {\n  vault_path: string;\n  title: string;\n  folder: string;\n  observation_types: Record<string, number>;\n  dominant_type: string;\n  updated_at: number;\n  word_count: number;\n}\n\nexport interface NoteEdge {\n  source: string;\n  target: string;\n  type: \"wikilink\" | \"semantic\";\n  weight: number;\n}\n\nexport interface GraphNeighborhoodResult {\n  nodes: NoteNode[];\n  edges: NoteEdge[];\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction folderFromPath(vaultPath: string): string {\n  const lastSlash = vaultPath.lastIndexOf(\"/\");\n  return lastSlash === -1 ? \"\" : vaultPath.slice(0, lastSlash);\n}\n\nfunction cosineSimilarity(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0;\n  let dot = 0;\n  let normA = 0;\n  let normB = 0;\n  for (let i = 0; i < a.length; i++) {\n    dot += a[i] * b[i];\n    normA += a[i] * a[i];\n    normB += b[i] * b[i];\n  }\n  if (normA === 0 || normB === 0) return 0;\n  return dot / (Math.sqrt(normA) * Math.sqrt(normB));\n}\n\nfunction dominantType(counts: Record<string, number>): string {\n  let dominant = \"unknown\";\n  let maxCount = 0;\n  for (const [type, n] of Object.entries(counts)) {\n    if (n > maxCount) {\n      maxCount = n;\n      dominant = type;\n    }\n  }\n  return dominant;\n}\n\n// ---------------------------------------------------------------------------\n// Observation type enrichment (same pattern as clusters.ts)\n// ---------------------------------------------------------------------------\n\nasync function fetchObservationTypes(\n  pool: Pool,\n  filePaths: string[],\n  projectId: number\n): Promise<Map<string, Record<string, number>>> {\n  if (filePaths.length === 0) return new Map();\n\n  try {\n    const params: (string[] | number)[] = [filePaths, projectId];\n\n    const result = await pool.query<{ path: string; type: string; cnt: string }>(\n      `SELECT unnested_path AS path, type, COUNT(*) AS cnt\n       FROM pai_observations,\n            LATERAL unnest(files_modified || files_read) AS unnested_path\n       WHERE unnested_path = ANY($1::text[])\n         AND project_id = $2\n       GROUP BY unnested_path, type`,\n      params\n    );\n\n    const byPath = new Map<string, Record<string, number>>();\n    for (const row of result.rows) {\n      const existing = byPath.get(row.path) ?? {};\n      existing[row.type] = (existing[row.type] ?? 0) + parseInt(row.cnt, 10);\n      byPath.set(row.path, existing);\n    }\n    return byPath;\n  } catch {\n    return new Map();\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Main handler\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphNeighborhood(\n  pool: Pool | null,\n  backend: StorageBackend,\n  params: GraphNeighborhoodParams\n): Promise<GraphNeighborhoodResult> {\n  const vaultPaths = params.vault_paths ?? [];\n  if (vaultPaths.length === 0) {\n    return { nodes: [], edges: [] };\n  }\n\n  const includeSemanticEdges = params.include_semantic_edges ?? false;\n  const semanticThreshold = params.semantic_threshold ?? 0.7;\n\n  // -------------------------------------------------------------------------\n  // 1. Fetch node metadata from vault_files\n  // -------------------------------------------------------------------------\n\n  const fileRows = await backend.getVaultFilesByPaths(vaultPaths);\n\n  const fileIndex = new Map<string, { vaultPath: string; title: string | null; indexedAt: number }>();\n  for (const row of fileRows) {\n    fileIndex.set(row.vaultPath, row);\n  }\n\n  // -------------------------------------------------------------------------\n  // 2. Fetch observation types (Postgres if available)\n  // -------------------------------------------------------------------------\n\n  const observationsByPath =\n    pool !== null\n      ? await fetchObservationTypes(pool, vaultPaths, params.project_id)\n      : new Map<string, Record<string, number>>();\n\n  // -------------------------------------------------------------------------\n  // 3. Build NoteNode array\n  // -------------------------------------------------------------------------\n\n  const nodes: NoteNode[] = vaultPaths.map((vp) => {\n    const fileRow = fileIndex.get(vp);\n    const fileName = vp.split(\"/\").pop() ?? vp;\n    const rawTitle = fileRow?.title ?? fileName.replace(/\\.md$/i, \"\");\n\n    const obsCounts = observationsByPath.get(vp) ?? {};\n\n    return {\n      vault_path: vp,\n      title: rawTitle,\n      folder: folderFromPath(vp),\n      observation_types: obsCounts,\n      dominant_type: dominantType(obsCounts),\n      updated_at: fileRow?.indexedAt ?? 0,\n      word_count: 0,\n    };\n  });\n\n  // -------------------------------------------------------------------------\n  // 4. Fetch wikilink edges between the provided paths\n  // -------------------------------------------------------------------------\n\n  const pathSet = new Set(vaultPaths);\n  const linkRows = await backend.getVaultLinksFromPaths(vaultPaths);\n\n  const edges: NoteEdge[] = [];\n\n  for (const row of linkRows) {\n    if (!row.targetPath || !pathSet.has(row.targetPath)) continue;\n\n    edges.push({\n      source: row.sourcePath,\n      target: row.targetPath,\n      type: \"wikilink\",\n      weight: 1.0,\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // 5. Optional: semantic edges\n  // -------------------------------------------------------------------------\n\n  if (includeSemanticEdges && vaultPaths.length > 1) {\n    // Fetch mean embeddings for all paths\n    const embeddings = new Map<string, number[]>();\n    for (const vp of vaultPaths) {\n      const chunkRows = await backend.getChunksForPath(params.project_id, vp);\n      const embRows = chunkRows.filter(r => r.embedding !== null) as Array<{ text: string; embedding: Buffer }>;\n      if (embRows.length === 0) continue;\n\n      let vecLen = 0;\n      const vectors: Float32Array[] = [];\n\n      for (const row of embRows) {\n        const arr = deserializeEmbedding(row.embedding);\n        if (vecLen === 0) vecLen = arr.length;\n        if (arr.length === vecLen) vectors.push(arr);\n      }\n\n      if (vectors.length === 0 || vecLen === 0) continue;\n\n      const mean = new Array<number>(vecLen).fill(0);\n      for (const vec of vectors) {\n        for (let i = 0; i < vecLen; i++) {\n          mean[i] += vec[i];\n        }\n      }\n      for (let i = 0; i < vecLen; i++) {\n        mean[i] /= vectors.length;\n      }\n      embeddings.set(vp, mean);\n    }\n\n    const existingEdgeKeys = new Set<string>(\n      edges.map((e) => `${e.source}|||${e.target}`)\n    );\n\n    const pathsWithEmbeddings = Array.from(embeddings.keys());\n    for (let i = 0; i < pathsWithEmbeddings.length; i++) {\n      for (let j = i + 1; j < pathsWithEmbeddings.length; j++) {\n        const pathA = pathsWithEmbeddings[i];\n        const pathB = pathsWithEmbeddings[j];\n\n        const vecA = embeddings.get(pathA)!;\n        const vecB = embeddings.get(pathB)!;\n\n        const sim = cosineSimilarity(vecA, vecB);\n        if (sim < semanticThreshold) continue;\n\n        const keyAB = `${pathA}|||${pathB}`;\n        const keyBA = `${pathB}|||${pathA}`;\n        if (existingEdgeKeys.has(keyAB) || existingEdgeKeys.has(keyBA)) continue;\n\n        edges.push({\n          source: pathA,\n          target: pathB,\n          type: \"semantic\",\n          weight: sim,\n        });\n        existingEdgeKeys.add(keyAB);\n      }\n    }\n  }\n\n  return { nodes, edges };\n}\n"],"mappings":";;;AAuDA,SAAS,eAAe,WAA2B;CACjD,MAAM,YAAY,UAAU,YAAY,IAAI;AAC5C,QAAO,cAAc,KAAK,KAAK,UAAU,MAAM,GAAG,UAAU;;AAG9D,SAAS,iBAAiB,GAAa,GAAqB;AAC1D,KAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;CACpD,IAAI,MAAM;CACV,IAAI,QAAQ;CACZ,IAAI,QAAQ;AACZ,MAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAO,EAAE,KAAK,EAAE;AAChB,WAAS,EAAE,KAAK,EAAE;AAClB,WAAS,EAAE,KAAK,EAAE;;AAEpB,KAAI,UAAU,KAAK,UAAU,EAAG,QAAO;AACvC,QAAO,OAAO,KAAK,KAAK,MAAM,GAAG,KAAK,KAAK,MAAM;;AAGnD,SAAS,aAAa,QAAwC;CAC5D,IAAI,WAAW;CACf,IAAI,WAAW;AACf,MAAK,MAAM,CAAC,MAAM,MAAM,OAAO,QAAQ,OAAO,CAC5C,KAAI,IAAI,UAAU;AAChB,aAAW;AACX,aAAW;;AAGf,QAAO;;AAOT,eAAe,sBACb,MACA,WACA,WAC8C;AAC9C,KAAI,UAAU,WAAW,EAAG,wBAAO,IAAI,KAAK;AAE5C,KAAI;EACF,MAAM,SAAgC,CAAC,WAAW,UAAU;EAE5D,MAAM,SAAS,MAAM,KAAK,MACxB;;;;;sCAMA,OACD;EAED,MAAM,yBAAS,IAAI,KAAqC;AACxD,OAAK,MAAM,OAAO,OAAO,MAAM;GAC7B,MAAM,WAAW,OAAO,IAAI,IAAI,KAAK,IAAI,EAAE;AAC3C,YAAS,IAAI,SAAS,SAAS,IAAI,SAAS,KAAK,SAAS,IAAI,KAAK,GAAG;AACtE,UAAO,IAAI,IAAI,MAAM,SAAS;;AAEhC,SAAO;SACD;AACN,yBAAO,IAAI,KAAK;;;AAQpB,eAAsB,wBACpB,MACA,SACA,QACkC;CAClC,MAAM,aAAa,OAAO,eAAe,EAAE;AAC3C,KAAI,WAAW,WAAW,EACxB,QAAO;EAAE,OAAO,EAAE;EAAE,OAAO,EAAE;EAAE;CAGjC,MAAM,uBAAuB,OAAO,0BAA0B;CAC9D,MAAM,oBAAoB,OAAO,sBAAsB;CAMvD,MAAM,WAAW,MAAM,QAAQ,qBAAqB,WAAW;CAE/D,MAAM,4BAAY,IAAI,KAA6E;AACnG,MAAK,MAAM,OAAO,SAChB,WAAU,IAAI,IAAI,WAAW,IAAI;CAOnC,MAAM,qBACJ,SAAS,OACL,MAAM,sBAAsB,MAAM,YAAY,OAAO,WAAW,mBAChE,IAAI,KAAqC;CAM/C,MAAM,QAAoB,WAAW,KAAK,OAAO;EAC/C,MAAM,UAAU,UAAU,IAAI,GAAG;EACjC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,KAAK,IAAI;EACxC,MAAM,WAAW,SAAS,SAAS,SAAS,QAAQ,UAAU,GAAG;EAEjE,MAAM,YAAY,mBAAmB,IAAI,GAAG,IAAI,EAAE;AAElD,SAAO;GACL,YAAY;GACZ,OAAO;GACP,QAAQ,eAAe,GAAG;GAC1B,mBAAmB;GACnB,eAAe,aAAa,UAAU;GACtC,YAAY,SAAS,aAAa;GAClC,YAAY;GACb;GACD;CAMF,MAAM,UAAU,IAAI,IAAI,WAAW;CACnC,MAAM,WAAW,MAAM,QAAQ,uBAAuB,WAAW;CAEjE,MAAM,QAAoB,EAAE;AAE5B,MAAK,MAAM,OAAO,UAAU;AAC1B,MAAI,CAAC,IAAI,cAAc,CAAC,QAAQ,IAAI,IAAI,WAAW,CAAE;AAErD,QAAM,KAAK;GACT,QAAQ,IAAI;GACZ,QAAQ,IAAI;GACZ,MAAM;GACN,QAAQ;GACT,CAAC;;AAOJ,KAAI,wBAAwB,WAAW,SAAS,GAAG;EAEjD,MAAM,6BAAa,IAAI,KAAuB;AAC9C,OAAK,MAAM,MAAM,YAAY;GAE3B,MAAM,WADY,MAAM,QAAQ,iBAAiB,OAAO,YAAY,GAAG,EAC7C,QAAO,MAAK,EAAE,cAAc,KAAK;AAC3D,OAAI,QAAQ,WAAW,EAAG;GAE1B,IAAI,SAAS;GACb,MAAM,UAA0B,EAAE;AAElC,QAAK,MAAM,OAAO,SAAS;IACzB,MAAM,MAAM,qBAAqB,IAAI,UAAU;AAC/C,QAAI,WAAW,EAAG,UAAS,IAAI;AAC/B,QAAI,IAAI,WAAW,OAAQ,SAAQ,KAAK,IAAI;;AAG9C,OAAI,QAAQ,WAAW,KAAK,WAAW,EAAG;GAE1C,MAAM,OAAO,IAAI,MAAc,OAAO,CAAC,KAAK,EAAE;AAC9C,QAAK,MAAM,OAAO,QAChB,MAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,IAAI;AAGnB,QAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,QAAQ;AAErB,cAAW,IAAI,IAAI,KAAK;;EAG1B,MAAM,mBAAmB,IAAI,IAC3B,MAAM,KAAK,MAAM,GAAG,EAAE,OAAO,KAAK,EAAE,SAAS,CAC9C;EAED,MAAM,sBAAsB,MAAM,KAAK,WAAW,MAAM,CAAC;AACzD,OAAK,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,IAC9C,MAAK,IAAI,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,KAAK;GACvD,MAAM,QAAQ,oBAAoB;GAClC,MAAM,QAAQ,oBAAoB;GAKlC,MAAM,MAAM,iBAHC,WAAW,IAAI,MAAM,EACrB,WAAW,IAAI,MAAM,CAEM;AACxC,OAAI,MAAM,kBAAmB;GAE7B,MAAM,QAAQ,GAAG,MAAM,KAAK;GAC5B,MAAM,QAAQ,GAAG,MAAM,KAAK;AAC5B,OAAI,iBAAiB,IAAI,MAAM,IAAI,iBAAiB,IAAI,MAAM,CAAE;AAEhE,SAAM,KAAK;IACT,QAAQ;IACR,QAAQ;IACR,MAAM;IACN,QAAQ;IACT,CAAC;AACF,oBAAiB,IAAI,MAAM;;;AAKjC,QAAO;EAAE;EAAO;EAAO"}
+{"version":3,"file":"neighborhood-lThN-MaQ.mjs","names":[],"sources":["../src/graph/neighborhood.ts"],"sourcesContent":["/**\n * neighborhood.ts — graph_neighborhood endpoint handler\n *\n * Given a set of vault note paths (typically the notes inside a cluster),\n * returns the individual note nodes and the wikilink edges between them.\n *\n * Optionally enriches with semantic edges computed from cosine similarity\n * between chunk embeddings stored in the federation database.\n */\n\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport type { Pool } from \"pg\";\nimport { deserializeEmbedding } from \"../memory/embeddings.js\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphNeighborhoodParams {\n  /** Vault-relative paths of notes in the cluster */\n  vault_paths: string[];\n  /** Numeric PAI project ID */\n  project_id: number;\n  /** Whether to compute semantic similarity edges (default: false) */\n  include_semantic_edges?: boolean;\n  /** Cosine similarity threshold for semantic edges (default: 0.7) */\n  semantic_threshold?: number;\n}\n\nexport interface NoteNode {\n  vault_path: string;\n  title: string;\n  folder: string;\n  observation_types: Record<string, number>;\n  dominant_type: string;\n  updated_at: number;\n  word_count: number;\n}\n\nexport interface NoteEdge {\n  source: string;\n  target: string;\n  type: \"wikilink\" | \"semantic\";\n  weight: number;\n}\n\nexport interface GraphNeighborhoodResult {\n  nodes: NoteNode[];\n  edges: NoteEdge[];\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction folderFromPath(vaultPath: string): string {\n  const lastSlash = vaultPath.lastIndexOf(\"/\");\n  return lastSlash === -1 ? \"\" : vaultPath.slice(0, lastSlash);\n}\n\nfunction cosineSimilarity(a: number[], b: number[]): number {\n  if (a.length !== b.length || a.length === 0) return 0;\n  let dot = 0;\n  let normA = 0;\n  let normB = 0;\n  for (let i = 0; i < a.length; i++) {\n    dot += a[i] * b[i];\n    normA += a[i] * a[i];\n    normB += b[i] * b[i];\n  }\n  if (normA === 0 || normB === 0) return 0;\n  return dot / (Math.sqrt(normA) * Math.sqrt(normB));\n}\n\nfunction dominantType(counts: Record<string, number>): string {\n  let dominant = \"unknown\";\n  let maxCount = 0;\n  for (const [type, n] of Object.entries(counts)) {\n    if (n > maxCount) {\n      maxCount = n;\n      dominant = type;\n    }\n  }\n  return dominant;\n}\n\n// ---------------------------------------------------------------------------\n// Observation type enrichment (same pattern as clusters.ts)\n// ---------------------------------------------------------------------------\n\nasync function fetchObservationTypes(\n  pool: Pool,\n  filePaths: string[],\n  projectId: number\n): Promise<Map<string, Record<string, number>>> {\n  if (filePaths.length === 0) return new Map();\n\n  try {\n    const params: (string[] | number)[] = [filePaths, projectId];\n\n    const result = await pool.query<{ path: string; type: string; cnt: string }>(\n      `SELECT unnested_path AS path, type, COUNT(*) AS cnt\n       FROM pai_observations,\n            LATERAL unnest(files_modified || files_read) AS unnested_path\n       WHERE unnested_path = ANY($1::text[])\n         AND project_id = $2\n       GROUP BY unnested_path, type`,\n      params\n    );\n\n    const byPath = new Map<string, Record<string, number>>();\n    for (const row of result.rows) {\n      const existing = byPath.get(row.path) ?? {};\n      existing[row.type] = (existing[row.type] ?? 0) + parseInt(row.cnt, 10);\n      byPath.set(row.path, existing);\n    }\n    return byPath;\n  } catch {\n    return new Map();\n  }\n}\n\n// ---------------------------------------------------------------------------\n// Main handler\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphNeighborhood(\n  pool: Pool | null,\n  backend: StorageBackend,\n  params: GraphNeighborhoodParams\n): Promise<GraphNeighborhoodResult> {\n  const vaultPaths = params.vault_paths ?? [];\n  if (vaultPaths.length === 0) {\n    return { nodes: [], edges: [] };\n  }\n\n  const includeSemanticEdges = params.include_semantic_edges ?? false;\n  const semanticThreshold = params.semantic_threshold ?? 0.7;\n\n  // -------------------------------------------------------------------------\n  // 1. Fetch node metadata from vault_files\n  // -------------------------------------------------------------------------\n\n  const fileRows = await backend.getVaultFilesByPaths(vaultPaths);\n\n  const fileIndex = new Map<string, { vaultPath: string; title: string | null; indexedAt: number }>();\n  for (const row of fileRows) {\n    fileIndex.set(row.vaultPath, row);\n  }\n\n  // -------------------------------------------------------------------------\n  // 2. Fetch observation types (Postgres if available)\n  // -------------------------------------------------------------------------\n\n  const observationsByPath =\n    pool !== null\n      ? await fetchObservationTypes(pool, vaultPaths, params.project_id)\n      : new Map<string, Record<string, number>>();\n\n  // -------------------------------------------------------------------------\n  // 3. Build NoteNode array\n  // -------------------------------------------------------------------------\n\n  const nodes: NoteNode[] = vaultPaths.map((vp) => {\n    const fileRow = fileIndex.get(vp);\n    const fileName = vp.split(\"/\").pop() ?? vp;\n    const rawTitle = fileRow?.title ?? fileName.replace(/\\.md$/i, \"\");\n\n    const obsCounts = observationsByPath.get(vp) ?? {};\n\n    return {\n      vault_path: vp,\n      title: rawTitle,\n      folder: folderFromPath(vp),\n      observation_types: obsCounts,\n      dominant_type: dominantType(obsCounts),\n      updated_at: fileRow?.indexedAt ?? 0,\n      word_count: 0,\n    };\n  });\n\n  // -------------------------------------------------------------------------\n  // 4. Fetch wikilink edges between the provided paths\n  // -------------------------------------------------------------------------\n\n  const pathSet = new Set(vaultPaths);\n  const linkRows = await backend.getVaultLinksFromPaths(vaultPaths);\n\n  const edges: NoteEdge[] = [];\n\n  for (const row of linkRows) {\n    if (!row.targetPath || !pathSet.has(row.targetPath)) continue;\n\n    edges.push({\n      source: row.sourcePath,\n      target: row.targetPath,\n      type: \"wikilink\",\n      weight: 1.0,\n    });\n  }\n\n  // -------------------------------------------------------------------------\n  // 5. Optional: semantic edges\n  // -------------------------------------------------------------------------\n\n  if (includeSemanticEdges && vaultPaths.length > 1) {\n    // Fetch mean embeddings for all paths\n    const embeddings = new Map<string, number[]>();\n    for (const vp of vaultPaths) {\n      const chunkRows = await backend.getChunksForPath(params.project_id, vp);\n      const embRows = chunkRows.filter(r => r.embedding !== null) as Array<{ text: string; embedding: Buffer }>;\n      if (embRows.length === 0) continue;\n\n      let vecLen = 0;\n      const vectors: Float32Array[] = [];\n\n      for (const row of embRows) {\n        const arr = deserializeEmbedding(row.embedding);\n        if (vecLen === 0) vecLen = arr.length;\n        if (arr.length === vecLen) vectors.push(arr);\n      }\n\n      if (vectors.length === 0 || vecLen === 0) continue;\n\n      const mean = new Array<number>(vecLen).fill(0);\n      for (const vec of vectors) {\n        for (let i = 0; i < vecLen; i++) {\n          mean[i] += vec[i];\n        }\n      }\n      for (let i = 0; i < vecLen; i++) {\n        mean[i] /= vectors.length;\n      }\n      embeddings.set(vp, mean);\n    }\n\n    const existingEdgeKeys = new Set<string>(\n      edges.map((e) => `${e.source}|||${e.target}`)\n    );\n\n    const pathsWithEmbeddings = Array.from(embeddings.keys());\n    for (let i = 0; i < pathsWithEmbeddings.length; i++) {\n      for (let j = i + 1; j < pathsWithEmbeddings.length; j++) {\n        const pathA = pathsWithEmbeddings[i];\n        const pathB = pathsWithEmbeddings[j];\n\n        const vecA = embeddings.get(pathA)!;\n        const vecB = embeddings.get(pathB)!;\n\n        const sim = cosineSimilarity(vecA, vecB);\n        if (sim < semanticThreshold) continue;\n\n        const keyAB = `${pathA}|||${pathB}`;\n        const keyBA = `${pathB}|||${pathA}`;\n        if (existingEdgeKeys.has(keyAB) || existingEdgeKeys.has(keyBA)) continue;\n\n        edges.push({\n          source: pathA,\n          target: pathB,\n          type: \"semantic\",\n          weight: sim,\n        });\n        existingEdgeKeys.add(keyAB);\n      }\n    }\n  }\n\n  return { nodes, edges };\n}\n"],"mappings":";;;AAuDA,SAAS,eAAe,WAA2B;CACjD,MAAM,YAAY,UAAU,YAAY,IAAI;AAC5C,QAAO,cAAc,KAAK,KAAK,UAAU,MAAM,GAAG,UAAU;;AAG9D,SAAS,iBAAiB,GAAa,GAAqB;AAC1D,KAAI,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAG,QAAO;CACpD,IAAI,MAAM;CACV,IAAI,QAAQ;CACZ,IAAI,QAAQ;AACZ,MAAK,IAAI,IAAI,GAAG,IAAI,EAAE,QAAQ,KAAK;AACjC,SAAO,EAAE,KAAK,EAAE;AAChB,WAAS,EAAE,KAAK,EAAE;AAClB,WAAS,EAAE,KAAK,EAAE;;AAEpB,KAAI,UAAU,KAAK,UAAU,EAAG,QAAO;AACvC,QAAO,OAAO,KAAK,KAAK,MAAM,GAAG,KAAK,KAAK,MAAM;;AAGnD,SAAS,aAAa,QAAwC;CAC5D,IAAI,WAAW;CACf,IAAI,WAAW;AACf,MAAK,MAAM,CAAC,MAAM,MAAM,OAAO,QAAQ,OAAO,CAC5C,KAAI,IAAI,UAAU;AAChB,aAAW;AACX,aAAW;;AAGf,QAAO;;AAOT,eAAe,sBACb,MACA,WACA,WAC8C;AAC9C,KAAI,UAAU,WAAW,EAAG,wBAAO,IAAI,KAAK;AAE5C,KAAI;EACF,MAAM,SAAgC,CAAC,WAAW,UAAU;EAE5D,MAAM,SAAS,MAAM,KAAK,MACxB;;;;;sCAMA,OACD;EAED,MAAM,yBAAS,IAAI,KAAqC;AACxD,OAAK,MAAM,OAAO,OAAO,MAAM;GAC7B,MAAM,WAAW,OAAO,IAAI,IAAI,KAAK,IAAI,EAAE;AAC3C,YAAS,IAAI,SAAS,SAAS,IAAI,SAAS,KAAK,SAAS,IAAI,KAAK,GAAG;AACtE,UAAO,IAAI,IAAI,MAAM,SAAS;;AAEhC,SAAO;SACD;AACN,yBAAO,IAAI,KAAK;;;AAQpB,eAAsB,wBACpB,MACA,SACA,QACkC;CAClC,MAAM,aAAa,OAAO,eAAe,EAAE;AAC3C,KAAI,WAAW,WAAW,EACxB,QAAO;EAAE,OAAO,EAAE;EAAE,OAAO,EAAE;EAAE;CAGjC,MAAM,uBAAuB,OAAO,0BAA0B;CAC9D,MAAM,oBAAoB,OAAO,sBAAsB;CAMvD,MAAM,WAAW,MAAM,QAAQ,qBAAqB,WAAW;CAE/D,MAAM,4BAAY,IAAI,KAA6E;AACnG,MAAK,MAAM,OAAO,SAChB,WAAU,IAAI,IAAI,WAAW,IAAI;CAOnC,MAAM,qBACJ,SAAS,OACL,MAAM,sBAAsB,MAAM,YAAY,OAAO,WAAW,mBAChE,IAAI,KAAqC;CAM/C,MAAM,QAAoB,WAAW,KAAK,OAAO;EAC/C,MAAM,UAAU,UAAU,IAAI,GAAG;EACjC,MAAM,WAAW,GAAG,MAAM,IAAI,CAAC,KAAK,IAAI;EACxC,MAAM,WAAW,SAAS,SAAS,SAAS,QAAQ,UAAU,GAAG;EAEjE,MAAM,YAAY,mBAAmB,IAAI,GAAG,IAAI,EAAE;AAElD,SAAO;GACL,YAAY;GACZ,OAAO;GACP,QAAQ,eAAe,GAAG;GAC1B,mBAAmB;GACnB,eAAe,aAAa,UAAU;GACtC,YAAY,SAAS,aAAa;GAClC,YAAY;GACb;GACD;CAMF,MAAM,UAAU,IAAI,IAAI,WAAW;CACnC,MAAM,WAAW,MAAM,QAAQ,uBAAuB,WAAW;CAEjE,MAAM,QAAoB,EAAE;AAE5B,MAAK,MAAM,OAAO,UAAU;AAC1B,MAAI,CAAC,IAAI,cAAc,CAAC,QAAQ,IAAI,IAAI,WAAW,CAAE;AAErD,QAAM,KAAK;GACT,QAAQ,IAAI;GACZ,QAAQ,IAAI;GACZ,MAAM;GACN,QAAQ;GACT,CAAC;;AAOJ,KAAI,wBAAwB,WAAW,SAAS,GAAG;EAEjD,MAAM,6BAAa,IAAI,KAAuB;AAC9C,OAAK,MAAM,MAAM,YAAY;GAE3B,MAAM,WADY,MAAM,QAAQ,iBAAiB,OAAO,YAAY,GAAG,EAC7C,QAAO,MAAK,EAAE,cAAc,KAAK;AAC3D,OAAI,QAAQ,WAAW,EAAG;GAE1B,IAAI,SAAS;GACb,MAAM,UAA0B,EAAE;AAElC,QAAK,MAAM,OAAO,SAAS;IACzB,MAAM,MAAM,qBAAqB,IAAI,UAAU;AAC/C,QAAI,WAAW,EAAG,UAAS,IAAI;AAC/B,QAAI,IAAI,WAAW,OAAQ,SAAQ,KAAK,IAAI;;AAG9C,OAAI,QAAQ,WAAW,KAAK,WAAW,EAAG;GAE1C,MAAM,OAAO,IAAI,MAAc,OAAO,CAAC,KAAK,EAAE;AAC9C,QAAK,MAAM,OAAO,QAChB,MAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,IAAI;AAGnB,QAAK,IAAI,IAAI,GAAG,IAAI,QAAQ,IAC1B,MAAK,MAAM,QAAQ;AAErB,cAAW,IAAI,IAAI,KAAK;;EAG1B,MAAM,mBAAmB,IAAI,IAC3B,MAAM,KAAK,MAAM,GAAG,EAAE,OAAO,KAAK,EAAE,SAAS,CAC9C;EAED,MAAM,sBAAsB,MAAM,KAAK,WAAW,MAAM,CAAC;AACzD,OAAK,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,IAC9C,MAAK,IAAI,IAAI,IAAI,GAAG,IAAI,oBAAoB,QAAQ,KAAK;GACvD,MAAM,QAAQ,oBAAoB;GAClC,MAAM,QAAQ,oBAAoB;GAKlC,MAAM,MAAM,iBAHC,WAAW,IAAI,MAAM,EACrB,WAAW,IAAI,MAAM,CAEM;AACxC,OAAI,MAAM,kBAAmB;GAE7B,MAAM,QAAQ,GAAG,MAAM,KAAK;GAC5B,MAAM,QAAQ,GAAG,MAAM,KAAK;AAC5B,OAAI,iBAAiB,IAAI,MAAM,IAAI,iBAAiB,IAAI,MAAM,CAAE;AAEhE,SAAM,KAAK;IACT,QAAQ;IACR,QAAQ;IACR,MAAM;IACN,QAAQ;IACT,CAAC;AACF,oBAAiB,IAAI,MAAM;;;AAKjC,QAAO;EAAE;EAAO;EAAO"}

package/dist/{note-context-d1wT_-GA.mjs → note-context-b6k0mAKi.mjs}

@@ -123,4 +123,4 @@ async function handleGraphNoteContext(pool, backend, params) {
 
 //#endregion
 export { handleGraphNoteContext };
-//# sourceMappingURL=note-context-d1wT_-GA.mjs.map
+//# sourceMappingURL=note-context-b6k0mAKi.mjs.map

package/dist/{note-context-d1wT_-GA.mjs.map → note-context-b6k0mAKi.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"note-context-d1wT_-GA.mjs","names":[],"sources":["../src/graph/note-context.ts"],"sourcesContent":["/**\n * note-context.ts — graph_note_context endpoint handler\n *\n * Given a single vault note path, returns ALL notes linked to or from it\n * across the entire vault (1-hop neighbourhood), plus the edges.\n */\n\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport type { Pool } from \"pg\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphNoteContextParams {\n  vault_path: string;\n  project_id: number;\n  max_neighbors?: number;\n  include_backlinks?: boolean;\n  include_outlinks?: boolean;\n}\n\nexport interface NoteNode {\n  vault_path: string;\n  title: string;\n  folder: string;\n  observation_types: Record<string, number>;\n  dominant_type: string;\n  updated_at: number;\n  word_count: number;\n}\n\nexport interface NoteEdge {\n  source: string;\n  target: string;\n  type: \"wikilink\" | \"semantic\";\n  weight: number;\n}\n\nexport interface GraphNoteContextResult {\n  focal: NoteNode;\n  neighbors: NoteNode[];\n  edges: NoteEdge[];\n  cluster_membership: Record<string, number>;\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction folderFromPath(vaultPath: string): string {\n  const lastSlash = vaultPath.lastIndexOf(\"/\");\n  return lastSlash === -1 ? \"\" : vaultPath.slice(0, lastSlash);\n}\n\nfunction dominantType(counts: Record<string, number>): string {\n  let best = \"unknown\";\n  let maxCount = 0;\n  for (const [type, n] of Object.entries(counts)) {\n    if (n > maxCount) {\n      maxCount = n;\n      best = type;\n    }\n  }\n  return best;\n}\n\n// ---------------------------------------------------------------------------\n// Observation type enrichment\n// ---------------------------------------------------------------------------\n\nasync function fetchObservationTypes(\n  pool: Pool,\n  filePaths: string[],\n  projectId: number\n): Promise<Map<string, Record<string, number>>> {\n  if (filePaths.length === 0) return new Map();\n\n  try {\n    const result = await pool.query<{ path: string; type: string; cnt: string }>(\n      `SELECT unnested_path AS path, type, COUNT(*) AS cnt\n       FROM pai_observations,\n            LATERAL unnest(files_modified || files_read) AS unnested_path\n       WHERE unnested_path = ANY($1::text[])\n         AND project_id = $2\n       GROUP BY unnested_path, type`,\n      [filePaths, projectId]\n    );\n\n    const byPath = new Map<string, Record<string, number>>();\n    for (const row of result.rows) {\n      const existing = byPath.get(row.path) ?? {};\n      existing[row.type] = (existing[row.type] ?? 0) + parseInt(row.cnt, 10);\n      byPath.set(row.path, existing);\n    }\n    return byPath;\n  } catch {\n    return new Map();\n  }\n}\n\nfunction buildNoteNode(\n  vaultPath: string,\n  fileIndex: Map<string, { title: string | null; indexedAt: number }>,\n  obsByPath: Map<string, Record<string, number>>\n): NoteNode {\n  const fileRow = fileIndex.get(vaultPath);\n  const fileName = vaultPath.split(\"/\").pop() ?? vaultPath;\n  const rawTitle = fileRow?.title ?? fileName.replace(/\\.md$/i, \"\");\n  const obsCounts = obsByPath.get(vaultPath) ?? {};\n\n  return {\n    vault_path: vaultPath,\n    title: rawTitle,\n    folder: folderFromPath(vaultPath),\n    observation_types: obsCounts,\n    dominant_type: dominantType(obsCounts),\n    updated_at: fileRow?.indexedAt ?? 0,\n    word_count: 0,\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Main handler\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphNoteContext(\n  pool: Pool | null,\n  backend: StorageBackend,\n  params: GraphNoteContextParams\n): Promise<GraphNoteContextResult> {\n  const focalPath = params.vault_path;\n  if (!focalPath) {\n    throw new Error(\"graph_note_context: vault_path is required\");\n  }\n\n  const maxNeighbors = params.max_neighbors ?? 50;\n  const includeBacklinks = params.include_backlinks !== false;\n  const includeOutlinks = params.include_outlinks !== false;\n\n  // -------------------------------------------------------------------------\n  // 1. Collect 1-hop neighbor paths via vault_links\n  // -------------------------------------------------------------------------\n\n  const neighborPaths = new Set<string>();\n  const rawEdges: Array<{ source: string; target: string }> = [];\n\n  if (includeOutlinks) {\n    const outLinks = await backend.getLinksFromSource(focalPath);\n    for (const link of outLinks) {\n      if (!link.targetPath) continue;\n      neighborPaths.add(link.targetPath);\n      rawEdges.push({ source: focalPath, target: link.targetPath });\n    }\n  }\n\n  if (includeBacklinks) {\n    const inLinks = await backend.getLinksToTarget(focalPath);\n    for (const link of inLinks) {\n      neighborPaths.add(link.sourcePath);\n      rawEdges.push({ source: link.sourcePath, target: focalPath });\n    }\n  }\n\n  // Cap neighbors at max_neighbors, keeping the most-linked ones\n  let neighborPathList = Array.from(neighborPaths);\n  if (neighborPathList.length > maxNeighbors) {\n    const linkCount = new Map<string, number>();\n    for (const e of rawEdges) {\n      const neighbor = e.source === focalPath ? e.target : e.source;\n      linkCount.set(neighbor, (linkCount.get(neighbor) ?? 0) + 1);\n    }\n    neighborPathList = neighborPathList\n      .sort((a, b) => (linkCount.get(b) ?? 0) - (linkCount.get(a) ?? 0))\n      .slice(0, maxNeighbors);\n  }\n\n  const retainedSet = new Set(neighborPathList);\n  const retainedEdges = rawEdges.filter((e) => {\n    const neighbor = e.source === focalPath ? e.target : e.source;\n    return retainedSet.has(neighbor);\n  });\n\n  // -------------------------------------------------------------------------\n  // 2. Fetch vault_files metadata for focal + all neighbors\n  // -------------------------------------------------------------------------\n\n  const allPaths = [focalPath, ...neighborPathList];\n  const fileRows = await backend.getVaultFilesByPaths(allPaths);\n  const fileIndex = new Map<string, { title: string | null; indexedAt: number }>(\n    fileRows.map(f => [f.vaultPath, { title: f.title, indexedAt: f.indexedAt }])\n  );\n\n  // -------------------------------------------------------------------------\n  // 3. Observation type enrichment (Postgres if available)\n  // -------------------------------------------------------------------------\n\n  const obsByPath =\n    pool !== null\n      ? await fetchObservationTypes(pool, allPaths, params.project_id)\n      : new Map<string, Record<string, number>>();\n\n  // -------------------------------------------------------------------------\n  // 4. Build focal NoteNode\n  // -------------------------------------------------------------------------\n\n  const focal = buildNoteNode(focalPath, fileIndex, obsByPath);\n\n  // -------------------------------------------------------------------------\n  // 5. Build neighbor NoteNode array\n  // -------------------------------------------------------------------------\n\n  const neighbors: NoteNode[] = neighborPathList.map((vp) =>\n    buildNoteNode(vp, fileIndex, obsByPath)\n  );\n\n  // -------------------------------------------------------------------------\n  // 6. Deduplicate edges\n  // -------------------------------------------------------------------------\n\n  const edgeKeys = new Set<string>();\n  const edges: NoteEdge[] = [];\n  for (const e of retainedEdges) {\n    const key = `${e.source}|||${e.target}`;\n    if (!edgeKeys.has(key)) {\n      edgeKeys.add(key);\n      edges.push({\n        source: e.source,\n        target: e.target,\n        type: \"wikilink\",\n        weight: 1.0,\n      });\n    }\n  }\n\n  return {\n    focal,\n    neighbors,\n    edges,\n    cluster_membership: {},\n  };\n}\n"],"mappings":";AAkDA,SAAS,eAAe,WAA2B;CACjD,MAAM,YAAY,UAAU,YAAY,IAAI;AAC5C,QAAO,cAAc,KAAK,KAAK,UAAU,MAAM,GAAG,UAAU;;AAG9D,SAAS,aAAa,QAAwC;CAC5D,IAAI,OAAO;CACX,IAAI,WAAW;AACf,MAAK,MAAM,CAAC,MAAM,MAAM,OAAO,QAAQ,OAAO,CAC5C,KAAI,IAAI,UAAU;AAChB,aAAW;AACX,SAAO;;AAGX,QAAO;;AAOT,eAAe,sBACb,MACA,WACA,WAC8C;AAC9C,KAAI,UAAU,WAAW,EAAG,wBAAO,IAAI,KAAK;AAE5C,KAAI;EACF,MAAM,SAAS,MAAM,KAAK,MACxB;;;;;sCAMA,CAAC,WAAW,UAAU,CACvB;EAED,MAAM,yBAAS,IAAI,KAAqC;AACxD,OAAK,MAAM,OAAO,OAAO,MAAM;GAC7B,MAAM,WAAW,OAAO,IAAI,IAAI,KAAK,IAAI,EAAE;AAC3C,YAAS,IAAI,SAAS,SAAS,IAAI,SAAS,KAAK,SAAS,IAAI,KAAK,GAAG;AACtE,UAAO,IAAI,IAAI,MAAM,SAAS;;AAEhC,SAAO;SACD;AACN,yBAAO,IAAI,KAAK;;;AAIpB,SAAS,cACP,WACA,WACA,WACU;CACV,MAAM,UAAU,UAAU,IAAI,UAAU;CACxC,MAAM,WAAW,UAAU,MAAM,IAAI,CAAC,KAAK,IAAI;CAC/C,MAAM,WAAW,SAAS,SAAS,SAAS,QAAQ,UAAU,GAAG;CACjE,MAAM,YAAY,UAAU,IAAI,UAAU,IAAI,EAAE;AAEhD,QAAO;EACL,YAAY;EACZ,OAAO;EACP,QAAQ,eAAe,UAAU;EACjC,mBAAmB;EACnB,eAAe,aAAa,UAAU;EACtC,YAAY,SAAS,aAAa;EAClC,YAAY;EACb;;AAOH,eAAsB,uBACpB,MACA,SACA,QACiC;CACjC,MAAM,YAAY,OAAO;AACzB,KAAI,CAAC,UACH,OAAM,IAAI,MAAM,6CAA6C;CAG/D,MAAM,eAAe,OAAO,iBAAiB;CAC7C,MAAM,mBAAmB,OAAO,sBAAsB;CACtD,MAAM,kBAAkB,OAAO,qBAAqB;CAMpD,MAAM,gCAAgB,IAAI,KAAa;CACvC,MAAM,WAAsD,EAAE;AAE9D,KAAI,iBAAiB;EACnB,MAAM,WAAW,MAAM,QAAQ,mBAAmB,UAAU;AAC5D,OAAK,MAAM,QAAQ,UAAU;AAC3B,OAAI,CAAC,KAAK,WAAY;AACtB,iBAAc,IAAI,KAAK,WAAW;AAClC,YAAS,KAAK;IAAE,QAAQ;IAAW,QAAQ,KAAK;IAAY,CAAC;;;AAIjE,KAAI,kBAAkB;EACpB,MAAM,UAAU,MAAM,QAAQ,iBAAiB,UAAU;AACzD,OAAK,MAAM,QAAQ,SAAS;AAC1B,iBAAc,IAAI,KAAK,WAAW;AAClC,YAAS,KAAK;IAAE,QAAQ,KAAK;IAAY,QAAQ;IAAW,CAAC;;;CAKjE,IAAI,mBAAmB,MAAM,KAAK,cAAc;AAChD,KAAI,iBAAiB,SAAS,cAAc;EAC1C,MAAM,4BAAY,IAAI,KAAqB;AAC3C,OAAK,MAAM,KAAK,UAAU;GACxB,MAAM,WAAW,EAAE,WAAW,YAAY,EAAE,SAAS,EAAE;AACvD,aAAU,IAAI,WAAW,UAAU,IAAI,SAAS,IAAI,KAAK,EAAE;;AAE7D,qBAAmB,iBAChB,MAAM,GAAG,OAAO,UAAU,IAAI,EAAE,IAAI,MAAM,UAAU,IAAI,EAAE,IAAI,GAAG,CACjE,MAAM,GAAG,aAAa;;CAG3B,MAAM,cAAc,IAAI,IAAI,iBAAiB;CAC7C,MAAM,gBAAgB,SAAS,QAAQ,MAAM;EAC3C,MAAM,WAAW,EAAE,WAAW,YAAY,EAAE,SAAS,EAAE;AACvD,SAAO,YAAY,IAAI,SAAS;GAChC;CAMF,MAAM,WAAW,CAAC,WAAW,GAAG,iBAAiB;CACjD,MAAM,WAAW,MAAM,QAAQ,qBAAqB,SAAS;CAC7D,MAAM,YAAY,IAAI,IACpB,SAAS,KAAI,MAAK,CAAC,EAAE,WAAW;EAAE,OAAO,EAAE;EAAO,WAAW,EAAE;EAAW,CAAC,CAAC,CAC7E;CAMD,MAAM,YACJ,SAAS,OACL,MAAM,sBAAsB,MAAM,UAAU,OAAO,WAAW,mBAC9D,IAAI,KAAqC;CAM/C,MAAM,QAAQ,cAAc,WAAW,WAAW,UAAU;CAM5D,MAAM,YAAwB,iBAAiB,KAAK,OAClD,cAAc,IAAI,WAAW,UAAU,CACxC;CAMD,MAAM,2BAAW,IAAI,KAAa;CAClC,MAAM,QAAoB,EAAE;AAC5B,MAAK,MAAM,KAAK,eAAe;EAC7B,MAAM,MAAM,GAAG,EAAE,OAAO,KAAK,EAAE;AAC/B,MAAI,CAAC,SAAS,IAAI,IAAI,EAAE;AACtB,YAAS,IAAI,IAAI;AACjB,SAAM,KAAK;IACT,QAAQ,EAAE;IACV,QAAQ,EAAE;IACV,MAAM;IACN,QAAQ;IACT,CAAC;;;AAIN,QAAO;EACL;EACA;EACA;EACA,oBAAoB,EAAE;EACvB"}
+{"version":3,"file":"note-context-b6k0mAKi.mjs","names":[],"sources":["../src/graph/note-context.ts"],"sourcesContent":["/**\n * note-context.ts — graph_note_context endpoint handler\n *\n * Given a single vault note path, returns ALL notes linked to or from it\n * across the entire vault (1-hop neighbourhood), plus the edges.\n */\n\nimport type { StorageBackend } from \"../storage/interface.js\";\nimport type { Pool } from \"pg\";\n\n// ---------------------------------------------------------------------------\n// Public param / result types\n// ---------------------------------------------------------------------------\n\nexport interface GraphNoteContextParams {\n  vault_path: string;\n  project_id: number;\n  max_neighbors?: number;\n  include_backlinks?: boolean;\n  include_outlinks?: boolean;\n}\n\nexport interface NoteNode {\n  vault_path: string;\n  title: string;\n  folder: string;\n  observation_types: Record<string, number>;\n  dominant_type: string;\n  updated_at: number;\n  word_count: number;\n}\n\nexport interface NoteEdge {\n  source: string;\n  target: string;\n  type: \"wikilink\" | \"semantic\";\n  weight: number;\n}\n\nexport interface GraphNoteContextResult {\n  focal: NoteNode;\n  neighbors: NoteNode[];\n  edges: NoteEdge[];\n  cluster_membership: Record<string, number>;\n}\n\n// ---------------------------------------------------------------------------\n// Helpers\n// ---------------------------------------------------------------------------\n\nfunction folderFromPath(vaultPath: string): string {\n  const lastSlash = vaultPath.lastIndexOf(\"/\");\n  return lastSlash === -1 ? \"\" : vaultPath.slice(0, lastSlash);\n}\n\nfunction dominantType(counts: Record<string, number>): string {\n  let best = \"unknown\";\n  let maxCount = 0;\n  for (const [type, n] of Object.entries(counts)) {\n    if (n > maxCount) {\n      maxCount = n;\n      best = type;\n    }\n  }\n  return best;\n}\n\n// ---------------------------------------------------------------------------\n// Observation type enrichment\n// ---------------------------------------------------------------------------\n\nasync function fetchObservationTypes(\n  pool: Pool,\n  filePaths: string[],\n  projectId: number\n): Promise<Map<string, Record<string, number>>> {\n  if (filePaths.length === 0) return new Map();\n\n  try {\n    const result = await pool.query<{ path: string; type: string; cnt: string }>(\n      `SELECT unnested_path AS path, type, COUNT(*) AS cnt\n       FROM pai_observations,\n            LATERAL unnest(files_modified || files_read) AS unnested_path\n       WHERE unnested_path = ANY($1::text[])\n         AND project_id = $2\n       GROUP BY unnested_path, type`,\n      [filePaths, projectId]\n    );\n\n    const byPath = new Map<string, Record<string, number>>();\n    for (const row of result.rows) {\n      const existing = byPath.get(row.path) ?? {};\n      existing[row.type] = (existing[row.type] ?? 0) + parseInt(row.cnt, 10);\n      byPath.set(row.path, existing);\n    }\n    return byPath;\n  } catch {\n    return new Map();\n  }\n}\n\nfunction buildNoteNode(\n  vaultPath: string,\n  fileIndex: Map<string, { title: string | null; indexedAt: number }>,\n  obsByPath: Map<string, Record<string, number>>\n): NoteNode {\n  const fileRow = fileIndex.get(vaultPath);\n  const fileName = vaultPath.split(\"/\").pop() ?? vaultPath;\n  const rawTitle = fileRow?.title ?? fileName.replace(/\\.md$/i, \"\");\n  const obsCounts = obsByPath.get(vaultPath) ?? {};\n\n  return {\n    vault_path: vaultPath,\n    title: rawTitle,\n    folder: folderFromPath(vaultPath),\n    observation_types: obsCounts,\n    dominant_type: dominantType(obsCounts),\n    updated_at: fileRow?.indexedAt ?? 0,\n    word_count: 0,\n  };\n}\n\n// ---------------------------------------------------------------------------\n// Main handler\n// ---------------------------------------------------------------------------\n\nexport async function handleGraphNoteContext(\n  pool: Pool | null,\n  backend: StorageBackend,\n  params: GraphNoteContextParams\n): Promise<GraphNoteContextResult> {\n  const focalPath = params.vault_path;\n  if (!focalPath) {\n    throw new Error(\"graph_note_context: vault_path is required\");\n  }\n\n  const maxNeighbors = params.max_neighbors ?? 50;\n  const includeBacklinks = params.include_backlinks !== false;\n  const includeOutlinks = params.include_outlinks !== false;\n\n  // -------------------------------------------------------------------------\n  // 1. Collect 1-hop neighbor paths via vault_links\n  // -------------------------------------------------------------------------\n\n  const neighborPaths = new Set<string>();\n  const rawEdges: Array<{ source: string; target: string }> = [];\n\n  if (includeOutlinks) {\n    const outLinks = await backend.getLinksFromSource(focalPath);\n    for (const link of outLinks) {\n      if (!link.targetPath) continue;\n      neighborPaths.add(link.targetPath);\n      rawEdges.push({ source: focalPath, target: link.targetPath });\n    }\n  }\n\n  if (includeBacklinks) {\n    const inLinks = await backend.getLinksToTarget(focalPath);\n    for (const link of inLinks) {\n      neighborPaths.add(link.sourcePath);\n      rawEdges.push({ source: link.sourcePath, target: focalPath });\n    }\n  }\n\n  // Cap neighbors at max_neighbors, keeping the most-linked ones\n  let neighborPathList = Array.from(neighborPaths);\n  if (neighborPathList.length > maxNeighbors) {\n    const linkCount = new Map<string, number>();\n    for (const e of rawEdges) {\n      const neighbor = e.source === focalPath ? e.target : e.source;\n      linkCount.set(neighbor, (linkCount.get(neighbor) ?? 0) + 1);\n    }\n    neighborPathList = neighborPathList\n      .sort((a, b) => (linkCount.get(b) ?? 0) - (linkCount.get(a) ?? 0))\n      .slice(0, maxNeighbors);\n  }\n\n  const retainedSet = new Set(neighborPathList);\n  const retainedEdges = rawEdges.filter((e) => {\n    const neighbor = e.source === focalPath ? e.target : e.source;\n    return retainedSet.has(neighbor);\n  });\n\n  // -------------------------------------------------------------------------\n  // 2. Fetch vault_files metadata for focal + all neighbors\n  // -------------------------------------------------------------------------\n\n  const allPaths = [focalPath, ...neighborPathList];\n  const fileRows = await backend.getVaultFilesByPaths(allPaths);\n  const fileIndex = new Map<string, { title: string | null; indexedAt: number }>(\n    fileRows.map(f => [f.vaultPath, { title: f.title, indexedAt: f.indexedAt }])\n  );\n\n  // -------------------------------------------------------------------------\n  // 3. Observation type enrichment (Postgres if available)\n  // -------------------------------------------------------------------------\n\n  const obsByPath =\n    pool !== null\n      ? await fetchObservationTypes(pool, allPaths, params.project_id)\n      : new Map<string, Record<string, number>>();\n\n  // -------------------------------------------------------------------------\n  // 4. Build focal NoteNode\n  // -------------------------------------------------------------------------\n\n  const focal = buildNoteNode(focalPath, fileIndex, obsByPath);\n\n  // -------------------------------------------------------------------------\n  // 5. Build neighbor NoteNode array\n  // -------------------------------------------------------------------------\n\n  const neighbors: NoteNode[] = neighborPathList.map((vp) =>\n    buildNoteNode(vp, fileIndex, obsByPath)\n  );\n\n  // -------------------------------------------------------------------------\n  // 6. Deduplicate edges\n  // -------------------------------------------------------------------------\n\n  const edgeKeys = new Set<string>();\n  const edges: NoteEdge[] = [];\n  for (const e of retainedEdges) {\n    const key = `${e.source}|||${e.target}`;\n    if (!edgeKeys.has(key)) {\n      edgeKeys.add(key);\n      edges.push({\n        source: e.source,\n        target: e.target,\n        type: \"wikilink\",\n        weight: 1.0,\n      });\n    }\n  }\n\n  return {\n    focal,\n    neighbors,\n    edges,\n    cluster_membership: {},\n  };\n}\n"],"mappings":";AAkDA,SAAS,eAAe,WAA2B;CACjD,MAAM,YAAY,UAAU,YAAY,IAAI;AAC5C,QAAO,cAAc,KAAK,KAAK,UAAU,MAAM,GAAG,UAAU;;AAG9D,SAAS,aAAa,QAAwC;CAC5D,IAAI,OAAO;CACX,IAAI,WAAW;AACf,MAAK,MAAM,CAAC,MAAM,MAAM,OAAO,QAAQ,OAAO,CAC5C,KAAI,IAAI,UAAU;AAChB,aAAW;AACX,SAAO;;AAGX,QAAO;;AAOT,eAAe,sBACb,MACA,WACA,WAC8C;AAC9C,KAAI,UAAU,WAAW,EAAG,wBAAO,IAAI,KAAK;AAE5C,KAAI;EACF,MAAM,SAAS,MAAM,KAAK,MACxB;;;;;sCAMA,CAAC,WAAW,UAAU,CACvB;EAED,MAAM,yBAAS,IAAI,KAAqC;AACxD,OAAK,MAAM,OAAO,OAAO,MAAM;GAC7B,MAAM,WAAW,OAAO,IAAI,IAAI,KAAK,IAAI,EAAE;AAC3C,YAAS,IAAI,SAAS,SAAS,IAAI,SAAS,KAAK,SAAS,IAAI,KAAK,GAAG;AACtE,UAAO,IAAI,IAAI,MAAM,SAAS;;AAEhC,SAAO;SACD;AACN,yBAAO,IAAI,KAAK;;;AAIpB,SAAS,cACP,WACA,WACA,WACU;CACV,MAAM,UAAU,UAAU,IAAI,UAAU;CACxC,MAAM,WAAW,UAAU,MAAM,IAAI,CAAC,KAAK,IAAI;CAC/C,MAAM,WAAW,SAAS,SAAS,SAAS,QAAQ,UAAU,GAAG;CACjE,MAAM,YAAY,UAAU,IAAI,UAAU,IAAI,EAAE;AAEhD,QAAO;EACL,YAAY;EACZ,OAAO;EACP,QAAQ,eAAe,UAAU;EACjC,mBAAmB;EACnB,eAAe,aAAa,UAAU;EACtC,YAAY,SAAS,aAAa;EAClC,YAAY;EACb;;AAOH,eAAsB,uBACpB,MACA,SACA,QACiC;CACjC,MAAM,YAAY,OAAO;AACzB,KAAI,CAAC,UACH,OAAM,IAAI,MAAM,6CAA6C;CAG/D,MAAM,eAAe,OAAO,iBAAiB;CAC7C,MAAM,mBAAmB,OAAO,sBAAsB;CACtD,MAAM,kBAAkB,OAAO,qBAAqB;CAMpD,MAAM,gCAAgB,IAAI,KAAa;CACvC,MAAM,WAAsD,EAAE;AAE9D,KAAI,iBAAiB;EACnB,MAAM,WAAW,MAAM,QAAQ,mBAAmB,UAAU;AAC5D,OAAK,MAAM,QAAQ,UAAU;AAC3B,OAAI,CAAC,KAAK,WAAY;AACtB,iBAAc,IAAI,KAAK,WAAW;AAClC,YAAS,KAAK;IAAE,QAAQ;IAAW,QAAQ,KAAK;IAAY,CAAC;;;AAIjE,KAAI,kBAAkB;EACpB,MAAM,UAAU,MAAM,QAAQ,iBAAiB,UAAU;AACzD,OAAK,MAAM,QAAQ,SAAS;AAC1B,iBAAc,IAAI,KAAK,WAAW;AAClC,YAAS,KAAK;IAAE,QAAQ,KAAK;IAAY,QAAQ;IAAW,CAAC;;;CAKjE,IAAI,mBAAmB,MAAM,KAAK,cAAc;AAChD,KAAI,iBAAiB,SAAS,cAAc;EAC1C,MAAM,4BAAY,IAAI,KAAqB;AAC3C,OAAK,MAAM,KAAK,UAAU;GACxB,MAAM,WAAW,EAAE,WAAW,YAAY,EAAE,SAAS,EAAE;AACvD,aAAU,IAAI,WAAW,UAAU,IAAI,SAAS,IAAI,KAAK,EAAE;;AAE7D,qBAAmB,iBAChB,MAAM,GAAG,OAAO,UAAU,IAAI,EAAE,IAAI,MAAM,UAAU,IAAI,EAAE,IAAI,GAAG,CACjE,MAAM,GAAG,aAAa;;CAG3B,MAAM,cAAc,IAAI,IAAI,iBAAiB;CAC7C,MAAM,gBAAgB,SAAS,QAAQ,MAAM;EAC3C,MAAM,WAAW,EAAE,WAAW,YAAY,EAAE,SAAS,EAAE;AACvD,SAAO,YAAY,IAAI,SAAS;GAChC;CAMF,MAAM,WAAW,CAAC,WAAW,GAAG,iBAAiB;CACjD,MAAM,WAAW,MAAM,QAAQ,qBAAqB,SAAS;CAC7D,MAAM,YAAY,IAAI,IACpB,SAAS,KAAI,MAAK,CAAC,EAAE,WAAW;EAAE,OAAO,EAAE;EAAO,WAAW,EAAE;EAAW,CAAC,CAAC,CAC7E;CAMD,MAAM,YACJ,SAAS,OACL,MAAM,sBAAsB,MAAM,UAAU,OAAO,WAAW,mBAC9D,IAAI,KAAqC;CAM/C,MAAM,QAAQ,cAAc,WAAW,WAAW,UAAU;CAM5D,MAAM,YAAwB,iBAAiB,KAAK,OAClD,cAAc,IAAI,WAAW,UAAU,CACxC;CAMD,MAAM,2BAAW,IAAI,KAAa;CAClC,MAAM,QAAoB,EAAE;AAC5B,MAAK,MAAM,KAAK,eAAe;EAC7B,MAAM,MAAM,GAAG,EAAE,OAAO,KAAK,EAAE;AAC/B,MAAI,CAAC,SAAS,IAAI,IAAI,EAAE;AACtB,YAAS,IAAI,IAAI;AACjB,SAAM,KAAK;IACT,QAAQ,EAAE;IACV,QAAQ,EAAE;IACV,MAAM;IACN,QAAQ;IACT,CAAC;;;AAIN,QAAO;EACL;EACA;EACA;EACA,oBAAoB,EAAE;EACvB"}

package/dist/planner-BDI7bE9B.mjs

@@ -0,0 +1,243 @@
+import "./config-B64vFg14.mjs";
+import { B as workersLogDir, I as parseRunnerArgs, J as readWorkersSection, L as shortText, M as loadStatuses, N as newWorkerId, P as appendLedger, i as runWorker, n as swapPromptArg, r as printResult, z as ledgerPath } from "./chain-CTtHligI.mjs";
+import { existsSync, mkdirSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+
+//#region src/workers/planner.ts
+/**
+* planner.ts — the plan class: plan, spawn, integrate.
+*
+* `pai worker run --class plan -p "<goal>"` does not run one worker but a
+* small orchestration:
+*
+*   1. a planner worker (the plan class's provider) reads the repository and
+*      writes `<logDir>/plans/<planner id>.json` — 5–50 sub-tasks, fewer only
+*      when the goal itself names a smaller count, each with title, brief,
+*      class, files and acceptance;
+*   2. the runner validates the plan and spawns the sub-tasks as children of
+*      the planner worker (`parent` = the planner id, so `ps` shows the
+*      tree), at most `workers.tree.maxChildren` at a time;
+*   3. each child that finishes delivers its structured report to the
+*      planner's inbox as a `kind: "result"` handoff (see handoff.ts);
+*   4. the run finishes with a summary report and the branches to merge.
+*
+* The planner's prompt carries the prompt rules from the self-driving
+* codebases write-up: domain-specific instructions only, constraints over
+* step lists, explicit quantity ranges, no checkbox style.
+*/
+/** Where a planner run's plan file lives: <logDir>/plans/<planner id>.json. */
+function planPathFor(logDir, plannerId) {
+	return join(logDir, "plans", `${plannerId}.json`);
+}
+const MIN_TASKS = 5;
+const MAX_TASKS = 50;
+/**
+* Validate a raw plan file into its tasks. Accepts 1–MAX_TASKS tasks: the
+* prompt asks for MIN_TASKS–MAX_TASKS, but a goal may itself name a smaller
+* explicit count ("one sub-task per file" with three files), and that
+* explicit range wins. Everything else — missing file, damaged JSON, tasks
+* without a title or brief — throws with a message the operator can act on.
+*/
+function validatePlan(raw) {
+	if (typeof raw !== "object" || raw === null || Array.isArray(raw)) throw new Error("plan file must be a JSON object");
+	const tasksRaw = raw.tasks;
+	if (!Array.isArray(tasksRaw)) throw new Error("plan file needs a \"tasks\" array");
+	if (tasksRaw.length < 1) throw new Error("plan file has no tasks");
+	if (tasksRaw.length > MAX_TASKS) throw new Error(`plan file has ${tasksRaw.length} tasks; the maximum is ${MAX_TASKS}`);
+	const tasks = [];
+	for (let i = 0; i < tasksRaw.length; i++) {
+		const t = tasksRaw[i];
+		if (typeof t !== "object" || t === null || Array.isArray(t)) throw new Error(`task ${i + 1} must be an object`);
+		const o = t;
+		const title = typeof o.title === "string" ? o.title.trim() : "";
+		const brief = typeof o.brief === "string" ? o.brief.trim() : "";
+		if (!title) throw new Error(`task ${i + 1} needs a non-empty "title"`);
+		if (!brief) throw new Error(`task ${i + 1} needs a non-empty "brief"`);
+		const strArr = (v) => Array.isArray(v) && v.every((x) => typeof x === "string") ? v : void 0;
+		const files = strArr(o.files);
+		const acceptance = strArr(o.acceptance);
+		tasks.push({
+			title,
+			brief,
+			...typeof o.class === "string" && o.class.trim() ? { class: o.class.trim() } : {},
+			...files?.length ? { files } : {},
+			...acceptance?.length ? { acceptance } : {}
+		});
+	}
+	return tasks;
+}
+/** The planner worker's instructions, goal and prompt rules included. */
+function plannerPrompt(goal, planFile, maxChildren) {
+	return [
+		"You are a PLANNER worker. Turn the goal below into a plan file and stop —",
+		"you do not implement anything, and you do not spawn workers yourself:",
+		"the runner executes your plan on your behalf (the sub-tasks run as your",
+		"children and report back to you).",
+		"",
+		`Write the plan with the Write tool to ${planFile} as one JSON object:`,
+		"{\"tasks\":[{\"title\":\"…\",\"brief\":\"…\",\"class\":\"implement\",\"files\":[\"…\"],\"acceptance\":[\"…\"]}]}",
+		"",
+		"Prompt rules for the plan itself (they are yours too):",
+		"- Domain-specific instructions only: name real files, commands and",
+		"  constraints of this repository, never generic advice.",
+		"- Constraints over step lists: state what must hold (and what must not",
+		"  change), not a numbered procedure to follow.",
+		"- Explicit quantity ranges: how many, how much, how long — never",
+		"  \"several\", \"some\", \"as needed\".",
+		"- No checkbox style: no `[ ]` items, no step numbering theatre.",
+		"",
+		`Emit between ${MIN_TASKS} and ${MAX_TASKS} sub-tasks — unless the goal itself`,
+		"names a smaller explicit count (\"one sub-task per file\" with three files",
+		"means three); an explicit count in the goal always wins.",
+		`At most ${maxChildren} of them run at a time, so independent tasks are better`,
+		"than long chains. Each task's class picks its provider: implement, draft,",
+		"review, research, spotcheck, simple, complex, image.",
+		"",
+		"Read the repository first (Glob/Grep/Read) so tasks name real files.",
+		"Every sub-task runs in its own worktree of the same repository: name files",
+		"relative to the repository root and never an absolute path or a worktree",
+		"directory in a title, brief, files or acceptance entry.",
+		"",
+		"## Goal",
+		"",
+		goal
+	].join("\n");
+}
+/** The child's prompt for one planned sub-task. */
+function taskPrompt(goal, task, index, of) {
+	return [
+		`You are sub-task ${index + 1} of ${of} of a planned goal. Do exactly this sub-task;`,
+		"the other sub-tasks are other workers' business.",
+		"",
+		`# ${task.title}`,
+		"",
+		task.brief,
+		...task.files?.length ? ["", "Files likely touched: " + task.files.join(", ")] : [],
+		...task.acceptance?.length ? [
+			"",
+			"Done means:",
+			...task.acceptance.map((a) => `- ${a}`)
+		] : [],
+		"",
+		"## The overall goal (context only — your scope is the sub-task above)",
+		"",
+		goal
+	].join("\n");
+}
+/**
+* The planner orchestration; returns the process exit code. The planner id is
+* minted here and preset on the phase-1 run, so the plan file's name is known
+* before the worker starts and the prompt can name its exact path.
+*/
+async function runPlanner(opts, deps = {}) {
+	const runStage = deps.runStage ?? runWorker;
+	const { workers: config } = readWorkersSection();
+	const logDir = deps.logDir ?? workersLogDir(config);
+	const maxChildren = deps.maxChildren ?? config.tree.maxChildren;
+	const parsed = parseRunnerArgs(opts.claudeArgs);
+	const goal = parsed.prompt ?? "";
+	const plannerId = newWorkerId();
+	const planFile = planPathFor(logDir, plannerId);
+	mkdirSync(join(logDir, "plans"), { recursive: true });
+	process.stderr.write(`planner ${plannerId}: writing plan (${planFile})\n`);
+	const rc1 = await runStage({
+		...opts,
+		id: plannerId,
+		claudeArgs: swapPromptArg(opts.claudeArgs, plannerPrompt(goal, planFile, maxChildren)),
+		quiet: true,
+		_planner: true
+	});
+	if (rc1 !== 0) return rc1;
+	if (!existsSync(planFile)) {
+		process.stderr.write(`planner ${plannerId}: no plan file at ${planFile} — the planner worker did not write one. Re-run, or write the plan yourself and run the tasks with pai worker run.\n`);
+		appendLedger(ledgerPath(logDir), "WORKER-PLAN-END", {
+			planner: plannerId,
+			rc: 1,
+			failed: "plan"
+		});
+		return 1;
+	}
+	let tasks;
+	try {
+		tasks = validatePlan(JSON.parse(readFileSync(planFile, "utf8")));
+	} catch (e) {
+		process.stderr.write(`planner ${plannerId}: invalid plan file ${planFile}: ${e.message}\n`);
+		appendLedger(ledgerPath(logDir), "WORKER-PLAN-END", {
+			planner: plannerId,
+			rc: 1,
+			failed: "plan"
+		});
+		return 1;
+	}
+	appendLedger(ledgerPath(logDir), "WORKER-PLAN", {
+		planner: plannerId,
+		tasks: tasks.length,
+		plan: planFile
+	});
+	const results = [];
+	for (let i = 0; i < tasks.length; i += maxChildren) {
+		const wave = tasks.slice(i, i + maxChildren);
+		process.stderr.write(`planner ${plannerId}: sub-tasks ${i + 1}–${i + wave.length} of ${tasks.length}\n`);
+		const rcs = await Promise.all(wave.map((task, w) => runStage({
+			className: task.class ?? "implement",
+			providerFlag: opts.providerFlag,
+			modelFlag: opts.modelFlag,
+			label: shortText(task.title, 40),
+			noPane: opts.noPane,
+			mcpFlag: opts.mcpFlag,
+			claudeArgs: swapPromptArg(opts.claudeArgs, taskPrompt(goal, task, i + w, tasks.length)),
+			cwd: opts.cwd,
+			worktreeFlag: opts.worktreeFlag,
+			parent: plannerId,
+			quiet: true
+		})));
+		for (let w = 0; w < wave.length; w++) results.push({
+			task: wave[w],
+			rc: rcs[w]
+		});
+	}
+	const failed = results.filter((r) => r.rc !== 0).length;
+	appendLedger(ledgerPath(logDir), "WORKER-PLAN-END", {
+		planner: plannerId,
+		rc: failed ? 1 : 0,
+		failed
+	});
+	printPlannerSummary(logDir, plannerId, results, parsed.outputFormat, opts.quiet === true);
+	return failed ? 1 : 0;
+}
+/** Compose and print the planner's summary report (text or json). */
+function printPlannerSummary(logDir, plannerId, results, fmt, quiet) {
+	const children = statusesOfChildren(logDir, plannerId);
+	const ok = results.filter((r) => r.rc === 0).length;
+	const unmerged = children.filter((c) => c.branch && !c.merged);
+	const report = {
+		checks: results.map((r, i) => ({
+			name: r.task.title,
+			ok: r.rc === 0,
+			detail: r.rc === 0 ? shortText(children[i]?.last, 80) : `rc=${r.rc}`
+		})),
+		...unmerged.length ? { open: unmerged.map((c) => `branch to merge: pai worker merge ${c.id} (${c.branch})`) } : {},
+		notes: `${ok}/${results.length} sub-tasks ok${unmerged.length ? `; ${unmerged.length} branch(es) to merge` : ""}`
+	};
+	if (quiet || fmt === "stream-json") return;
+	printResult(fmt, {
+		type: "result",
+		result: [
+			`planner ${plannerId}: ${report.notes}`,
+			...unmerged.length ? ["branches to merge:"] : [],
+			...unmerged.map((c) => `  pai worker merge ${c.id}   # ${c.branch}`)
+		].join("\n"),
+		is_error: ok !== results.length
+	}, ok === results.length ? 0 : 1, logDir, plannerId, report, {
+		plan: results.length,
+		...unmerged.length ? { branches: unmerged.map((c) => c.branch) } : {}
+	});
+}
+/** Children of the planner, oldest first, from the status files. */
+function statusesOfChildren(logDir, plannerId) {
+	return loadStatuses(logDir).filter((s) => s.parent === plannerId);
+}
+
+//#endregion
+export { runPlanner };
+//# sourceMappingURL=planner-BDI7bE9B.mjs.map