@tekmidian/pai 0.54.1 → 0.55.0

package/dist/kg-entity-Cj8Q6YmU.mjs

@@ -0,0 +1,29 @@
+import { createHash } from "node:crypto";
+
+//#region src/memory/kg-entity.ts
+/**
+* kg-entity.ts — Entity content-addressing with multi-tenant support.
+*
+* Provides UUID5-style deterministic content hashes for KG entities and edges,
+* ensuring that the same entity name always maps to the same ID within a tenant.
+* This enables idempotent upserts and stable foreign keys for kg_triples.
+*
+* Multi-tenant support: each tenant namespace gets its own entity ID space.
+* The default tenant is "default" for single-user deployments.
+*/
+/**
+* Generate a deterministic entity ID (UUID5-style) for a given name and tenant.
+*
+* The ID is a hex digest derived from "tenant_id:name" so the same entity
+* always receives the same ID within a tenant namespace.
+*
+* @param name      Entity name (case-preserved)
+* @param tenantId  Tenant namespace (default: "default")
+*/
+function entityContentId(name, tenantId = "default") {
+	return createHash("sha256").update(`entity:${tenantId}:${name}`).digest("hex").slice(0, 32);
+}
+
+//#endregion
+export { entityContentId as t };
+//# sourceMappingURL=kg-entity-Cj8Q6YmU.mjs.map

package/dist/kg-entity-Cj8Q6YmU.mjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"kg-entity-Cj8Q6YmU.mjs","names":[],"sources":["../src/memory/kg-entity.ts"],"sourcesContent":["/**\n * kg-entity.ts — Entity content-addressing with multi-tenant support.\n *\n * Provides UUID5-style deterministic content hashes for KG entities and edges,\n * ensuring that the same entity name always maps to the same ID within a tenant.\n * This enables idempotent upserts and stable foreign keys for kg_triples.\n *\n * Multi-tenant support: each tenant namespace gets its own entity ID space.\n * The default tenant is \"default\" for single-user deployments.\n */\n\nimport { createHash } from \"node:crypto\";\n\n// ---------------------------------------------------------------------------\n// Types\n// ---------------------------------------------------------------------------\n\nexport interface KgEntity {\n  entity_id: string;\n  tenant_id: string;\n  name: string;\n  type: string;\n  description?: string;\n  first_seen?: number;\n  last_seen?: number;\n  mention_count: number;\n  feedback_weight: number;\n}\n\nexport interface KgEntityUpsertParams {\n  name: string;\n  type?: string;\n  description?: string;\n  tenantId?: string;\n}\n\n// ---------------------------------------------------------------------------\n// Content addressing\n// ---------------------------------------------------------------------------\n\n/**\n * Generate a deterministic entity ID (UUID5-style) for a given name and tenant.\n *\n * The ID is a hex digest derived from \"tenant_id:name\" so the same entity\n * always receives the same ID within a tenant namespace.\n *\n * @param name      Entity name (case-preserved)\n * @param tenantId  Tenant namespace (default: \"default\")\n */\nexport function entityContentId(name: string, tenantId = \"default\"): string {\n  return createHash(\"sha256\")\n    .update(`entity:${tenantId}:${name}`)\n    .digest(\"hex\")\n    .slice(0, 32); // 128-bit hex string — UUID5-compatible length\n}\n\n/**\n * Generate a deterministic edge ID for a (source, relation, target) triple\n * within a tenant namespace.\n *\n * @param source    Source entity name\n * @param relation  Relation/predicate verb phrase\n * @param target    Target entity name\n * @param tenantId  Tenant namespace (default: \"default\")\n */\nexport function edgeContentId(\n  source: string,\n  relation: string,\n  target: string,\n  tenantId = \"default\"\n): string {\n  return createHash(\"sha256\")\n    .update(`edge:${tenantId}:${source}:${relation}:${target}`)\n    .digest(\"hex\")\n    .slice(0, 32);\n}\n\n// SQLite's kg_entities CRUD (upsertKgEntity/findKgEntity/listKgEntities/\n// updateEntityFeedbackWeight) lives in storage/sqlite/kg-entity.ts —\n// PostgresBackend implements the same operations inline against its own\n// pool, so there is no cross-backend function to share here.\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AAiDA,SAAgB,gBAAgB,MAAc,WAAW,WAAmB;AAC1E,QAAO,WAAW,SAAS,CACxB,OAAO,UAAU,SAAS,GAAG,OAAO,CACpC,OAAO,MAAM,CACb,MAAM,GAAG,GAAG"}

package/dist/{latent-ideas-BfL4nCQ3.mjs → latent-ideas-BRy6HNwg.mjs}

@@ -1,6 +1,6 @@
-import "./embeddings-DvRKcQyt.mjs";
-import { n as TITLE_STOP_WORDS } from "./stop-words-BaMEGVeY.mjs";
-import { t as zettelThemes } from "./themes-BVoAnLZj.mjs";
+import { n as TITLE_STOP_WORDS } from "./stop-words-ClCwO8mZ.mjs";
+import "./embeddings-DfrYNIng.mjs";
+import { t as zettelThemes } from "./themes-B2v7RNnA.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-BfL4nCQ3.mjs.map
+//# sourceMappingURL=latent-ideas-BRy6HNwg.mjs.map

package/dist/{latent-ideas-BfL4nCQ3.mjs.map → latent-ideas-BRy6HNwg.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"latent-ideas-BfL4nCQ3.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-BRy6HNwg.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-C5PaNdzm.mjs → link-boost-BptbAPuv.mjs}

@@ -31,4 +31,4 @@ function applyLinkBoost(results, edges, opts) {
 
 //#endregion
 export { applyLinkBoost };
-//# sourceMappingURL=link-boost-C5PaNdzm.mjs.map
+//# sourceMappingURL=link-boost-BptbAPuv.mjs.map

package/dist/{link-boost-C5PaNdzm.mjs.map → link-boost-BptbAPuv.mjs.map}

@@ -1 +1 @@
-{"version":3,"file":"link-boost-C5PaNdzm.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-BptbAPuv.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/{main-resolver-Bt1f6jJc.mjs → main-resolver-Dx-pxic1.mjs}

@@ -1,12 +1,12 @@
-import { a as paiHomePath, o as resolvePaiFile, r as migratePaiFile } from "./pai-home-ChAOupZc.mjs";
-import { _ as warn, c as ok, g as smartDecodeDir, i as err, n as dim, o as header, u as renderTable } from "./utils-CJymHkYm.mjs";
-import { C as readWorkersSection } from "./config-B3EPzRjf.mjs";
-import { a as switchToSession, n as fetchLiveSessions } from "./aibroker-client-DzkTpVjx.mjs";
+import { _ as warn, c as ok, g as smartDecodeDir, i as err, n as dim, o as header, u as renderTable } from "./utils-C9HsYDpO.mjs";
+import { a as paiHomePath, o as resolvePaiFile, r as migratePaiFile } from "./pai-home-CeOIYltV.mjs";
+import { C as readWorkersSection } from "./config-B93YS8tP.mjs";
+import { a as switchToSession, n as fetchLiveSessions } from "./aibroker-client-BK1h--W_.mjs";
 import { closeSync, copyFileSync, createReadStream, existsSync, linkSync, mkdirSync, openSync, readFileSync, readSync, readdirSync, realpathSync, renameSync, statSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";
 import { basename, join } from "node:path";
-import { spawnSync } from "node:child_process";
 import chalk from "chalk";
+import { spawnSync } from "node:child_process";
 import { createInterface } from "node:readline";
 
 //#region src/cli/lib/session-scan.ts
@@ -65,12 +65,12 @@ function buildClcInfoMap() {
 	return map;
 }
 /** Build encoded_dir → root_path map from PAI registry (authoritative cwd). */
-function buildRegistryRootPathMap(db) {
+async function buildRegistryRootPathMap() {
 	try {
-		const rows = db.prepare(`SELECT root_path, encoded_dir FROM projects
-         WHERE encoded_dir IS NOT NULL AND encoded_dir != ''`).all();
+		const { getRegistryBackend } = await import("./factory-C9wGUe30.mjs");
+		const projects = await (await getRegistryBackend()).listProjects();
 		const map = /* @__PURE__ */ new Map();
-		for (const row of rows) map.set(row.encoded_dir, row.root_path);
+		for (const p of projects) if (p.encoded_dir) map.set(p.encoded_dir, p.root_path);
 		return map;
 	} catch {
 		return /* @__PURE__ */ new Map();
@@ -297,12 +297,12 @@ function resolveFilter(opts) {
 *
 * Returns results sorted by mtime descending.
 */
-function scanSessions(db, opts = {}) {
+async function scanSessions(opts = {}) {
 	const limit = opts.limit ?? 200;
 	const filterMode = resolveFilter(opts);
 	if (!existsSync(CLAUDE_PROJECTS_DIR)) return [];
 	const clcInfoMap = buildClcInfoMap();
-	const rootPathMap = buildRegistryRootPathMap(db);
+	const rootPathMap = await buildRegistryRootPathMap();
 	const results = [];
 	const seenUuids = /* @__PURE__ */ new Set();
 	const attachedClcUuids = /* @__PURE__ */ new Set();
@@ -1119,6 +1119,27 @@ async function searchHistory(query, maxResults) {
 
 //#endregion
 //#region src/cli/lib/dedup-sessions.ts
+/**
+* dedup-sessions.ts — Shared dedup catalog for `pai` and listing commands.
+*
+* Both `pai` (no args) and any listing command must show IDENTICAL output.
+* This module is the single source of truth for:
+*   - Name normalization  (strips spinner prefixes and " (node)" suffixes)
+*   - Dedup grouping      (one row per normalized name)
+*   - Priority ranking    (live > resumable > transcript-only > stub > project > orphan)
+*   - Registered projects (shown as status "idle" when no live/disk session matches)
+*   - Filtering           (hide 0-session / cold / archived projects by default)
+*   - Table rendering     (shared columns and formatting)
+*
+* Default visibility rules (overridden by showAll=true):
+*   SHOW: live sessions (always)
+*   SHOW: resumable sessions (always)
+*   SHOW: registered projects with session_count > 0 AND last_active within 90 days
+*         OR session_count >= 3 (i.e. well-used projects regardless of age)
+*   HIDE: projects with session_count = 0 (never used)
+*   HIDE: projects with last_active > 90 days AND session_count < 3 (cold)
+*   HIDE: archived projects (status != 'active')
+*/
 const STATUS_PRIORITY = {
 	live: 0,
 	resumable: 1,
@@ -1349,6 +1370,23 @@ function renderDedupedSessions(entries, maxRows) {
 //#endregion
 //#region src/cli/commands/main-resolver.ts
 /**
+* main-resolver.ts
+*
+* `pai [<query>]`  —  topic-first session discovery and launcher (v0.11.1)
+*
+* Decision tree:
+*   1. No arg          → show deduped session listing (one row per name)
+*   2. UUID prefix     → universal filesystem scan; auto-launch the match
+*   3. Any string:
+*      a. Live match (by normalized paiName) → aibroker_switch → iTerm tab to front. Done.
+*      b. Any disk/project match             → fresh launch in the project dir, on the
+*                                              configured route (worker or claude)
+*      c. --resume                           → probe + claude --resume <newest uuid>
+*      d. No name match                      → free-text history search → picker
+*
+* Dedup + name normalization logic: src/cli/lib/dedup-sessions.ts (shared with listing).
+*/
+/**
 * Which directory should we open for this session?
 *
 * A session carries up to three ideas of where it lives, they disagree in
@@ -1552,36 +1590,23 @@ async function openMatch(entry, allSessions, dryRun, resume, auto) {
 	}
 	return false;
 }
-function getRegisteredProjects(db, all = false) {
+async function getRegisteredProjects(all = false) {
 	try {
-		const statusClause = all ? "" : "WHERE p.status = 'active'";
-		return db.prepare(`
-        SELECT
-          p.slug,
-          p.display_name,
-          p.root_path,
-          p.status,
-          COUNT(s.id) AS session_count,
-          MAX(s.created_at) AS last_active
-        FROM projects p
-        LEFT JOIN sessions s ON s.project_id = p.id
-        ${statusClause}
-        GROUP BY p.id
-        ORDER BY last_active DESC NULLS LAST, p.updated_at DESC
-      `).all();
+		const { getRegistryBackend } = await import("./factory-C9wGUe30.mjs");
+		return await (await getRegistryBackend()).listProjectsWithSessionStats(all ? {} : { status: "active" });
 	} catch {
 		return [];
 	}
 }
-async function cmdMain(db, query, pickN, opts) {
+async function cmdMain(query, pickN, opts) {
 	const maxResults = parseInt(opts.n ?? "20", 10);
 	const showAll = opts.all ?? false;
 	const livePromise = !query ? fetchLiveSessions().catch(() => []) : Promise.resolve([]);
-	const allSessions = scanSessions(db, {
+	const allSessions = await scanSessions({
 		limit: 500,
 		filter: "named"
 	});
-	const registeredProjects = getRegisteredProjects(db, showAll);
+	const registeredProjects = await getRegisteredProjects(showAll);
 	if (!query) {
 		renderDedupedSessions(buildDeduped(await livePromise, allSessions, registeredProjects, showAll), showAll ? void 0 : maxResults);
 		return;
@@ -1750,4 +1775,4 @@ async function cmdMain(db, query, pickN, opts) {
 
 //#endregion
 export { scanSessions as _, buildDeduped as a, hasConversation as c, restoreTopLevel as d, printExitDir as f, scanCacheFilePath as g, resolveSessionByNameOrId as h, resumeTargetFor as i, launchInDir as l, migrateSessionScanCache as m, resolveSessionDir as n, normalizeName as o, fmtAge as p, resumeCandidateFor as r, renderDedupedSessions as s, cmdMain as t, probeResume as u };
-//# sourceMappingURL=main-resolver-Bt1f6jJc.mjs.map
+//# sourceMappingURL=main-resolver-Dx-pxic1.mjs.map