sensemaking 0.5.0 → 0.7.0

package/dist/esm/verbs.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/verbs.ts"],"sourcesContent":["import posix from 'node:path/posix';\nimport type { DatabaseSync } from 'node:sqlite';\nimport type { Config } from './config.ts';\nimport { featureEnabled } from './config.ts';\nimport { SenseError } from './errors.ts';\nimport { linkEdges } from './features/index.ts';\nimport { personalizedRank } from './graph.ts';\nimport type { Row } from './output.ts';\n\n// The three layer verbs: mapTree (orient), find (locate), peek (structure).\n// Each returns data; cli.ts renders. All of them degrade when a feature is off.\n\nconst WEIGHTED_BM25 = 'bm25(content, 10.0, 5.0, 1.0)';\nconst RRF_K = 60;\n\nexport interface FindOptions {\n  k?: number;\n  where?: string; // SQL fragment against frontmatter alias `f`, e.g. \"f.status = 'active'\"\n}\n\n// Layer 1: BM25 + link-graph expansion, fused by reciprocal rank. `via` says which\n// signal produced each row so the agent knows what evidence it is trusting.\nexport function find(db: DatabaseSync, cfg: Config, terms: string, opts: FindOptions = {}): Row[] {\n  const k = opts.k ?? 10;\n  const fetch = Math.max(k * 3, 30);\n\n  // Terms pass verbatim to FTS5 MATCH: bare words AND-join, operators are the caller's.\n  // Invalid syntax propagates as an error, zero matches return zero -- no silent rewrites.\n  // --where applies inside the candidate query (a post-filter over the top-N would drop\n  // matches ranked past the pool) and again on the final select for link-derived rows.\n  const whereJoin = opts.where ? `JOIN frontmatter f ON f.\"path\" = content.path` : '';\n  const whereCond = opts.where ? `AND (${opts.where})` : '';\n  const matchSql = `SELECT content.path AS path, snippet(content, -1, '«', '»', '…', 10) AS hit FROM content ${whereJoin} WHERE content MATCH ? ${whereCond} ORDER BY ${WEIGHTED_BM25} LIMIT ${fetch}`;\n  const matchRows = db.prepare(matchSql).all(terms) as Array<{ path: string; hit: string }>;\n\n  const hits = new Map(matchRows.map((r) => [r.path, r.hit]));\n  const candidates = new Map<string, { score: number; via: string }>();\n  matchRows.forEach((r, i) => {\n    candidates.set(r.path, { score: 1 / (RRF_K + i), via: 'match' });\n  });\n\n  if (featureEnabled(cfg, 'links') && matchRows.length > 0) {\n    const nodes = (db.prepare('SELECT \"path\" FROM frontmatter').all() as Array<{ path: string }>).map((r) => r.path);\n    const seeds = new Map(matchRows.map((r, i) => [r.path, 1 / (i + 1)]));\n    const ranked = [...personalizedRank(nodes, linkEdges(db), seeds)]\n      .filter(([, score]) => score > 1e-9)\n      .sort((a, b) => b[1] - a[1])\n      .slice(0, fetch);\n    ranked.forEach(([path], i) => {\n      const existing = candidates.get(path);\n      if (existing) {\n        existing.score += 1 / (RRF_K + i);\n        existing.via = 'match+link';\n      } else {\n        candidates.set(path, { score: 1 / (RRF_K + i), via: 'link' });\n      }\n    });\n  }\n\n  db.exec('CREATE TEMP TABLE IF NOT EXISTS _find (\"path\" TEXT PRIMARY KEY, score REAL, via TEXT, hit TEXT)');\n  db.exec('DELETE FROM _find');\n  const insert = db.prepare('INSERT INTO _find (\"path\", score, via, hit) VALUES (?, ?, ?, ?)');\n  for (const [path, c] of candidates) insert.run(path, c.score, c.via, hits.get(path) ?? null);\n\n  const where = opts.where ? `WHERE ${opts.where}` : '';\n  return db\n    .prepare(\n      `SELECT f.\"path\" AS path, content.title, content.summary, _find.hit, _find.via, round(_find.score, 4) AS score\n       FROM _find JOIN frontmatter f ON f.\"path\" = _find.\"path\" JOIN content ON content.path = _find.\"path\"\n       ${where} ORDER BY _find.score DESC LIMIT ?`\n    )\n    .all(k) as Row[];\n}\n\nexport interface TreeMap {\n  docs: { count: number; bytes: number };\n  fields: Row[]; // top 20 by coverage; fieldsTotal carries the real count\n  fieldsTotal: number;\n  hubs: Row[];\n  recent: Row[];\n}\n\nconst INTERNAL_COLUMNS = new Set(['path', '_mtime', '_size', '_rank']);\n\n// Layer 0: what is this tree. Fixed-size output regardless of tree size.\nexport function mapTree(db: DatabaseSync, cfg: Config): TreeMap {\n  const docs = db.prepare('SELECT COUNT(*) AS count, COALESCE(SUM(\"_size\"), 0) AS bytes FROM frontmatter').get() as { count: number; bytes: number };\n\n  const columns = (db.prepare('PRAGMA table_info(frontmatter)').all() as Array<{ name: string }>).map((r) => r.name).filter((name) => !INTERNAL_COLUMNS.has(name));\n  const allFields = columns\n    .map((name) => {\n      const { n } = db.prepare(`SELECT COUNT(\"${name.split('\"').join('\"\"')}\") AS n FROM frontmatter`).get() as { n: number };\n      return { field: name, coverage: n };\n    })\n    .sort((a, b) => (b.coverage as number) - (a.coverage as number)) as Row[];\n  const fields = allFields.slice(0, 20);\n\n  const hubs = featureEnabled(cfg, 'rank') ? (db.prepare(`SELECT f.\"path\" AS path, round(f.\"_rank\" * 100, 2) AS rank, content.title FROM frontmatter f JOIN content ON content.path = f.\"path\" WHERE f.\"_rank\" IS NOT NULL ORDER BY f.\"_rank\" DESC LIMIT 8`).all() as Row[]) : [];\n\n  const recent = db.prepare(`SELECT \"path\", datetime(\"_mtime\" / 1000, 'unixepoch') AS modified FROM frontmatter ORDER BY \"_mtime\" DESC LIMIT 5`).all() as Row[];\n\n  return { docs, fields, fieldsTotal: allFields.length, hubs, recent };\n}\n\nexport interface Peek {\n  path: string;\n  tokens: number;\n  frontmatter: Row;\n  sections: Row[];\n  outbound: string[];\n  backlinks: string[];\n  unresolved: string[];\n  // Totals before truncation: a hub can have thousands of backlinks, and peek's whole\n  // point is bounded output. Query the links table directly for the full list.\n  outboundTotal: number;\n  backlinksTotal: number;\n  unresolvedTotal: number;\n}\n\nconst PEEK_LINK_LIMIT = 20;\n\n// Layer 2: everything about one note except its prose -- frontmatter, outline with line\n// ranges + token estimates (so the follow-up Read is a range, not the file), links both ways.\nexport function peek(db: DatabaseSync, cfg: Config, pathArg: string): Peek {\n  const paths = (db.prepare('SELECT \"path\" FROM frontmatter').all() as Array<{ path: string }>).map((r) => r.path);\n  let path = paths.find((p) => p === pathArg);\n  if (!path) {\n    const base = posix.basename(pathArg).replace(/\\.md$/i, '').toLowerCase();\n    const matches = paths.filter((p) => posix.basename(p).replace(/\\.md$/i, '').toLowerCase() === base);\n    if (matches.length === 1) path = matches[0];\n    else if (matches.length > 1) throw new SenseError('NOTE_AMBIGUOUS', `\"${pathArg}\" is ambiguous: ${matches.join(', ')}`);\n    else throw new SenseError('NOTE_NOT_FOUND', `no note matches \"${pathArg}\"`);\n  }\n\n  const row = db.prepare('SELECT * FROM frontmatter WHERE \"path\" = ?').get(path) as Row;\n  const frontmatter: Row = {};\n  for (const [key, value] of Object.entries(row)) {\n    if (!INTERNAL_COLUMNS.has(key) && value !== null) frontmatter[key] = value;\n  }\n\n  const sections = featureEnabled(cfg, 'sections') ? (db.prepare('SELECT level, heading, start_line, end_line, tokens FROM sections WHERE \"path\" = ? ORDER BY idx').all(path) as Row[]) : [];\n\n  let outbound: string[] = [];\n  let backlinks: string[] = [];\n  let unresolved: string[] = [];\n  let backlinksTotal = 0;\n  if (featureEnabled(cfg, 'links')) {\n    const out = db.prepare('SELECT target, dst FROM links WHERE src = ? ORDER BY target').all(path) as Array<{ target: string; dst: string | null }>;\n    outbound = [...new Set(out.filter((l) => l.dst !== null).map((l) => l.dst as string))];\n    unresolved = out.filter((l) => l.dst === null).map((l) => l.target);\n    backlinksTotal = (db.prepare('SELECT COUNT(DISTINCT src) AS n FROM links WHERE dst = ?').get(path) as { n: number }).n;\n    backlinks = (db.prepare('SELECT DISTINCT src FROM links WHERE dst = ? ORDER BY src LIMIT ?').all(path, PEEK_LINK_LIMIT) as Array<{ src: string }>).map((r) => r.src);\n  }\n\n  return {\n    path,\n    tokens: Math.ceil(((row._size as number) ?? 0) / 4),\n    frontmatter,\n    sections,\n    outbound: outbound.slice(0, PEEK_LINK_LIMIT),\n    backlinks,\n    unresolved: unresolved.slice(0, PEEK_LINK_LIMIT),\n    outboundTotal: outbound.length,\n    backlinksTotal,\n    unresolvedTotal: unresolved.length,\n  };\n}\n"],"names":["posix","featureEnabled","SenseError","linkEdges","personalizedRank","WEIGHTED_BM25","RRF_K","find","db","cfg","terms","opts","hits","k","fetch","Math","max","whereJoin","where","whereCond","matchSql","matchRows","prepare","all","Map","map","r","path","hit","candidates","forEach","i","set","score","via","length","nodes","seeds","ranked","filter","sort","a","b","slice","existing","get","exec","insert","c","run","INTERNAL_COLUMNS","Set","mapTree","docs","columns","name","has","allFields","n","split","join","field","coverage","fields","hubs","recent","fieldsTotal","PEEK_LINK_LIMIT","peek","pathArg","row","paths","p","base","basename","replace","toLowerCase","matches","frontmatter","key","value","Object","entries","sections","outbound","backlinks","unresolved","backlinksTotal","out","l","dst","target","src","tokens","ceil","_size","outboundTotal","unresolvedTotal"],"mappings":"AAAA,OAAOA,WAAW,kBAAkB;AAGpC,SAASC,cAAc,QAAQ,cAAc;AAC7C,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,SAAS,QAAQ,sBAAsB;AAChD,SAASC,gBAAgB,QAAQ,aAAa;AAG9C,4EAA4E;AAC5E,gFAAgF;AAEhF,MAAMC,gBAAgB;AACtB,MAAMC,QAAQ;AAOd,mFAAmF;AACnF,4EAA4E;AAC5E,OAAO,SAASC,KAAKC,EAAgB,EAAEC,GAAW,EAAEC,KAAa,EAAEC,OAAoB,CAAC,CAAC;QAC7EA,SAuC2DC;IAvCrE,MAAMC,KAAIF,UAAAA,KAAKE,CAAC,cAANF,qBAAAA,UAAU;IACpB,MAAMG,QAAQC,KAAKC,GAAG,CAACH,IAAI,GAAG;IAE9B,sFAAsF;IACtF,yFAAyF;IACzF,sFAAsF;IACtF,qFAAqF;IACrF,MAAMI,YAAYN,KAAKO,KAAK,GAAG,CAAC,6CAA6C,CAAC,GAAG;IACjF,MAAMC,YAAYR,KAAKO,KAAK,GAAG,CAAC,KAAK,EAAEP,KAAKO,KAAK,CAAC,CAAC,CAAC,GAAG;IACvD,MAAME,WAAW,CAAC,yFAAyF,EAAEH,UAAU,uBAAuB,EAAEE,UAAU,UAAU,EAAEd,cAAc,OAAO,EAAES,OAAO;IACpM,MAAMO,YAAYb,GAAGc,OAAO,CAACF,UAAUG,GAAG,CAACb;IAE3C,MAAME,OAAO,IAAIY,IAAIH,UAAUI,GAAG,CAAC,CAACC,IAAM;YAACA,EAAEC,IAAI;YAAED,EAAEE,GAAG;SAAC;IACzD,MAAMC,aAAa,IAAIL;IACvBH,UAAUS,OAAO,CAAC,CAACJ,GAAGK;QACpBF,WAAWG,GAAG,CAACN,EAAEC,IAAI,EAAE;YAAEM,OAAO,IAAK3B,CAAAA,QAAQyB,CAAAA;YAAIG,KAAK;QAAQ;IAChE;IAEA,IAAIjC,eAAeQ,KAAK,YAAYY,UAAUc,MAAM,GAAG,GAAG;QACxD,MAAMC,QAAQ,AAAC5B,GAAGc,OAAO,CAAC,kCAAkCC,GAAG,GAA+BE,GAAG,CAAC,CAACC,IAAMA,EAAEC,IAAI;QAC/G,MAAMU,QAAQ,IAAIb,IAAIH,UAAUI,GAAG,CAAC,CAACC,GAAGK,IAAM;gBAACL,EAAEC,IAAI;gBAAE,IAAKI,CAAAA,IAAI,CAAA;aAAG;QACnE,MAAMO,SAAS;eAAIlC,iBAAiBgC,OAAOjC,UAAUK,KAAK6B;SAAO,CAC9DE,MAAM,CAAC,CAAC,GAAGN,MAAM,GAAKA,QAAQ,MAC9BO,IAAI,CAAC,CAACC,GAAGC,IAAMA,CAAC,CAAC,EAAE,GAAGD,CAAC,CAAC,EAAE,EAC1BE,KAAK,CAAC,GAAG7B;QACZwB,OAAOR,OAAO,CAAC,CAAC,CAACH,KAAK,EAAEI;YACtB,MAAMa,WAAWf,WAAWgB,GAAG,CAAClB;YAChC,IAAIiB,UAAU;gBACZA,SAASX,KAAK,IAAI,IAAK3B,CAAAA,QAAQyB,CAAAA;gBAC/Ba,SAASV,GAAG,GAAG;YACjB,OAAO;gBACLL,WAAWG,GAAG,CAACL,MAAM;oBAAEM,OAAO,IAAK3B,CAAAA,QAAQyB,CAAAA;oBAAIG,KAAK;gBAAO;YAC7D;QACF;IACF;IAEA1B,GAAGsC,IAAI,CAAC;IACRtC,GAAGsC,IAAI,CAAC;IACR,MAAMC,SAASvC,GAAGc,OAAO,CAAC;IAC1B,KAAK,MAAM,CAACK,MAAMqB,EAAE,IAAInB,WAAYkB,OAAOE,GAAG,CAACtB,MAAMqB,EAAEf,KAAK,EAAEe,EAAEd,GAAG,GAAEtB,YAAAA,KAAKiC,GAAG,CAAClB,mBAATf,uBAAAA,YAAkB;IAEvF,MAAMM,QAAQP,KAAKO,KAAK,GAAG,CAAC,MAAM,EAAEP,KAAKO,KAAK,EAAE,GAAG;IACnD,OAAOV,GACJc,OAAO,CACN,CAAC;;OAEA,EAAEJ,MAAM,kCAAkC,CAAC,EAE7CK,GAAG,CAACV;AACT;AAUA,MAAMqC,mBAAmB,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAS;CAAQ;AAErE,yEAAyE;AACzE,OAAO,SAASC,QAAQ5C,EAAgB,EAAEC,GAAW;IACnD,MAAM4C,OAAO7C,GAAGc,OAAO,CAAC,iFAAiFuB,GAAG;IAE5G,MAAMS,UAAU,AAAC9C,GAAGc,OAAO,CAAC,kCAAkCC,GAAG,GAA+BE,GAAG,CAAC,CAACC,IAAMA,EAAE6B,IAAI,EAAEhB,MAAM,CAAC,CAACgB,OAAS,CAACL,iBAAiBM,GAAG,CAACD;IAC1J,MAAME,YAAYH,QACf7B,GAAG,CAAC,CAAC8B;QACJ,MAAM,EAAEG,CAAC,EAAE,GAAGlD,GAAGc,OAAO,CAAC,CAAC,cAAc,EAAEiC,KAAKI,KAAK,CAAC,KAAKC,IAAI,CAAC,MAAM,wBAAwB,CAAC,EAAEf,GAAG;QACnG,OAAO;YAAEgB,OAAON;YAAMO,UAAUJ;QAAE;IACpC,GACClB,IAAI,CAAC,CAACC,GAAGC,IAAM,AAACA,EAAEoB,QAAQ,GAAerB,EAAEqB,QAAQ;IACtD,MAAMC,SAASN,UAAUd,KAAK,CAAC,GAAG;IAElC,MAAMqB,OAAO/D,eAAeQ,KAAK,UAAWD,GAAGc,OAAO,CAAC,CAAC,gMAAgM,CAAC,EAAEC,GAAG,KAAe,EAAE;IAE/Q,MAAM0C,SAASzD,GAAGc,OAAO,CAAC,CAAC,iHAAiH,CAAC,EAAEC,GAAG;IAElJ,OAAO;QAAE8B;QAAMU;QAAQG,aAAaT,UAAUtB,MAAM;QAAE6B;QAAMC;IAAO;AACrE;AAiBA,MAAME,kBAAkB;AAExB,wFAAwF;AACxF,8FAA8F;AAC9F,OAAO,SAASC,KAAK5D,EAAgB,EAAEC,GAAW,EAAE4D,OAAe;QAiC3CC;IAhCtB,MAAMC,QAAQ,AAAC/D,GAAGc,OAAO,CAAC,kCAAkCC,GAAG,GAA+BE,GAAG,CAAC,CAACC,IAAMA,EAAEC,IAAI;IAC/G,IAAIA,OAAO4C,MAAMhE,IAAI,CAAC,CAACiE,IAAMA,MAAMH;IACnC,IAAI,CAAC1C,MAAM;QACT,MAAM8C,OAAOzE,MAAM0E,QAAQ,CAACL,SAASM,OAAO,CAAC,UAAU,IAAIC,WAAW;QACtE,MAAMC,UAAUN,MAAMhC,MAAM,CAAC,CAACiC,IAAMxE,MAAM0E,QAAQ,CAACF,GAAGG,OAAO,CAAC,UAAU,IAAIC,WAAW,OAAOH;QAC9F,IAAII,QAAQ1C,MAAM,KAAK,GAAGR,OAAOkD,OAAO,CAAC,EAAE;aACtC,IAAIA,QAAQ1C,MAAM,GAAG,GAAG,MAAM,IAAIjC,WAAW,kBAAkB,CAAC,CAAC,EAAEmE,QAAQ,gBAAgB,EAAEQ,QAAQjB,IAAI,CAAC,OAAO;aACjH,MAAM,IAAI1D,WAAW,kBAAkB,CAAC,iBAAiB,EAAEmE,QAAQ,CAAC,CAAC;IAC5E;IAEA,MAAMC,MAAM9D,GAAGc,OAAO,CAAC,8CAA8CuB,GAAG,CAAClB;IACzE,MAAMmD,cAAmB,CAAC;IAC1B,KAAK,MAAM,CAACC,KAAKC,MAAM,IAAIC,OAAOC,OAAO,CAACZ,KAAM;QAC9C,IAAI,CAACpB,iBAAiBM,GAAG,CAACuB,QAAQC,UAAU,MAAMF,WAAW,CAACC,IAAI,GAAGC;IACvE;IAEA,MAAMG,WAAWlF,eAAeQ,KAAK,cAAeD,GAAGc,OAAO,CAAC,mGAAmGC,GAAG,CAACI,QAAkB,EAAE;IAE1L,IAAIyD,WAAqB,EAAE;IAC3B,IAAIC,YAAsB,EAAE;IAC5B,IAAIC,aAAuB,EAAE;IAC7B,IAAIC,iBAAiB;IACrB,IAAItF,eAAeQ,KAAK,UAAU;QAChC,MAAM+E,MAAMhF,GAAGc,OAAO,CAAC,+DAA+DC,GAAG,CAACI;QAC1FyD,WAAW;eAAI,IAAIjC,IAAIqC,IAAIjD,MAAM,CAAC,CAACkD,IAAMA,EAAEC,GAAG,KAAK,MAAMjE,GAAG,CAAC,CAACgE,IAAMA,EAAEC,GAAG;SAAa;QACtFJ,aAAaE,IAAIjD,MAAM,CAAC,CAACkD,IAAMA,EAAEC,GAAG,KAAK,MAAMjE,GAAG,CAAC,CAACgE,IAAMA,EAAEE,MAAM;QAClEJ,iBAAiB,AAAC/E,GAAGc,OAAO,CAAC,4DAA4DuB,GAAG,CAAClB,MAAwB+B,CAAC;QACtH2B,YAAY,AAAC7E,GAAGc,OAAO,CAAC,qEAAqEC,GAAG,CAACI,MAAMwC,iBAA4C1C,GAAG,CAAC,CAACC,IAAMA,EAAEkE,GAAG;IACrK;IAEA,OAAO;QACLjE;QACAkE,QAAQ9E,KAAK+E,IAAI,CAAC,EAAExB,aAAAA,IAAIyB,KAAK,cAATzB,wBAAAA,aAAwB,KAAK;QACjDQ;QACAK;QACAC,UAAUA,SAASzC,KAAK,CAAC,GAAGwB;QAC5BkB;QACAC,YAAYA,WAAW3C,KAAK,CAAC,GAAGwB;QAChC6B,eAAeZ,SAASjD,MAAM;QAC9BoD;QACAU,iBAAiBX,WAAWnD,MAAM;IACpC;AACF"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/verbs.ts"],"sourcesContent":["import posix from 'node:path/posix';\nimport type { DatabaseSync } from 'node:sqlite';\nimport type { Config, FeatureName } from './config.ts';\nimport { featureEnabled, featureStates } from './config.ts';\nimport { SenseError } from './errors.ts';\nimport { semanticCandidates } from './features/embed.ts';\nimport { linkEdges } from './features/index.ts';\nimport { personalizedRank } from './graph.ts';\nimport type { Row } from './output.ts';\n\n// The three layer verbs: mapTree (orient), find (locate), peek (structure).\n// Each returns data; cli.ts renders. All of them degrade when a feature is off.\n\nconst WEIGHTED_BM25 = 'bm25(content, 10.0, 5.0, 1.0)';\nconst RRF_K = 60;\n\n// FTS5 reads these as operators, so a bare term containing one is a syntax error or --\n// worse -- a column filter: `end-to-end` parses as a filter on column `to`, and SQLite\n// reports `no such column: to`, which is a true statement about the parse and a false\n// one about the input. Both field reports on 0.6.0 misdiagnosed that message.\nconst FTS5_OPERATORS = /[-'\"/.:^*()]/;\n\nfunction searchError(err: Error, terms: string, scope?: string): Error {\n  const message = err.message;\n  if (!/no such column|fts5: syntax error|malformed MATCH/.test(message)) return err;\n  const suspects = (terms.match(/\\S+/g) ?? []).filter((t) => !t.startsWith('\"') && FTS5_OPERATORS.test(t));\n  // Blame the terms only when the failing token actually came from one -- a typo'd column\n  // in --where (or the tree's default scope) raises \"no such column\" through this same\n  // statement, and naming a term for it would state a false fact about the input.\n  const col = /no such column: (\\S+)/.exec(message)?.[1];\n  const fromTerms = col === undefined ? suspects.length > 0 : suspects.some((t) => t.split(/[^\\p{L}\\p{N}]+/u).includes(col));\n  if (fromTerms && suspects.length > 0) {\n    return new SenseError('SEARCH_SYNTAX', `${message} -- the punctuation in ${suspects.map((t) => `\\`${t}\\``).join(', ')} is FTS5 syntax, not literal text; search for it literally by double-quoting: '\"${suspects[0]}\"'. Searchable columns are title, summary, text.`);\n  }\n  if (col !== undefined && scope !== undefined) {\n    return new SenseError('SEARCH_SYNTAX', `${message} -- the where condition (${scope}) references it; frontmatter columns are listed by sense query \"SELECT name FROM pragma_table_info('frontmatter')\".`);\n  }\n  return new SenseError('SEARCH_SYNTAX', `${message} -- searchable columns are title, summary, text; frontmatter fields are queried with --where or sense query (list them with pragma_table_info('frontmatter')).`);\n}\n\nexport interface FindOptions {\n  k?: number;\n  where?: string; // SQL fragment against frontmatter alias `f`, e.g. \"f.status = 'active'\"\n  semantic?: boolean; // invoke vector expansion (requires features.embed); rows gain via 'vector' and a lines column\n}\n\n// Layer 1: BM25 + link-graph expansion, fused by reciprocal rank. `via` says which\n// signal produced each row so the agent knows what evidence it is trusting.\n// Semantic expansion is per-query opt-in: without opts.semantic the result is\n// byte-for-byte independent of the embed feature.\nexport async function find(db: DatabaseSync, cfg: Config, terms: string, opts: FindOptions = {}): Promise<Row[]> {\n  const k = opts.k ?? 10;\n  const fetch = Math.max(k * 3, 30);\n\n  // Terms pass verbatim to FTS5 MATCH: bare words AND-join, operators are the caller's.\n  // Invalid syntax propagates as an error, zero matches return zero -- no silent rewrites.\n  // --where applies inside the candidate query (a post-filter over the top-N would drop\n  // matches ranked past the pool) and again on the final select for link-derived rows.\n  // An explicit --where replaces the tree's declared default rather than ANDing with it,\n  // so a caller can always widen back to the whole tree.\n  const scope = opts.where ?? cfg.defaults?.find?.where;\n  const whereJoin = scope ? `JOIN frontmatter f ON f.\"path\" = content.path` : '';\n  const whereCond = scope ? `AND (${scope})` : '';\n  const matchSql = `SELECT content.path AS path, snippet(content, -1, '«', '»', '…', 10) AS hit FROM content ${whereJoin} WHERE content MATCH ? ${whereCond} ORDER BY ${WEIGHTED_BM25} LIMIT ${fetch}`;\n  let matchRows: Array<{ path: string; hit: string }>;\n  try {\n    matchRows = db.prepare(matchSql).all(terms) as Array<{ path: string; hit: string }>;\n  } catch (err) {\n    throw searchError(err as Error, terms, scope);\n  }\n\n  const hits = new Map(matchRows.map((r) => [r.path, r.hit]));\n  const candidates = new Map<string, { score: number; via: string }>();\n  matchRows.forEach((r, i) => {\n    candidates.set(r.path, { score: 1 / (RRF_K + i), via: 'match' });\n  });\n\n  const edges = featureEnabled(cfg, 'links') && matchRows.length > 0 ? linkEdges(db) : [];\n  if (edges.length > 0) {\n    // `linked` gates the label only, not the score: PPR restart mass gives every seed a\n    // nonzero rank even without an incident edge, which is not link evidence — but dropping\n    // that mass from the score list reweights fusion toward connectivity and measurably\n    // wrecks ranking on link-dense corpora (FEVER hit@10 0.997 -> 0.907; fusion-tuning.md).\n    const linked = new Set(edges.flat());\n    const nodes = (db.prepare('SELECT \"path\" FROM frontmatter').all() as Array<{ path: string }>).map((r) => r.path);\n    const seeds = new Map(matchRows.map((r, i) => [r.path, 1 / (i + 1)]));\n    const ranked = [...personalizedRank(nodes, edges, seeds)]\n      .filter(([, score]) => score > 1e-9)\n      .sort((a, b) => b[1] - a[1])\n      .slice(0, fetch);\n    ranked.forEach(([path], i) => {\n      const existing = candidates.get(path);\n      if (existing) {\n        existing.score += 1 / (RRF_K + i);\n        if (linked.has(path)) existing.via = 'match+link';\n      } else {\n        candidates.set(path, { score: 1 / (RRF_K + i), via: 'link' });\n      }\n    });\n  }\n\n  // Vector expansion, invoked only: a third RRF list at the swept flat-region constants\n  // (weight 1, pool = fetch). Each row carries its best chunk's line range.\n  const chunkLines = new Map<string, string>();\n  const chunkSimilarity = new Map<string, number>();\n  if (opts.semantic) {\n    const vec = await semanticCandidates(db, cfg, terms, fetch);\n    vec.forEach(({ path, lines, similarity }, i) => {\n      chunkLines.set(path, lines);\n      chunkSimilarity.set(path, similarity);\n      const existing = candidates.get(path);\n      if (existing) {\n        existing.score += 1 / (RRF_K + i);\n        existing.via = `${existing.via}+vector`;\n      } else {\n        candidates.set(path, { score: 1 / (RRF_K + i), via: 'vector' });\n      }\n    });\n  }\n\n  db.exec('DROP TABLE IF EXISTS _find');\n  db.exec('CREATE TEMP TABLE _find (\"path\" TEXT PRIMARY KEY, score REAL, via TEXT, hit TEXT, lines TEXT, similarity REAL)');\n  const insert = db.prepare('INSERT INTO _find (\"path\", score, via, hit, lines, similarity) VALUES (?, ?, ?, ?, ?, ?)');\n  for (const [path, c] of candidates) insert.run(path, c.score, c.via, hits.get(path) ?? null, chunkLines.get(path) ?? null, chunkSimilarity.get(path) ?? null);\n\n  const where = scope ? `WHERE ${scope}` : '';\n  const linesCol = opts.semantic ? ', _find.lines, _find.similarity' : '';\n  return db\n    .prepare(\n      `SELECT f.\"path\" AS path, content.title, content.summary, _find.hit, _find.via, round(_find.score, 4) AS score${linesCol}\n       FROM _find JOIN frontmatter f ON f.\"path\" = _find.\"path\" JOIN content ON content.path = _find.\"path\"\n       ${where} ORDER BY _find.score DESC LIMIT ?`\n    )\n    .all(k) as Row[];\n}\n\nexport interface TreeMap {\n  docs: { count: number; bytes: number };\n  fields: Row[]; // top 20 by coverage; fieldsTotal carries the real count\n  fieldsTotal: number;\n  features: { on: FeatureName[]; off: FeatureName[] };\n  hubs: Row[];\n  recent: Row[];\n}\n\nconst INTERNAL_COLUMNS = new Set(['path', '_mtime', '_size', '_rank']);\n\n// Layer 0: what is this tree. Fixed-size output regardless of tree size.\nexport function mapTree(db: DatabaseSync, cfg: Config): TreeMap {\n  const docs = db.prepare('SELECT COUNT(*) AS count, COALESCE(SUM(\"_size\"), 0) AS bytes FROM frontmatter').get() as { count: number; bytes: number };\n\n  const columns = (db.prepare('PRAGMA table_info(frontmatter)').all() as Array<{ name: string }>).map((r) => r.name).filter((name) => !INTERNAL_COLUMNS.has(name));\n  const allFields = columns\n    .map((name) => {\n      const quoted = `\"${name.split('\"').join('\"\"')}\"`;\n      // Observed storage class, not a declared one: these columns are added dynamically and\n      // SQLite types per value, so a field can be text in most notes and numeric in a few.\n      // Listing every distinct type makes both the type and any drift visible.\n      const { n, types } = db.prepare(`SELECT COUNT(${quoted}) AS n, GROUP_CONCAT(DISTINCT typeof(${quoted})) AS types FROM frontmatter WHERE ${quoted} IS NOT NULL`).get() as { n: number; types: string | null };\n      return { field: name, coverage: n, type: types ?? '' };\n    })\n    .sort((a, b) => (b.coverage as number) - (a.coverage as number)) as Row[];\n  const fields = allFields.slice(0, 20);\n\n  const hubs = featureEnabled(cfg, 'rank') ? (db.prepare(`SELECT f.\"path\" AS path, round(f.\"_rank\" * 100, 2) AS rank, content.title FROM frontmatter f JOIN content ON content.path = f.\"path\" WHERE f.\"_rank\" IS NOT NULL ORDER BY f.\"_rank\" DESC LIMIT 8`).all() as Row[]) : [];\n\n  const recent = db.prepare(`SELECT \"path\", datetime(\"_mtime\" / 1000, 'unixepoch') AS modified FROM frontmatter ORDER BY \"_mtime\" DESC LIMIT 5`).all() as Row[];\n\n  return { docs, fields, fieldsTotal: allFields.length, features: featureStates(cfg), hubs, recent };\n}\n\nexport interface Peek {\n  path: string;\n  tokens: number;\n  frontmatter: Row;\n  sections: Row[];\n  outbound: string[];\n  backlinks: string[];\n  unresolved: string[];\n  // Totals before truncation: a hub can have thousands of backlinks, and peek's whole\n  // point is bounded output. Query the links table directly for the full list.\n  outboundTotal: number;\n  backlinksTotal: number;\n  unresolvedTotal: number;\n  off: FeatureName[]; // disabled features whose blocks are omitted (not empty)\n}\n\nconst PEEK_LINK_LIMIT = 20;\n\n// Layer 2: everything about one note except its prose -- frontmatter, outline with line\n// ranges + token estimates (so the follow-up Read is a range, not the file), links both ways.\nexport function peek(db: DatabaseSync, cfg: Config, pathArg: string): Peek {\n  const paths = (db.prepare('SELECT \"path\" FROM frontmatter').all() as Array<{ path: string }>).map((r) => r.path);\n  let path = paths.find((p) => p === pathArg);\n  if (!path) {\n    const base = posix.basename(pathArg).replace(/\\.md$/i, '').toLowerCase();\n    const matches = paths.filter((p) => posix.basename(p).replace(/\\.md$/i, '').toLowerCase() === base);\n    if (matches.length === 1) path = matches[0];\n    else if (matches.length > 1) throw new SenseError('NOTE_AMBIGUOUS', `\"${pathArg}\" is ambiguous: ${matches.join(', ')}`);\n    else throw new SenseError('NOTE_NOT_FOUND', `no note matches \"${pathArg}\"`);\n  }\n\n  const row = db.prepare('SELECT * FROM frontmatter WHERE \"path\" = ?').get(path) as Row;\n  const frontmatter: Row = {};\n  for (const [key, value] of Object.entries(row)) {\n    if (!INTERNAL_COLUMNS.has(key) && value !== null) frontmatter[key] = value;\n  }\n\n  const sections = featureEnabled(cfg, 'sections') ? (db.prepare('SELECT level, heading, start_line, end_line, tokens FROM sections WHERE \"path\" = ? ORDER BY idx').all(path) as Row[]) : [];\n\n  let outbound: string[] = [];\n  let backlinks: string[] = [];\n  let unresolved: string[] = [];\n  let backlinksTotal = 0;\n  if (featureEnabled(cfg, 'links')) {\n    const out = db.prepare('SELECT target, dst FROM links WHERE src = ? ORDER BY target').all(path) as Array<{ target: string; dst: string | null }>;\n    outbound = [...new Set(out.filter((l) => l.dst !== null).map((l) => l.dst as string))];\n    unresolved = out.filter((l) => l.dst === null).map((l) => l.target);\n    backlinksTotal = (db.prepare('SELECT COUNT(DISTINCT src) AS n FROM links WHERE dst = ?').get(path) as { n: number }).n;\n    backlinks = (db.prepare('SELECT DISTINCT src FROM links WHERE dst = ? ORDER BY src LIMIT ?').all(path, PEEK_LINK_LIMIT) as Array<{ src: string }>).map((r) => r.src);\n  }\n\n  return {\n    path,\n    tokens: Math.ceil(((row._size as number) ?? 0) / 4),\n    frontmatter,\n    sections,\n    outbound: outbound.slice(0, PEEK_LINK_LIMIT),\n    backlinks,\n    unresolved: unresolved.slice(0, PEEK_LINK_LIMIT),\n    outboundTotal: outbound.length,\n    backlinksTotal,\n    unresolvedTotal: unresolved.length,\n    off: (['sections', 'links'] as FeatureName[]).filter((name) => !featureEnabled(cfg, name)),\n  };\n}\n"],"names":["posix","featureEnabled","featureStates","SenseError","semanticCandidates","linkEdges","personalizedRank","WEIGHTED_BM25","RRF_K","FTS5_OPERATORS","searchError","err","terms","scope","message","test","suspects","match","filter","t","startsWith","col","exec","fromTerms","undefined","length","some","split","includes","map","join","find","db","cfg","opts","hits","chunkLines","chunkSimilarity","k","fetch","Math","max","where","defaults","whereJoin","whereCond","matchSql","matchRows","prepare","all","Map","r","path","hit","candidates","forEach","i","set","score","via","edges","linked","Set","flat","nodes","seeds","ranked","sort","a","b","slice","existing","get","has","semantic","vec","lines","similarity","insert","c","run","linesCol","INTERNAL_COLUMNS","mapTree","docs","columns","name","allFields","quoted","n","types","field","coverage","type","fields","hubs","recent","fieldsTotal","features","PEEK_LINK_LIMIT","peek","pathArg","row","paths","p","base","basename","replace","toLowerCase","matches","frontmatter","key","value","Object","entries","sections","outbound","backlinks","unresolved","backlinksTotal","out","l","dst","target","src","tokens","ceil","_size","outboundTotal","unresolvedTotal","off"],"mappings":"AAAA,OAAOA,WAAW,kBAAkB;AAGpC,SAASC,cAAc,EAAEC,aAAa,QAAQ,cAAc;AAC5D,SAASC,UAAU,QAAQ,cAAc;AACzC,SAASC,kBAAkB,QAAQ,sBAAsB;AACzD,SAASC,SAAS,QAAQ,sBAAsB;AAChD,SAASC,gBAAgB,QAAQ,aAAa;AAG9C,4EAA4E;AAC5E,gFAAgF;AAEhF,MAAMC,gBAAgB;AACtB,MAAMC,QAAQ;AAEd,uFAAuF;AACvF,uFAAuF;AACvF,sFAAsF;AACtF,8EAA8E;AAC9E,MAAMC,iBAAiB;AAEvB,SAASC,YAAYC,GAAU,EAAEC,KAAa,EAAEC,KAAc;QAG1CD;QAIN;IANZ,MAAME,UAAUH,IAAIG,OAAO;IAC3B,IAAI,CAAC,oDAAoDC,IAAI,CAACD,UAAU,OAAOH;IAC/E,MAAMK,WAAW,EAACJ,eAAAA,MAAMK,KAAK,CAAC,qBAAZL,0BAAAA,eAAuB,EAAE,EAAEM,MAAM,CAAC,CAACC,IAAM,CAACA,EAAEC,UAAU,CAAC,QAAQX,eAAeM,IAAI,CAACI;IACrG,wFAAwF;IACxF,qFAAqF;IACrF,gFAAgF;IAChF,MAAME,OAAM,QAAA,wBAAwBC,IAAI,CAACR,sBAA7B,4BAAA,KAAuC,CAAC,EAAE;IACtD,MAAMS,YAAYF,QAAQG,YAAYR,SAASS,MAAM,GAAG,IAAIT,SAASU,IAAI,CAAC,CAACP,IAAMA,EAAEQ,KAAK,CAAC,mBAAmBC,QAAQ,CAACP;IACrH,IAAIE,aAAaP,SAASS,MAAM,GAAG,GAAG;QACpC,OAAO,IAAItB,WAAW,iBAAiB,GAAGW,QAAQ,uBAAuB,EAAEE,SAASa,GAAG,CAAC,CAACV,IAAM,CAAC,EAAE,EAAEA,EAAE,EAAE,CAAC,EAAEW,IAAI,CAAC,MAAM,gFAAgF,EAAEd,QAAQ,CAAC,EAAE,CAAC,gDAAgD,CAAC;IACvQ;IACA,IAAIK,QAAQG,aAAaX,UAAUW,WAAW;QAC5C,OAAO,IAAIrB,WAAW,iBAAiB,GAAGW,QAAQ,yBAAyB,EAAED,MAAM,mHAAmH,CAAC;IACzM;IACA,OAAO,IAAIV,WAAW,iBAAiB,GAAGW,QAAQ,8JAA8J,CAAC;AACnN;AAQA,mFAAmF;AACnF,4EAA4E;AAC5E,8EAA8E;AAC9E,kDAAkD;AAClD,OAAO,eAAeiB,KAAKC,EAAgB,EAAEC,GAAW,EAAErB,KAAa,EAAEsB,OAAoB,CAAC,CAAC;QACnFA,SASIA,aA+DuDC,WAAwBC,iBAA8BC;QA/D/FJ,oBAAAA;IAT5B,MAAMK,KAAIJ,UAAAA,KAAKI,CAAC,cAANJ,qBAAAA,UAAU;IACpB,MAAMK,QAAQC,KAAKC,GAAG,CAACH,IAAI,GAAG;IAE9B,sFAAsF;IACtF,yFAAyF;IACzF,sFAAsF;IACtF,qFAAqF;IACrF,uFAAuF;IACvF,uDAAuD;IACvD,MAAMzB,SAAQqB,cAAAA,KAAKQ,KAAK,cAAVR,yBAAAA,eAAcD,gBAAAA,IAAIU,QAAQ,cAAZV,qCAAAA,qBAAAA,cAAcF,IAAI,cAAlBE,yCAAAA,mBAAoBS,KAAK;IACrD,MAAME,YAAY/B,QAAQ,CAAC,6CAA6C,CAAC,GAAG;IAC5E,MAAMgC,YAAYhC,QAAQ,CAAC,KAAK,EAAEA,MAAM,CAAC,CAAC,GAAG;IAC7C,MAAMiC,WAAW,CAAC,yFAAyF,EAAEF,UAAU,uBAAuB,EAAEC,UAAU,UAAU,EAAEtC,cAAc,OAAO,EAAEgC,OAAO;IACpM,IAAIQ;IACJ,IAAI;QACFA,YAAYf,GAAGgB,OAAO,CAACF,UAAUG,GAAG,CAACrC;IACvC,EAAE,OAAOD,KAAK;QACZ,MAAMD,YAAYC,KAAcC,OAAOC;IACzC;IAEA,MAAMsB,OAAO,IAAIe,IAAIH,UAAUlB,GAAG,CAAC,CAACsB,IAAM;YAACA,EAAEC,IAAI;YAAED,EAAEE,GAAG;SAAC;IACzD,MAAMC,aAAa,IAAIJ;IACvBH,UAAUQ,OAAO,CAAC,CAACJ,GAAGK;QACpBF,WAAWG,GAAG,CAACN,EAAEC,IAAI,EAAE;YAAEM,OAAO,IAAKlD,CAAAA,QAAQgD,CAAAA;YAAIG,KAAK;QAAQ;IAChE;IAEA,MAAMC,QAAQ3D,eAAegC,KAAK,YAAYc,UAAUtB,MAAM,GAAG,IAAIpB,UAAU2B,MAAM,EAAE;IACvF,IAAI4B,MAAMnC,MAAM,GAAG,GAAG;QACpB,oFAAoF;QACpF,wFAAwF;QACxF,oFAAoF;QACpF,wFAAwF;QACxF,MAAMoC,SAAS,IAAIC,IAAIF,MAAMG,IAAI;QACjC,MAAMC,QAAQ,AAAChC,GAAGgB,OAAO,CAAC,kCAAkCC,GAAG,GAA+BpB,GAAG,CAAC,CAACsB,IAAMA,EAAEC,IAAI;QAC/G,MAAMa,QAAQ,IAAIf,IAAIH,UAAUlB,GAAG,CAAC,CAACsB,GAAGK,IAAM;gBAACL,EAAEC,IAAI;gBAAE,IAAKI,CAAAA,IAAI,CAAA;aAAG;QACnE,MAAMU,SAAS;eAAI5D,iBAAiB0D,OAAOJ,OAAOK;SAAO,CACtD/C,MAAM,CAAC,CAAC,GAAGwC,MAAM,GAAKA,QAAQ,MAC9BS,IAAI,CAAC,CAACC,GAAGC,IAAMA,CAAC,CAAC,EAAE,GAAGD,CAAC,CAAC,EAAE,EAC1BE,KAAK,CAAC,GAAG/B;QACZ2B,OAAOX,OAAO,CAAC,CAAC,CAACH,KAAK,EAAEI;YACtB,MAAMe,WAAWjB,WAAWkB,GAAG,CAACpB;YAChC,IAAImB,UAAU;gBACZA,SAASb,KAAK,IAAI,IAAKlD,CAAAA,QAAQgD,CAAAA;gBAC/B,IAAIK,OAAOY,GAAG,CAACrB,OAAOmB,SAASZ,GAAG,GAAG;YACvC,OAAO;gBACLL,WAAWG,GAAG,CAACL,MAAM;oBAAEM,OAAO,IAAKlD,CAAAA,QAAQgD,CAAAA;oBAAIG,KAAK;gBAAO;YAC7D;QACF;IACF;IAEA,sFAAsF;IACtF,0EAA0E;IAC1E,MAAMvB,aAAa,IAAIc;IACvB,MAAMb,kBAAkB,IAAIa;IAC5B,IAAIhB,KAAKwC,QAAQ,EAAE;QACjB,MAAMC,MAAM,MAAMvE,mBAAmB4B,IAAIC,KAAKrB,OAAO2B;QACrDoC,IAAIpB,OAAO,CAAC,CAAC,EAAEH,IAAI,EAAEwB,KAAK,EAAEC,UAAU,EAAE,EAAErB;YACxCpB,WAAWqB,GAAG,CAACL,MAAMwB;YACrBvC,gBAAgBoB,GAAG,CAACL,MAAMyB;YAC1B,MAAMN,WAAWjB,WAAWkB,GAAG,CAACpB;YAChC,IAAImB,UAAU;gBACZA,SAASb,KAAK,IAAI,IAAKlD,CAAAA,QAAQgD,CAAAA;gBAC/Be,SAASZ,GAAG,GAAG,GAAGY,SAASZ,GAAG,CAAC,OAAO,CAAC;YACzC,OAAO;gBACLL,WAAWG,GAAG,CAACL,MAAM;oBAAEM,OAAO,IAAKlD,CAAAA,QAAQgD,CAAAA;oBAAIG,KAAK;gBAAS;YAC/D;QACF;IACF;IAEA3B,GAAGV,IAAI,CAAC;IACRU,GAAGV,IAAI,CAAC;IACR,MAAMwD,SAAS9C,GAAGgB,OAAO,CAAC;IAC1B,KAAK,MAAM,CAACI,MAAM2B,EAAE,IAAIzB,WAAYwB,OAAOE,GAAG,CAAC5B,MAAM2B,EAAErB,KAAK,EAAEqB,EAAEpB,GAAG,GAAExB,YAAAA,KAAKqC,GAAG,CAACpB,mBAATjB,uBAAAA,YAAkB,OAAMC,kBAAAA,WAAWoC,GAAG,CAACpB,mBAAfhB,6BAAAA,kBAAwB,OAAMC,uBAAAA,gBAAgBmC,GAAG,CAACpB,mBAApBf,kCAAAA,uBAA6B;IAExJ,MAAMK,QAAQ7B,QAAQ,CAAC,MAAM,EAAEA,OAAO,GAAG;IACzC,MAAMoE,WAAW/C,KAAKwC,QAAQ,GAAG,oCAAoC;IACrE,OAAO1C,GACJgB,OAAO,CACN,CAAC,6GAA6G,EAAEiC,SAAS;;OAExH,EAAEvC,MAAM,kCAAkC,CAAC,EAE7CO,GAAG,CAACX;AACT;AAWA,MAAM4C,mBAAmB,IAAIpB,IAAI;IAAC;IAAQ;IAAU;IAAS;CAAQ;AAErE,yEAAyE;AACzE,OAAO,SAASqB,QAAQnD,EAAgB,EAAEC,GAAW;IACnD,MAAMmD,OAAOpD,GAAGgB,OAAO,CAAC,iFAAiFwB,GAAG;IAE5G,MAAMa,UAAU,AAACrD,GAAGgB,OAAO,CAAC,kCAAkCC,GAAG,GAA+BpB,GAAG,CAAC,CAACsB,IAAMA,EAAEmC,IAAI,EAAEpE,MAAM,CAAC,CAACoE,OAAS,CAACJ,iBAAiBT,GAAG,CAACa;IAC1J,MAAMC,YAAYF,QACfxD,GAAG,CAAC,CAACyD;QACJ,MAAME,SAAS,CAAC,CAAC,EAAEF,KAAK3D,KAAK,CAAC,KAAKG,IAAI,CAAC,MAAM,CAAC,CAAC;QAChD,sFAAsF;QACtF,qFAAqF;QACrF,yEAAyE;QACzE,MAAM,EAAE2D,CAAC,EAAEC,KAAK,EAAE,GAAG1D,GAAGgB,OAAO,CAAC,CAAC,aAAa,EAAEwC,OAAO,qCAAqC,EAAEA,OAAO,mCAAmC,EAAEA,OAAO,YAAY,CAAC,EAAEhB,GAAG;QACnK,OAAO;YAAEmB,OAAOL;YAAMM,UAAUH;YAAGI,IAAI,EAAEH,kBAAAA,mBAAAA,QAAS;QAAG;IACvD,GACCvB,IAAI,CAAC,CAACC,GAAGC,IAAM,AAACA,EAAEuB,QAAQ,GAAexB,EAAEwB,QAAQ;IACtD,MAAME,SAASP,UAAUjB,KAAK,CAAC,GAAG;IAElC,MAAMyB,OAAO9F,eAAegC,KAAK,UAAWD,GAAGgB,OAAO,CAAC,CAAC,gMAAgM,CAAC,EAAEC,GAAG,KAAe,EAAE;IAE/Q,MAAM+C,SAAShE,GAAGgB,OAAO,CAAC,CAAC,iHAAiH,CAAC,EAAEC,GAAG;IAElJ,OAAO;QAAEmC;QAAMU;QAAQG,aAAaV,UAAU9D,MAAM;QAAEyE,UAAUhG,cAAc+B;QAAM8D;QAAMC;IAAO;AACnG;AAkBA,MAAMG,kBAAkB;AAExB,wFAAwF;AACxF,8FAA8F;AAC9F,OAAO,SAASC,KAAKpE,EAAgB,EAAEC,GAAW,EAAEoE,OAAe;QAiC3CC;IAhCtB,MAAMC,QAAQ,AAACvE,GAAGgB,OAAO,CAAC,kCAAkCC,GAAG,GAA+BpB,GAAG,CAAC,CAACsB,IAAMA,EAAEC,IAAI;IAC/G,IAAIA,OAAOmD,MAAMxE,IAAI,CAAC,CAACyE,IAAMA,MAAMH;IACnC,IAAI,CAACjD,MAAM;QACT,MAAMqD,OAAOzG,MAAM0G,QAAQ,CAACL,SAASM,OAAO,CAAC,UAAU,IAAIC,WAAW;QACtE,MAAMC,UAAUN,MAAMrF,MAAM,CAAC,CAACsF,IAAMxG,MAAM0G,QAAQ,CAACF,GAAGG,OAAO,CAAC,UAAU,IAAIC,WAAW,OAAOH;QAC9F,IAAII,QAAQpF,MAAM,KAAK,GAAG2B,OAAOyD,OAAO,CAAC,EAAE;aACtC,IAAIA,QAAQpF,MAAM,GAAG,GAAG,MAAM,IAAItB,WAAW,kBAAkB,CAAC,CAAC,EAAEkG,QAAQ,gBAAgB,EAAEQ,QAAQ/E,IAAI,CAAC,OAAO;aACjH,MAAM,IAAI3B,WAAW,kBAAkB,CAAC,iBAAiB,EAAEkG,QAAQ,CAAC,CAAC;IAC5E;IAEA,MAAMC,MAAMtE,GAAGgB,OAAO,CAAC,8CAA8CwB,GAAG,CAACpB;IACzE,MAAM0D,cAAmB,CAAC;IAC1B,KAAK,MAAM,CAACC,KAAKC,MAAM,IAAIC,OAAOC,OAAO,CAACZ,KAAM;QAC9C,IAAI,CAACpB,iBAAiBT,GAAG,CAACsC,QAAQC,UAAU,MAAMF,WAAW,CAACC,IAAI,GAAGC;IACvE;IAEA,MAAMG,WAAWlH,eAAegC,KAAK,cAAeD,GAAGgB,OAAO,CAAC,mGAAmGC,GAAG,CAACG,QAAkB,EAAE;IAE1L,IAAIgE,WAAqB,EAAE;IAC3B,IAAIC,YAAsB,EAAE;IAC5B,IAAIC,aAAuB,EAAE;IAC7B,IAAIC,iBAAiB;IACrB,IAAItH,eAAegC,KAAK,UAAU;QAChC,MAAMuF,MAAMxF,GAAGgB,OAAO,CAAC,+DAA+DC,GAAG,CAACG;QAC1FgE,WAAW;eAAI,IAAItD,IAAI0D,IAAItG,MAAM,CAAC,CAACuG,IAAMA,EAAEC,GAAG,KAAK,MAAM7F,GAAG,CAAC,CAAC4F,IAAMA,EAAEC,GAAG;SAAa;QACtFJ,aAAaE,IAAItG,MAAM,CAAC,CAACuG,IAAMA,EAAEC,GAAG,KAAK,MAAM7F,GAAG,CAAC,CAAC4F,IAAMA,EAAEE,MAAM;QAClEJ,iBAAiB,AAACvF,GAAGgB,OAAO,CAAC,4DAA4DwB,GAAG,CAACpB,MAAwBqC,CAAC;QACtH4B,YAAY,AAACrF,GAAGgB,OAAO,CAAC,qEAAqEC,GAAG,CAACG,MAAM+C,iBAA4CtE,GAAG,CAAC,CAACsB,IAAMA,EAAEyE,GAAG;IACrK;IAEA,OAAO;QACLxE;QACAyE,QAAQrF,KAAKsF,IAAI,CAAC,EAAExB,aAAAA,IAAIyB,KAAK,cAATzB,wBAAAA,aAAwB,KAAK;QACjDQ;QACAK;QACAC,UAAUA,SAAS9C,KAAK,CAAC,GAAG6B;QAC5BkB;QACAC,YAAYA,WAAWhD,KAAK,CAAC,GAAG6B;QAChC6B,eAAeZ,SAAS3F,MAAM;QAC9B8F;QACAU,iBAAiBX,WAAW7F,MAAM;QAClCyG,KAAK,AAAC;YAAC;YAAY;SAAQ,CAAmBhH,MAAM,CAAC,CAACoE,OAAS,CAACrF,eAAegC,KAAKqD;IACtF;AACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sensemaking",
-  "version": "0.5.0",
+  "version": "0.7.0",
   "description": "Query a knowledge base you build with an agent: filter notes by frontmatter, then search inside them",
   "keywords": [
     "markdown",
@@ -47,6 +47,7 @@
     "version": "tsds version"
   },
   "dependencies": {
+    "@huggingface/tokenizers": "^0.1.3",
     "fast-glob": "^3.3.3",
     "remove-markdown": "^0.6.4",
     "yaml": "^2.9.0"

package/schema.json

@@ -45,12 +45,46 @@
         "rank": {
           "type": "boolean",
           "description": "PageRank over resolved links into `frontmatter._rank` at reconcile -- a static importance prior. Powers `map`'s hub list; usable in any ORDER BY. Requires `links`."
+        },
+        "embed": {
+          "description": "Semantic vectors, opt-in (absent = off; most trees don't need them). When on, vectors are computed and kept fresh at reconcile; expansion runs only on `find --semantic` invocations -- default `find` results are unchanged. `true` = defaults (static type, minishlab/potion-retrieval-32M, downloaded to ~/.cache/sensemaking on first use). Object form: `model` (Hugging Face id or local path), `type` (`static` = built-in pure-JS Model2Vec loader; `api` = OpenAI-compatible POST /embeddings), `url` (api base, e.g. http://localhost:11434/v1), `key` (name of the env var holding the bearer token). Changing model or type rebuilds the cache like a feature toggle.",
+          "oneOf": [
+            { "type": "boolean" },
+            {
+              "type": "object",
+              "additionalProperties": false,
+              "properties": {
+                "model": { "type": "string" },
+                "type": { "type": "string", "enum": ["static", "api"] },
+                "url": { "type": "string" },
+                "key": { "type": "string" }
+              }
+            }
+          ]
+        }
+      }
+    },
+    "defaults": {
+      "type": "object",
+      "additionalProperties": false,
+      "description": "Tree-declared defaults for the verbs.",
+      "properties": {
+        "find": {
+          "type": "object",
+          "additionalProperties": false,
+          "description": "Default scope for `find`. Use when a tree holds a layer that should stay queryable but out of ordinary search -- generated output, or unverified source extractions that must not be cited as conclusions.",
+          "properties": {
+            "where": {
+              "type": "string",
+              "description": "SQL condition against frontmatter alias `f`, e.g. \"f.type != 'raw'\". Applied when `find` runs without --where; an explicit --where replaces it (so `--where \"1=1\"` searches the whole tree). Reported by `sense status`."
+            }
+          }
         }
       }
     },
     "queries": {
       "type": "object",
-      "description": "Named SQL queries runnable as `sense <name> [params...]`. Tables: `frontmatter` (one row per file, one column per discovered frontmatter key, plus `path`/`_mtime`/`_size`/`_rank`), `content` (FTS5: `title`, `summary`, `text`, `path`), `links` (`src`, `target`, `dst`), and `sections` (`path`, `idx`, `level`, `heading`, `start_line`, `end_line`, `tokens`). `?` placeholders bind to CLI positional args in order. `has(field, value)`: array membership on a JSON-array field, substring match on a string (so has(f.status, 'active') also matches 'inactive'), false on NULL. Exact matches: `=` for scalars, `EXISTS (SELECT 1 FROM json_each(f.tags) WHERE value = ?)` for array members. Canonical query: `SELECT f.path, content.title, content.summary, snippet(content, -1, '«', '»', '…', 10) AS hit FROM frontmatter f JOIN content ON content.path = f.path WHERE content MATCH ? ORDER BY bm25(content, 10.0, 5.0, 1.0) LIMIT 10`. Reserved frontmatter keys: `path`, `_mtime`, `_size`, `_rank`, `content`, `links`, `sections`. Reserved query names (unreachable as subcommands): `init`, `query`, `find`, `map`, `peek`, `watch`, `status`, `rebuild`.",
+      "description": "Named SQL queries runnable as `sense <name> [params...]`. Tables: `frontmatter` (one row per file, one column per discovered frontmatter key, plus `path`/`_mtime`/`_size`/`_rank`), `content` (FTS5: `title`, `summary`, `text`, `path`), `links` (`src`, `target`, `dst`), and `sections` (`path`, `idx`, `level`, `heading`, `start_line`, `end_line`, `tokens`). `?` placeholders bind to CLI positional args in order. `has(field, value)`: array membership on a JSON-array field, substring match on a string (so has(f.status, 'active') also matches 'inactive'), false on NULL. Exact matches: `=` for scalars, `EXISTS (SELECT 1 FROM json_each(f.tags) WHERE value = ?)` for array members. Canonical query: `SELECT f.path, content.title, content.summary, snippet(content, -1, '«', '»', '…', 10) AS hit FROM frontmatter f JOIN content ON content.path = f.path WHERE content MATCH ? ORDER BY bm25(content, 10.0, 5.0, 1.0) LIMIT 10`. Reserved frontmatter keys: `path`, `_mtime`, `_size`, `_rank`, `content`, `links`, `sections`. Reserved query names (unreachable as subcommands): `init`, `query`, `find`, `map`, `peek`, `watch`, `status`, `rebuild`, `check`.",
       "additionalProperties": { "type": "string" }
     }
   }

package/skills/sense/EXAMPLES.md

@@ -75,6 +75,22 @@ sense --list                       # named queries someone already saved
 sense query "SELECT DISTINCT type FROM frontmatter"    # what a field's values are
 ```
 
+## F. "The notes say it in different words"
+
+```
+sense find "children dying from poor nutrition" --semantic --k 3 --format json
+```
+
+```json
+[
+  { "path": "notes/malnutrition-outcomes.md", "title": "Malnutrition outcomes",
+    "summary": "Stunting and mortality by region", "hit": null, "via": "vector", "score": 0.0167, "lines": "L14-52" }
+]
+```
+
+A `via: "vector"` row never contained the terms — it is semantically near them; `lines` is the
+best-matching section, a direct `Read` range. Only on trees whose config enables `features.embed`.
+
 ## Consequences
 
 | query | what happens | bounded alternative |

package/skills/sense/SKILL.md

@@ -10,22 +10,32 @@ SQL over a markdown tree, kept fresh by a filesystem check on every query. Four
 `title`, `summary`, `text`), `links` (`src`, `target`, `dst` — `NULL` dst = dead link),
 `sections` (heading outline with line ranges and token estimates).
 
-## The descent
+## What each tool is for
 
-Spend tokens in this order (progressive disclosure: metadata first, payloads just in time); go
-only as deep as the question needs.
+Every result is a reference (path, metadata, excerpt), never file contents; prose enters
+context only when you Read it. Costs: `map` is fixed-size, a `find` row ~30 tokens, `peek`
+~17% of reading the file. Which tool fits is a property of the question:
 
-1. `sense map` — orient once: fields, hub notes, recent changes. Fixed-size output.
-2. `sense find "<terms>"` — locate: ranked references with excerpts, ~30 tokens/row.
-   FTS5 standard: bare words AND-join (one absent word = zero rows) — write
-   `a OR b OR c` for any-word matching.
-3. `sense peek <path>` — structure before reading: outline with `[L143-162, ~380t]` ranges,
-   links both ways. ~17% the cost of reading the file.
-4. `Read` — the payload. On large files, peek's line ranges let you read just the section
-   you need; small files are often cheaper to read whole.
+- A deterministic, factual answer over known fields — counts, filters, "which notes have
+  X" — is SQL: `sense query`, a named query, or `find --where`. Enumerates every match;
+  same result regardless of phrasing.
+- Locating notes by words in their prose is `find` — ranked lexical match. Results shift as
+  phrasing shifts, and bare words AND-join (one absent word = zero rows): write
+  `a OR b OR c` for any-word matching.
+- A conceptual question the notes phrase in different words is `find --semantic` (exists only
+  on trees whose config enables `embed`): adds meaning-based candidates labeled `via: vector`.
+  Conceptual similarity, not typo-tolerance; false positives are expected, labeled, and
+  bounded by `--k`.
+- `map` answers "what is this tree" — fields, hub notes, recent changes — when the tree is
+  unfamiliar.
+- `peek <path>` prices a file before you pay for it: outline with `[L143-162, ~380t]`
+  ranges, links both ways.
+- When you know the file and need its contents, `Read` it — sense adds nothing there. On
+  large files peek's ranges let you read just one section; small files are often cheaper
+  whole.
 
-Every result is a reference, never file contents. Output defaults to a table, built for
-humans; `--format json` returns the same rows machine-parseable.
+Output defaults to a table, built for humans; `--format json` returns the same rows
+machine-parseable.
 
 ## Verbs
 
@@ -35,7 +45,7 @@ sense peek notes/pricing-model.md               # a unique basename also works
 sense map
 sense query "<sql>" [params...]                 # ad-hoc SQL; ? binds positional args, count-checked
 sense <name> [params...]                        # named query from sense.config.json
-sense --list | status | rebuild
+sense --list | status | rebuild | check
 ```
 
 - Terms pass verbatim to FTS5 MATCH. Bare words AND-join — one absent word means zero rows —
@@ -44,15 +54,37 @@ sense --list | status | rebuild
   rules apply to search commands you write into subagent briefs.
 - When a search misses, the recall levers are: OR-in synonyms and concrete instances (the
   index only knows the words in the files — a note about a specific tool rarely names its
-  category), and raise `--k` (a row costs ~30 tokens). Each widening adds candidates and
-  dilutes ranking, so the noise trade-off runs both ways.
+  category), raise `--k` (a row costs ~30 tokens), and on embed-enabled trees `--semantic`
+  (matches meaning where term overlap fails). Each widening adds candidates and dilutes
+  ranking, so the noise trade-off runs both ways.
 - A frontmatter query enumerates its matches deterministically; search ranks by term overlap,
   so results shift as phrasing shifts. Trade-off: a query needs a known field, search doesn't.
 - `find` fuses BM25 with link-graph expansion; the `via` column says what produced each row —
-  `match` (terms hit), `link` (connected to notes that hit), `match+link` (both).
-- `--where` takes a frontmatter condition against alias `f`, e.g. `"f.status = 'active' AND has(f.tags, 'x')"`.
+  `match` (terms hit), `link` (connected to notes that hit), `match+link` (both). With
+  `--semantic`, `vector` joins the composition and rows gain a `lines` column pointing at the
+  best-matching section, a direct `Read` range.
+- `--where` takes any SQL condition against frontmatter alias `f` — not only field equality:
+  `"f.status = 'active' AND has(f.tags, 'x')"`, `"f.path NOT LIKE 'generated/%'"`,
+  `"f.created >= datetime(?)"`. A tree can declare a default scope in `sense.config.json`
+  (`defaults.find.where`); an explicit `--where` replaces it, so `--where "1=1"` searches
+  everything. `sense status` prints the active default.
+- `score` is a rank-fusion value: it ranks rows within one result set and is not comparable
+  across queries, not a relevance magnitude — a perfect lexical hit and a weak vector-only
+  hit can both read ~0.017, because the number encodes how many signals fired and at what
+  rank. With `--semantic`, rows carry `similarity` (cosine, -1 to 1) which does measure
+  match quality: on a query whose words are absent from the tree, similarities sit near zero
+  (and can be negative), while a genuine paraphrase match runs high.
+- Lexical `find` returns 0 rows when nothing matches, so it answers "is this in the tree at
+  all". `--semantic` always returns up to `k` rows — nearest-neighbour search has a nearest
+  neighbour for any input — so absence is a lexical question; with `--semantic` the
+  `similarity` column is what separates a real hit from the best of a bad lot.
+- `sense check` prepares every saved query (catching syntax and unknown-column errors), runs
+  the ones taking no parameters, and prints row counts: a saved query returning 0 rows looks
+  the same as a true empty result until something distinguishes them.
 
-## SQL, when the verbs aren't enough
+## SQL
+
+The verbs are shorthands over the same four tables; anything they don't express, SQL does.
 
 ```
 sense query "SELECT name FROM pragma_table_info('frontmatter')"          # what fields exist
@@ -72,6 +104,10 @@ sense query "SELECT j.value, COUNT(*) n FROM frontmatter, json_each(frontmatter.
   `snippet(content, -1, '«', '»', '…', 10)`.
 - Select `content.title`/`content.summary` (always exist, empty when absent) rather than
   `f.title`/`f.summary` (discovered columns — error on trees that never declare them).
+- Frontmatter values keep their YAML type: strings are TEXT, whole numbers and booleans are
+  INTEGER (`true` stores as 1, so `WHERE flag = 1` matches and `WHERE flag = 'true'` matches
+  nothing), fractions are REAL, lists and maps are JSON text. `map` prints the observed type
+  per field, and a field showing two types (`integer,text`) has drifted across notes.
 - `has(field, value)`: array membership on JSON-array fields, substring on strings, false on NULL
   — the `includes()` convention. Substring means `has(f.status, 'active')` also matches
   `inactive`; exact scalar match is `f.status = ?`, deliberate substring is `LIKE`, exact array
@@ -92,7 +128,11 @@ Worked traces: [EXAMPLES.md](EXAMPLES.md).
 ## Setup and upkeep
 
 - Missing CLI: `npm install -g sensemaking`. Missing config: `sense init` at the tree root.
-  Discovery walks up from cwd; `--config <path>` overrides.
+  Discovery walks up from cwd; `--config <path>` overrides. Setting up or restructuring a
+  tree (features, frontmatter conventions, note design) is the `sense-setup` skill.
+- `map` and `status` report feature state (`features: links, sections, rank · off: embed
+  (features.embed)`). Invoking a capability whose feature is off is an error naming the
+  config key to enable — nothing silently falls back.
 - Save a query into `sense.config.json` only when it will be reused; run ad-hoc otherwise.
 - A one-line `summary:` per note is optional and pays twice: it appears in result rows and is a
   weighted search field. Date comparisons work for dates written as ISO 8601 (`2026-08-12`, or

package/skills/sense-setup/SKILL.md

@@ -0,0 +1,91 @@
+---
+name: sense-setup
+description: Set up the sense CLI on a markdown tree and make the tree-design decisions that shape it — sense init, enabling features (links, sections, rank, embed), and the trade-offs of frontmatter conventions, summaries, folder layout, and note size. Use when creating or restructuring a markdown knowledge base, running sense init, editing sense.config.json, enabling semantic search, or deciding how notes should be written for an agent to query later.
+---
+
+# sense: setup and tree design
+
+Querying an existing tree is the `sense` skill. This one covers making a tree:
+installing, configuring features, and the design decisions a tree owner faces.
+
+## Setup
+
+- `npm install -g sensemaking`, then `sense init` at the tree root writes
+  `sense.config.json` (all opt-out features on, `embed` off). Config discovery
+  walks up from cwd; `--config <path>` overrides.
+- `scan.include` globs resolve relative to the config file, never the cwd.
+- `sense map` and `sense status` report feature state
+  (`features: links, sections, rank · off: embed (features.embed)`), so the
+  current config is always discoverable from output.
+
+## Features
+
+| feature | powers | cost when on | config key |
+|---|---|---|---|
+| `links` | backlinks, dead-link queries, `find`'s link expansion, `peek`'s link lists | link re-resolution at reconcile | `features.links` (default on) |
+| `sections` | `peek`'s outline, line-range reads, per-section token estimates | heading extraction at parse | `features.sections` (default on) |
+| `rank` | `map`'s hubs, `_rank` in any ORDER BY | PageRank pass at reconcile; requires `links` | `features.rank` (default on) |
+| `embed` | `find --semantic` (meaning-based expansion) | vectors computed at reconcile; model download to `~/.cache/sensemaking` on first use; ~40 ms model load per semantic invocation | `features.embed` (default off) |
+
+- `embed` accepts `true` (built-in static model) or an object: `model` (Hugging
+  Face id or local path), `type` (`static` pure-JS built-in, or `api` for any
+  OpenAI-compatible `/v1/embeddings` endpoint — Ollama, LM Studio, llama.cpp,
+  hosted), `url`, `key` (env var name). A local `model` path is fully offline.
+- The two types differ in what they match, not only in cost: `static` is a
+  context-free distilled model that handles paraphrase and reworded concepts
+  (a query like "delegating without micromanaging" can reach a note that uses
+  neither word); tight near-synonyms and domain jargon ("heart attack" for
+  "myocardial infarction") are where it misses and an `api` transformer model
+  tends to succeed. The gain concentrates where the searcher's vocabulary
+  differs from the notes' — measured in the retrieval eval (BENCHMARKING.md,
+  "Retrieval quality"): on a vocabulary-gap corpus semantic expansion adds
+  recall; where vocabulary overlaps, BM25 plus link expansion already answers
+  most queries and vectors mostly reorder.
+- Enabling `embed` changes no default `find` result — expansion runs only when
+  a query passes `--semantic`. Invoking `--semantic` on a tree without `embed`
+  is an error naming the config key.
+- Toggling any feature rebuilds the cache on the next query (safe, automatic).
+- Disabled features degrade output, visibly: `peek` prints
+  `sections: off (features.sections)` rather than an empty outline.
+
+## Tree design decisions
+
+These belong to the tree's owner. sense works with any of them and reads no
+instruction files of its own; each choice below only changes what queries can
+do, and every consequence is listed so the choice can be made deliberately.
+
+- **Frontmatter fields.** Columns are discovered per tree — whatever keys notes
+  declare become queryable. Consistent fields across notes make SQL filters
+  and named queries possible (`WHERE status = 'active'`). Reserved keys
+  (dropped with a warning): `path`, `_mtime`, `_size`, `_rank`, `content`,
+  `links`, `sections`. Values keep their YAML type: strings TEXT, whole numbers
+  and booleans INTEGER (`true` is 1), fractions REAL, lists and maps JSON text;
+  `map` prints the observed type per field.
+- **What a note omits is also a filter.** A layer that deliberately carries none
+  of the fields the saved views filter on is excluded from all of them without
+  any view naming the layer; only a view filtering solely on a field the layer
+  does share needs an explicit condition (`AND type != 'raw'`). Sparse fields cut
+  both ways: less of the tree filters when you want breadth, and exactly this
+  separation when layers differ in authority — source extractions vs.
+  conclusions, generated output vs. notes.
+- **Dates.** `datetime()` comparisons work for dates written as ISO 8601 —
+  the only format it parses. A tree that mixes date formats can store them,
+  but can't compare them in SQL.
+- **Summaries.** A one-line `summary:` is optional and pays twice: it shows in
+  every result row (often answering a question with no file read) and is a
+  weighted search field ranked above body text. The cost is writing and
+  maintaining the line as notes change.
+- **Folder shape.** sense is structure-indifferent: globs find the files,
+  paths are queryable text, links resolve by basename at any depth. Folders
+  are for the humans and agents navigating the tree, not for the index —
+  flat and nested trees query identically.
+- **Note size.** Many small notes: precise `find` hits, whole-file reads stay
+  cheap, more links to maintain. Fewer large notes: `sections` and `peek`
+  carry the cost down to line-range reads. Both work; per-section token
+  estimates exist either way.
+- **Where decisions live.** Choices that should outlive one conversation can
+  be recorded in the agent's own instruction or skill files, or in a note in
+  the tree itself; a one-off search over an existing corpus needs none of
+  that. Whether to settle a choice with the user or proceed on the corpus as
+  found depends on whether the agent is only querying or also authoring —
+  an authoring agent's choices compound; a querying agent's don't.