sensemaking 0.11.3 → 0.11.4

package/dist/cjs/scan.d.cts

@@ -25,6 +25,7 @@ export interface ParsedDoc {
     };
     extracted: Record<string, unknown>;
 }
+export declare function looksLikeDatetime(value: string): boolean;
 export declare function normalizeDate(value: string): string;
 export declare function parseFile(file: FileStat, extractors?: Feature[]): {
     doc: ParsedDoc;

package/dist/cjs/scan.d.ts

@@ -25,6 +25,7 @@ export interface ParsedDoc {
     };
     extracted: Record<string, unknown>;
 }
+export declare function looksLikeDatetime(value: string): boolean;
 export declare function normalizeDate(value: string): string;
 export declare function parseFile(file: FileStat, extractors?: Feature[]): {
     doc: ParsedDoc;

package/dist/cjs/scan.js

@@ -15,6 +15,9 @@ _export(exports, {
     get listFiles () {
         return listFiles;
     },
+    get looksLikeDatetime () {
+        return looksLikeDatetime;
+    },
     get normalizeDate () {
         return normalizeDate;
     },
@@ -234,12 +237,19 @@ function listFiles(cfg, baseDir) {
 }
 // SQLite's datetime() rejects a colonless offset (`-0800`) and a space separator, which ISO 8601
 // allows and producers emit. A rejected date is invisible, not excluded: every comparison is NULL.
-var ISO_DATETIME = /^(\d{4}-\d{2}-\d{2})[T ](\d{2}:\d{2}(?::\d{2})?(?:\.\d+)?)(Z|[+-]\d{2}:?\d{2})?$/;
+var ISO_DATETIME = /^(\d{4}-\d{2}-\d{2})[T ](\d{2}:\d{2}(?::\d{2})?(?:\.\d+)?)(Z|[+-]\d{2}(?::?\d{2})?)?$/;
+// A value opening `YYYY-MM-DDT` was meant to be a datetime; prose never is. Reported when it
+// cannot be normalized, so a typo surfaces on the next crawl instead of at some later audit.
+var MEANT_AS_DATETIME = /^\d{4}-\d{2}-\d{2}[T ]\d/;
+function looksLikeDatetime(value) {
+    return MEANT_AS_DATETIME.test(value);
+}
 function normalizeDate(value) {
     var m = ISO_DATETIME.exec(value);
     if (m === null) return value;
     var _m = _sliced_to_array(m, 4), date = _m[1], time = _m[2], zone = _m[3];
-    var offset = zone !== undefined && zone !== 'Z' && !zone.includes(':') ? "".concat(zone.slice(0, 3), ":").concat(zone.slice(3)) : zone !== null && zone !== void 0 ? zone : '';
+    var digits = zone === undefined || zone === 'Z' ? '' : zone.replace(':', '');
+    var offset = digits === '' ? zone !== null && zone !== void 0 ? zone : '' : digits.length === 3 ? "".concat(digits, ":00") : "".concat(digits.slice(0, 3), ":").concat(digits.slice(3));
     var normalized = "".concat(date, "T").concat(time).concat(offset);
     return Number.isNaN(Date.parse(normalized)) ? value : normalized;
 }
@@ -366,7 +376,11 @@ function parseFile(file) {
                 warnings.push("warning: ".concat(file.relPath, ' has a frontmatter key named "').concat(key, '", which is reserved; ignoring it'));
                 continue;
             }
-            mapped[key] = mapValue(data[key]);
+            var value = mapValue(data[key]);
+            if (typeof value === 'string' && looksLikeDatetime(value) && Number.isNaN(Date.parse(value))) {
+                warnings.push("warning: ".concat(file.relPath, ": ").concat(key, " is not a valid date (").concat(value, "), so it is invisible to every date comparison"));
+            }
+            mapped[key] = value;
         }
     } catch (err) {
         _didIteratorError = true;

package/dist/cjs/scan.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/scan.ts"],"sourcesContent":["import { globSync, readFileSync, statSync } from 'node:fs';\nimport { join, sep } from 'node:path';\nimport removeMarkdown from 'remove-markdown';\nimport { isCollection, parseDocument, visit } from 'yaml';\nimport type { Config } from './config.ts';\nimport { embedEnabled, presetNames, presetSemanticEnabled } from './config.ts';\nimport type { Feature } from './features/types.ts';\n\n// Filesystem -> rows. Pure data in, data + warnings out; db.ts does the SQL.\n\n// Frontmatter keys that would collide with table columns. Exported so db.ts's upsert can tell\n// a feature-owned column (`_rank`) from a parsed one and leave it alone on reparse.\nexport const RESERVED_COLUMNS = new Set(['path', '_mtime', '_size', '_rank', '_parse_error', 'content', 'links', 'sections']);\n\n// YAML error codes whose recovery is unambiguous, so the parse is accepted rather than\n// quarantined. Only one qualifies: YAML 1.2 reserves `@` and `` ` `` at the start of a plain\n// scalar for future use, so they can never be valid and the text can only be what was typed\n// (`aliases: [@handle]` -> [\"@handle\"]). Every other code has a second reading -- an unquoted\n// `:` swallows the keys after it, an unquoted `[..](..)` drops the URL, a duplicate key picks\n// one value in silence -- so it writes values nobody wrote. See plans/frontmatter-parse-policy.md.\nconst ACCEPTED_YAML_CODES = new Set(['BAD_SCALAR_START']);\n\nfunction normalizeText(value: unknown): string {\n  if (value === null || value === undefined) return '';\n  return String(value).replace(/\\s+/g, ' ').trim();\n}\n\n// Keeps URL query strings, asset filenames, and HTML attributes out of the index: rare terms\n// carry high IDF, so they outrank prose. remove-markdown misses wikilinks and tables.\nfunction stripText(value: string): string {\n  const withoutWikilinks = value.replace(/\\[\\[([^\\]|]+)\\|([^\\]]+)\\]\\]/g, '$2').replace(/\\[\\[([^\\]]+)\\]\\]/g, '$1');\n  const withoutMarkdown = removeMarkdown(withoutWikilinks);\n  const withoutTables = withoutMarkdown.replace(/^\\s*\\|?[-\\s|:]+\\|\\s*$/gm, '').replace(/\\|/g, ' ');\n  return normalizeText(withoutTables);\n}\n\nexport interface FileStat {\n  relPath: string;\n  absPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[]; // every declared preset covering this file (>= 1; union, overlap allowed)\n  embed: boolean; // true iff a model is named and some covering preset has semantic on\n}\n\n// Presets are views, not partitions: they overlap freely, and a file's covering set (not one\n// owner) drives indexing. Globs resolve relative to baseDir; unmatched files are not indexed.\nexport function toPosixPath(relPath: string, separator: string = sep): string {\n  return separator === '\\\\' ? relPath.split(separator).join('/') : relPath;\n}\n\n// Every command pays listFiles before it answers (the freshness check stats each file), so\n// per-file work here is the hottest path in the package. Everything derivable from the config\n// alone is computed once, above the loop.\nconst NO_THROW = { throwIfNoEntry: false } as const;\n\nexport function listFiles(cfg: Config, baseDir: string): FileStat[] {\n  const coverage = new Map<string, Set<string>>();\n  const posixNeeded = sep === '\\\\';\n  for (const name of presetNames(cfg)) {\n    const preset = cfg.presets[name];\n    for (const matched of globSync(preset.include, { cwd: baseDir, exclude: preset.exclude })) {\n      const relPath = posixNeeded ? toPosixPath(matched) : matched;\n      const set = coverage.get(relPath) ?? new Set<string>();\n      set.add(name);\n      coverage.set(relPath, set);\n    }\n  }\n\n  // Which presets want vectors is a property of the config, not of any file.\n  const embedding = embedEnabled(cfg);\n  const semanticPresets = embedding ? new Set(presetNames(cfg).filter((name) => presetSemanticEnabled(cfg, name))) : null;\n\n  const files: FileStat[] = [];\n  for (const relPath of [...coverage.keys()].sort()) {\n    const absPath = join(baseDir, relPath); // join re-applies the platform separator for fs calls\n    // node:fs glob matches directories and dangling symlinks; fast-glob returned neither, so\n    // one stat filters both back out (throwIfNoEntry keeps a dangling link from throwing).\n    const st = statSync(absPath, NO_THROW);\n    if (!st?.isFile()) continue;\n    const presets = [...(coverage.get(relPath) as Set<string>)].sort();\n    const embed = semanticPresets !== null && presets.some((name) => semanticPresets.has(name));\n    files.push({ relPath, absPath, mtimeMs: st.mtimeMs, size: st.size, presets, embed });\n  }\n  return files;\n}\n\nexport interface ParsedDoc {\n  relPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[];\n  data: Record<string, string | number | bigint | null>;\n  // NULL when the frontmatter parsed, the first YAML message otherwise. In the row rather than\n  // a side table so `SELECT *` and any `IS NULL` investigation trip over it without being asked.\n  parseError: string | null;\n  // title/summary are duplicated from frontmatter so bm25() can weight them above the body text.\n  search: { title: string; summary: string; text: string };\n  // Per-feature extraction results, keyed by feature name; features store them at reconcile.\n  extracted: Record<string, unknown>;\n}\n\n// SQLite's datetime() rejects a colonless offset (`-0800`) and a space separator, which ISO 8601\n// allows and producers emit. A rejected date is invisible, not excluded: every comparison is NULL.\nconst ISO_DATETIME = /^(\\d{4}-\\d{2}-\\d{2})[T ](\\d{2}:\\d{2}(?::\\d{2})?(?:\\.\\d+)?)(Z|[+-]\\d{2}:?\\d{2})?$/;\n\n// Punctuation only, never a timezone conversion: the offset survives, so substr(d,1,10) is still\n// the local date. A shape that is not a real instant is left as written, and stays auditable.\nexport function normalizeDate(value: string): string {\n  const m = ISO_DATETIME.exec(value);\n  if (m === null) return value;\n  const [, date, time, zone] = m;\n  const offset = zone !== undefined && zone !== 'Z' && !zone.includes(':') ? `${zone.slice(0, 3)}:${zone.slice(3)}` : (zone ?? '');\n  const normalized = `${date}T${time}${offset}`;\n  return Number.isNaN(Date.parse(normalized)) ? value : normalized;\n}\n\n// Storage class follows the YAML scalar. Booleans store as 1/0, so `WHERE flag = 1` matches\n// and `WHERE flag = 'true'` cannot; `map` prints observed types so the mismatch is visible.\nfunction mapValue(value: unknown): string | number | bigint | null {\n  if (value === null || value === undefined) return null;\n  if (typeof value === 'boolean') return BigInt(value ? 1 : 0);\n  if (typeof value === 'number') return Number.isSafeInteger(value) ? BigInt(value) : value;\n  if (typeof value === 'string') return normalizeDate(value);\n  return JSON.stringify(value);\n}\n\n// The delimiter split is all this package used gray-matter for.\nfunction splitFrontmatter(raw: string): { fm: string | null; body: string } {\n  const open = raw.match(/^---\\r?\\n/);\n  if (!open) return { fm: null, body: raw };\n  const rest = raw.slice(open[0].length);\n  const close = rest.match(/^---\\r?(\\n|$)/m);\n  if (!close || close.index === undefined) return { fm: null, body: raw };\n  return { fm: rest.slice(0, close.index), body: rest.slice(close.index + close[0].length) };\n}\n\n// A well-formed document can still hold a value nobody meant: `created: {{date}}` is valid\n// YAML for a flow map used as a mapping key, so it raises no error and stores\n// {\"{ date }\": null}. No error code can catch that, but yaml notices the stringified key, so\n// this reports it with the path instead (yaml's own warning has none, fires once per document,\n// and is what trains readers to discard stderr).\nfunction warnStringifiedKeys(relPath: string, doc: ReturnType<typeof parseDocument>, warnings: string[]): void {\n  let found = false;\n  // Nested, not top level: `created: {{date}}` puts the collection key one level down, inside\n  // the flow map that `{{...}}` parses as.\n  visit(doc, {\n    Pair(_key, pair) {\n      if (!isCollection(pair.key)) return undefined;\n      found = true;\n      return visit.BREAK;\n    },\n  });\n  // One per file: a template repeats the same mistake on every field it stamps.\n  if (found) warnings.push(`warning: ${relPath} frontmatter has a key that is itself a list or mapping, stored as text; this is usually an unrendered template placeholder like {{date}}`);\n}\n\n// Accept a clean parse, and one whose every error is unambiguous (ACCEPTED_YAML_CODES).\n// Anything else is quarantined: no frontmatter columns at all, and `_parse_error` carries the\n// reason. Recovering it would write values nobody wrote, which is worse than absence because\n// no query can see it. The file is still indexed -- content, links and sections never touch\n// frontmatter -- so a broken note stays searchable while it is being hunted for.\n// yaml's message continues onto a source excerpt, so the first line is the sentence -- minus\n// the colon that introduced the part being dropped.\nfunction firstLine(message: string): string {\n  return message.split('\\n')[0].replace(/:\\s*$/, '');\n}\n\nfunction parseFrontmatter(relPath: string, fm: string, warnings: string[]): { data: Record<string, unknown>; parseError: string | null } {\n  // logLevel silences yaml's own pathless warnings; warnStringifiedKeys re-reports the one\n  // that carries information, with the file it came from.\n  const doc = parseDocument(fm, { logLevel: 'silent' });\n  const refused = doc.errors.filter((err) => !ACCEPTED_YAML_CODES.has(err.code));\n  if (refused.length > 0) {\n    const detail = refused.length > 1 ? ` (and ${refused.length - 1} more)` : '';\n    const parseError = `${firstLine(refused[0].message)}${detail}`;\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  let data: unknown;\n  try {\n    data = doc.toJS();\n  } catch (err) {\n    // Reaches here with doc.errors empty: `title: **Bold**` parses, then opens an alias on\n    // materialisation. An empty error list is not a successful parse.\n    const parseError = firstLine((err as Error).message);\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  if (data === null || data === undefined) return { data: {}, parseError: null };\n  if (typeof data !== 'object' || Array.isArray(data)) {\n    const parseError = 'frontmatter is not a key-value mapping';\n    warnings.push(`warning: ${relPath} ${parseError}; none of it is indexed`);\n    return { data: {}, parseError };\n  }\n  warnStringifiedKeys(relPath, doc, warnings);\n  return { data: data as Record<string, unknown>, parseError: null };\n}\n\nexport function parseFile(file: FileStat, extractors: Feature[] = []): { doc: ParsedDoc; warnings: string[] } {\n  const raw = readFileSync(file.absPath, 'utf8');\n  const warnings: string[] = [];\n\n  const { fm, body: content } = splitFrontmatter(raw);\n  const { data, parseError } = fm === null ? { data: {} as Record<string, unknown>, parseError: null } : parseFrontmatter(file.relPath, fm, warnings);\n  const mapped: Record<string, string | number | bigint | null> = {};\n\n  for (const key of Object.keys(data)) {\n    if (RESERVED_COLUMNS.has(key)) {\n      warnings.push(`warning: ${file.relPath} has a frontmatter key named \"${key}\", which is reserved; ignoring it`);\n      continue;\n    }\n    mapped[key] = mapValue(data[key]);\n  }\n\n  // title/summary are plain YAML strings -- whitespace-collapse only;\n  // the prose gets the full markdown strip.\n  const search = { title: normalizeText(data.title), summary: normalizeText(data.summary), text: stripText(content) };\n\n  return {\n    doc: {\n      relPath: file.relPath,\n      mtimeMs: file.mtimeMs,\n      size: file.size,\n      presets: file.presets,\n      data: mapped,\n      parseError,\n      search,\n      extracted: Object.fromEntries(extractors.filter((f) => f.extract).map((f) => [f.name, f.extract?.(raw, content, search)])),\n    },\n    warnings,\n  };\n}\n"],"names":["RESERVED_COLUMNS","listFiles","normalizeDate","parseFile","toPosixPath","Set","ACCEPTED_YAML_CODES","normalizeText","value","undefined","String","replace","trim","stripText","withoutWikilinks","withoutMarkdown","removeMarkdown","withoutTables","relPath","separator","sep","split","join","NO_THROW","throwIfNoEntry","cfg","baseDir","coverage","Map","posixNeeded","presetNames","name","preset","presets","globSync","include","cwd","exclude","matched","set","get","add","embedding","embedEnabled","semanticPresets","filter","presetSemanticEnabled","files","keys","sort","absPath","st","statSync","isFile","embed","some","has","push","mtimeMs","size","ISO_DATETIME","m","exec","date","time","zone","offset","includes","slice","normalized","Number","isNaN","Date","parse","mapValue","BigInt","isSafeInteger","JSON","stringify","splitFrontmatter","raw","open","match","fm","body","rest","length","close","index","warnStringifiedKeys","doc","warnings","found","visit","Pair","_key","pair","isCollection","key","BREAK","firstLine","message","parseFrontmatter","parseDocument","logLevel","refused","errors","err","code","detail","parseError","data","toJS","Array","isArray","file","extractors","readFileSync","content","mapped","Object","search","title","summary","text","extracted","fromEntries","f","extract","map"],"mappings":";;;;;;;;;;;QAYaA;eAAAA;;QA4CGC;eAAAA;;QAoDAC;eAAAA;;QA6FAC;eAAAA;;QA1JAC;eAAAA;;;sBA/CiC;wBACvB;qEACC;oBACwB;wBAEc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAO1D,IAAMJ,mBAAmB,IAAIK,IAAI;IAAC;IAAQ;IAAU;IAAS;IAAS;IAAgB;IAAW;IAAS;CAAW;AAE5H,uFAAuF;AACvF,6FAA6F;AAC7F,4FAA4F;AAC5F,8FAA8F;AAC9F,8FAA8F;AAC9F,mGAAmG;AACnG,IAAMC,sBAAsB,IAAID,IAAI;IAAC;CAAmB;AAExD,SAASE,cAAcC,KAAc;IACnC,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,OAAOC,OAAOF,OAAOG,OAAO,CAAC,QAAQ,KAAKC,IAAI;AAChD;AAEA,6FAA6F;AAC7F,sFAAsF;AACtF,SAASC,UAAUL,KAAa;IAC9B,IAAMM,mBAAmBN,MAAMG,OAAO,CAAC,gCAAgC,MAAMA,OAAO,CAAC,qBAAqB;IAC1G,IAAMI,kBAAkBC,IAAAA,uBAAc,EAACF;IACvC,IAAMG,gBAAgBF,gBAAgBJ,OAAO,CAAC,2BAA2B,IAAIA,OAAO,CAAC,OAAO;IAC5F,OAAOJ,cAAcU;AACvB;AAaO,SAASb,YAAYc,OAAe;QAAEC,YAAAA,iEAAoBC,aAAG;IAClE,OAAOD,cAAc,OAAOD,QAAQG,KAAK,CAACF,WAAWG,IAAI,CAAC,OAAOJ;AACnE;AAEA,2FAA2F;AAC3F,8FAA8F;AAC9F,0CAA0C;AAC1C,IAAMK,WAAW;IAAEC,gBAAgB;AAAM;AAElC,SAASvB,UAAUwB,GAAW,EAAEC,OAAe;IACpD,IAAMC,WAAW,IAAIC;IACrB,IAAMC,cAAcT,aAAG,KAAK;QACvB,kCAAA,2BAAA;;QAAL,QAAK,YAAcU,IAAAA,qBAAW,EAACL,yBAA1B,SAAA,6BAAA,QAAA,yBAAA,iCAAgC;YAAhC,IAAMM,OAAN;YACH,IAAMC,SAASP,IAAIQ,OAAO,CAACF,KAAK;gBAC3B,mCAAA,4BAAA;;gBAAL,QAAK,aAAiBG,IAAAA,gBAAQ,EAACF,OAAOG,OAAO,EAAE;oBAAEC,KAAKV;oBAASW,SAASL,OAAOK,OAAO;gBAAC,uBAAlF,UAAA,8BAAA,SAAA,0BAAA,kCAAsF;oBAAtF,IAAMC,UAAN;wBAESX;oBADZ,IAAMT,UAAUW,cAAczB,YAAYkC,WAAWA;oBACrD,IAAMC,OAAMZ,gBAAAA,SAASa,GAAG,CAACtB,sBAAbS,2BAAAA,gBAAyB,IAAItB;oBACzCkC,IAAIE,GAAG,CAACV;oBACRJ,SAASY,GAAG,CAACrB,SAASqB;gBACxB;;gBALK;gBAAA;;;yBAAA,8BAAA;wBAAA;;;wBAAA;8BAAA;;;;QAMP;;QARK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAUL,2EAA2E;IAC3E,IAAMG,YAAYC,IAAAA,sBAAY,EAAClB;IAC/B,IAAMmB,kBAAkBF,YAAY,IAAIrC,IAAIyB,IAAAA,qBAAW,EAACL,KAAKoB,MAAM,CAAC,SAACd;eAASe,IAAAA,+BAAqB,EAACrB,KAAKM;UAAU;IAEnH,IAAMgB,QAAoB,EAAE;QACvB,mCAAA,4BAAA;;QAAL,QAAK,aAAiB,AAAC,qBAAGpB,SAASqB,IAAI,IAAIC,IAAI,uBAA1C,UAAA,8BAAA,SAAA,0BAAA,kCAA8C;YAA9C,IAAM/B,WAAN;YACH,IAAMgC,UAAU5B,IAAAA,cAAI,EAACI,SAASR,WAAU,sDAAsD;YAC9F,yFAAyF;YACzF,uFAAuF;YACvF,IAAMiC,KAAKC,IAAAA,gBAAQ,EAACF,SAAS3B;YAC7B,IAAI,EAAC4B,eAAAA,yBAAAA,GAAIE,MAAM,KAAI;YACnB,IAAMpB,UAAU,AAAC,qBAAIN,SAASa,GAAG,CAACtB,WAA0B+B,IAAI;YAChE,IAAMK,QAAQV,oBAAoB,QAAQX,QAAQsB,IAAI,CAAC,SAACxB;uBAASa,gBAAgBY,GAAG,CAACzB;;YACrFgB,MAAMU,IAAI,CAAC;gBAAEvC,SAAAA;gBAASgC,SAAAA;gBAASQ,SAASP,GAAGO,OAAO;gBAAEC,MAAMR,GAAGQ,IAAI;gBAAE1B,SAAAA;gBAASqB,OAAAA;YAAM;QACpF;;QATK;QAAA;;;iBAAA,8BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAUL,OAAOP;AACT;AAiBA,iGAAiG;AACjG,mGAAmG;AACnG,IAAMa,eAAe;AAId,SAAS1D,cAAcM,KAAa;IACzC,IAAMqD,IAAID,aAAaE,IAAI,CAACtD;IAC5B,IAAIqD,MAAM,MAAM,OAAOrD;IACvB,IAA6BqD,sBAAAA,OAApBE,OAAoBF,OAAdG,OAAcH,OAARI,OAAQJ;IAC7B,IAAMK,SAASD,SAASxD,aAAawD,SAAS,OAAO,CAACA,KAAKE,QAAQ,CAAC,OAAO,AAAC,GAAsBF,OAApBA,KAAKG,KAAK,CAAC,GAAG,IAAG,KAAiB,OAAdH,KAAKG,KAAK,CAAC,MAAQH,iBAAAA,kBAAAA,OAAQ;IAC7H,IAAMI,aAAa,AAAC,GAAUL,OAARD,MAAK,KAAUG,OAAPF,MAAc,OAAPE;IACrC,OAAOI,OAAOC,KAAK,CAACC,KAAKC,KAAK,CAACJ,eAAe7D,QAAQ6D;AACxD;AAEA,4FAA4F;AAC5F,4FAA4F;AAC5F,SAASK,SAASlE,KAAc;IAC9B,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,IAAI,OAAOD,UAAU,WAAW,OAAOmE,OAAOnE,QAAQ,IAAI;IAC1D,IAAI,OAAOA,UAAU,UAAU,OAAO8D,OAAOM,aAAa,CAACpE,SAASmE,OAAOnE,SAASA;IACpF,IAAI,OAAOA,UAAU,UAAU,OAAON,cAAcM;IACpD,OAAOqE,KAAKC,SAAS,CAACtE;AACxB;AAEA,gEAAgE;AAChE,SAASuE,iBAAiBC,GAAW;IACnC,IAAMC,OAAOD,IAAIE,KAAK,CAAC;IACvB,IAAI,CAACD,MAAM,OAAO;QAAEE,IAAI;QAAMC,MAAMJ;IAAI;IACxC,IAAMK,OAAOL,IAAIZ,KAAK,CAACa,IAAI,CAAC,EAAE,CAACK,MAAM;IACrC,IAAMC,QAAQF,KAAKH,KAAK,CAAC;IACzB,IAAI,CAACK,SAASA,MAAMC,KAAK,KAAK/E,WAAW,OAAO;QAAE0E,IAAI;QAAMC,MAAMJ;IAAI;IACtE,OAAO;QAAEG,IAAIE,KAAKjB,KAAK,CAAC,GAAGmB,MAAMC,KAAK;QAAGJ,MAAMC,KAAKjB,KAAK,CAACmB,MAAMC,KAAK,GAAGD,KAAK,CAAC,EAAE,CAACD,MAAM;IAAE;AAC3F;AAEA,2FAA2F;AAC3F,8EAA8E;AAC9E,6FAA6F;AAC7F,+FAA+F;AAC/F,iDAAiD;AACjD,SAASG,oBAAoBvE,OAAe,EAAEwE,GAAqC,EAAEC,QAAkB;IACrG,IAAIC,QAAQ;IACZ,4FAA4F;IAC5F,yCAAyC;IACzCC,IAAAA,WAAK,EAACH,KAAK;QACTI,MAAAA,SAAAA,KAAKC,IAAI,EAAEC,IAAI;YACb,IAAI,CAACC,IAAAA,kBAAY,EAACD,KAAKE,GAAG,GAAG,OAAOzF;YACpCmF,QAAQ;YACR,OAAOC,WAAK,CAACM,KAAK;QACpB;IACF;IACA,8EAA8E;IAC9E,IAAIP,OAAOD,SAASlC,IAAI,CAAC,AAAC,YAAmB,OAARvC,SAAQ;AAC/C;AAEA,wFAAwF;AACxF,8FAA8F;AAC9F,6FAA6F;AAC7F,4FAA4F;AAC5F,iFAAiF;AACjF,6FAA6F;AAC7F,oDAAoD;AACpD,SAASkF,UAAUC,OAAe;IAChC,OAAOA,QAAQhF,KAAK,CAAC,KAAK,CAAC,EAAE,CAACV,OAAO,CAAC,SAAS;AACjD;AAEA,SAAS2F,iBAAiBpF,OAAe,EAAEiE,EAAU,EAAEQ,QAAkB;IACvE,yFAAyF;IACzF,wDAAwD;IACxD,IAAMD,MAAMa,IAAAA,mBAAa,EAACpB,IAAI;QAAEqB,UAAU;IAAS;IACnD,IAAMC,UAAUf,IAAIgB,MAAM,CAAC7D,MAAM,CAAC,SAAC8D;eAAQ,CAACrG,oBAAoBkD,GAAG,CAACmD,IAAIC,IAAI;;IAC5E,IAAIH,QAAQnB,MAAM,GAAG,GAAG;QACtB,IAAMuB,SAASJ,QAAQnB,MAAM,GAAG,IAAI,AAAC,SAA2B,OAAnBmB,QAAQnB,MAAM,GAAG,GAAE,YAAU;QAC1E,IAAMwB,aAAa,AAAC,GAAkCD,OAAhCT,UAAUK,OAAO,CAAC,EAAE,CAACJ,OAAO,GAAW,OAAPQ;QACtDlB,SAASlC,IAAI,CAAC,AAAC,YAA2EqD,OAAhE5F,SAAQ,0DAAmE,OAAX4F;QAC1F,OAAO;YAAEC,MAAM,CAAC;YAAGD,YAAAA;QAAW;IAChC;IAEA,IAAIC;IACJ,IAAI;QACFA,OAAOrB,IAAIsB,IAAI;IACjB,EAAE,OAAOL,KAAK;QACZ,uFAAuF;QACvF,kEAAkE;QAClE,IAAMG,cAAaV,UAAU,AAACO,IAAcN,OAAO;QACnDV,SAASlC,IAAI,CAAC,AAAC,YAA2EqD,OAAhE5F,SAAQ,0DAAmE,OAAX4F;QAC1F,OAAO;YAAEC,MAAM,CAAC;YAAGD,YAAAA;QAAW;IAChC;IAEA,IAAIC,SAAS,QAAQA,SAAStG,WAAW,OAAO;QAAEsG,MAAM,CAAC;QAAGD,YAAY;IAAK;IAC7E,IAAI,CAAA,OAAOC,qCAAP,SAAOA,KAAG,MAAM,YAAYE,MAAMC,OAAO,CAACH,OAAO;QACnD,IAAMD,cAAa;QACnBnB,SAASlC,IAAI,CAAC,AAAC,YAAsBqD,OAAX5F,SAAQ,KAAc,OAAX4F,aAAW;QAChD,OAAO;YAAEC,MAAM,CAAC;YAAGD,YAAAA;QAAW;IAChC;IACArB,oBAAoBvE,SAASwE,KAAKC;IAClC,OAAO;QAAEoB,MAAMA;QAAiCD,YAAY;IAAK;AACnE;AAEO,SAAS3G,UAAUgH,IAAc;QAAEC,aAAAA,iEAAwB,EAAE;IAClE,IAAMpC,MAAMqC,IAAAA,oBAAY,EAACF,KAAKjE,OAAO,EAAE;IACvC,IAAMyC,WAAqB,EAAE;IAE7B,IAA8BZ,oBAAAA,iBAAiBC,MAAvCG,KAAsBJ,kBAAtBI,IAAIC,AAAMkC,UAAYvC,kBAAlBK;IACZ,IAA6BD,OAAAA,OAAO,OAAO;QAAE4B,MAAM,CAAC;QAA8BD,YAAY;IAAK,IAAIR,iBAAiBa,KAAKjG,OAAO,EAAEiE,IAAIQ,WAAlIoB,OAAqB5B,KAArB4B,MAAMD,aAAe3B,KAAf2B;IACd,IAAMS,SAA0D,CAAC;QAE5D,kCAAA,2BAAA;;QAAL,QAAK,YAAaC,OAAOxE,IAAI,CAAC+D,0BAAzB,SAAA,6BAAA,QAAA,yBAAA,iCAAgC;YAAhC,IAAMb,MAAN;YACH,IAAIlG,iBAAiBwD,GAAG,CAAC0C,MAAM;gBAC7BP,SAASlC,IAAI,CAAC,AAAC,YAAwDyC,OAA7CiB,KAAKjG,OAAO,EAAC,kCAAoC,OAAJgF,KAAI;gBAC3E;YACF;YACAqB,MAAM,CAACrB,IAAI,GAAGxB,SAASqC,IAAI,CAACb,IAAI;QAClC;;QANK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAQL,oEAAoE;IACpE,0CAA0C;IAC1C,IAAMuB,SAAS;QAAEC,OAAOnH,cAAcwG,KAAKW,KAAK;QAAGC,SAASpH,cAAcwG,KAAKY,OAAO;QAAGC,MAAM/G,UAAUyG;IAAS;IAElH,OAAO;QACL5B,KAAK;YACHxE,SAASiG,KAAKjG,OAAO;YACrBwC,SAASyD,KAAKzD,OAAO;YACrBC,MAAMwD,KAAKxD,IAAI;YACf1B,SAASkF,KAAKlF,OAAO;YACrB8E,MAAMQ;YACNT,YAAAA;YACAW,QAAAA;YACAI,WAAWL,OAAOM,WAAW,CAACV,WAAWvE,MAAM,CAAC,SAACkF;uBAAMA,EAAEC,OAAO;eAAEC,GAAG,CAAC,SAACF;oBAAeA;uBAAT;oBAACA,EAAEhG,IAAI;qBAAEgG,aAAAA,EAAEC,OAAO,cAATD,iCAAAA,gBAAAA,GAAY/C,KAAKsC,SAASG;iBAAQ;;QAC1H;QACA9B,UAAAA;IACF;AACF"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/scan.ts"],"sourcesContent":["import { globSync, readFileSync, statSync } from 'node:fs';\nimport { join, sep } from 'node:path';\nimport removeMarkdown from 'remove-markdown';\nimport { isCollection, parseDocument, visit } from 'yaml';\nimport type { Config } from './config.ts';\nimport { embedEnabled, presetNames, presetSemanticEnabled } from './config.ts';\nimport type { Feature } from './features/types.ts';\n\n// Filesystem -> rows. Pure data in, data + warnings out; db.ts does the SQL.\n\n// Frontmatter keys that would collide with table columns. Exported so db.ts's upsert can tell\n// a feature-owned column (`_rank`) from a parsed one and leave it alone on reparse.\nexport const RESERVED_COLUMNS = new Set(['path', '_mtime', '_size', '_rank', '_parse_error', 'content', 'links', 'sections']);\n\n// YAML error codes whose recovery is unambiguous, so the parse is accepted rather than\n// quarantined. Only one qualifies: YAML 1.2 reserves `@` and `` ` `` at the start of a plain\n// scalar for future use, so they can never be valid and the text can only be what was typed\n// (`aliases: [@handle]` -> [\"@handle\"]). Every other code has a second reading -- an unquoted\n// `:` swallows the keys after it, an unquoted `[..](..)` drops the URL, a duplicate key picks\n// one value in silence -- so it writes values nobody wrote. See plans/frontmatter-parse-policy.md.\nconst ACCEPTED_YAML_CODES = new Set(['BAD_SCALAR_START']);\n\nfunction normalizeText(value: unknown): string {\n  if (value === null || value === undefined) return '';\n  return String(value).replace(/\\s+/g, ' ').trim();\n}\n\n// Keeps URL query strings, asset filenames, and HTML attributes out of the index: rare terms\n// carry high IDF, so they outrank prose. remove-markdown misses wikilinks and tables.\nfunction stripText(value: string): string {\n  const withoutWikilinks = value.replace(/\\[\\[([^\\]|]+)\\|([^\\]]+)\\]\\]/g, '$2').replace(/\\[\\[([^\\]]+)\\]\\]/g, '$1');\n  const withoutMarkdown = removeMarkdown(withoutWikilinks);\n  const withoutTables = withoutMarkdown.replace(/^\\s*\\|?[-\\s|:]+\\|\\s*$/gm, '').replace(/\\|/g, ' ');\n  return normalizeText(withoutTables);\n}\n\nexport interface FileStat {\n  relPath: string;\n  absPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[]; // every declared preset covering this file (>= 1; union, overlap allowed)\n  embed: boolean; // true iff a model is named and some covering preset has semantic on\n}\n\n// Presets are views, not partitions: they overlap freely, and a file's covering set (not one\n// owner) drives indexing. Globs resolve relative to baseDir; unmatched files are not indexed.\nexport function toPosixPath(relPath: string, separator: string = sep): string {\n  return separator === '\\\\' ? relPath.split(separator).join('/') : relPath;\n}\n\n// Every command pays listFiles before it answers (the freshness check stats each file), so\n// per-file work here is the hottest path in the package. Everything derivable from the config\n// alone is computed once, above the loop.\nconst NO_THROW = { throwIfNoEntry: false } as const;\n\nexport function listFiles(cfg: Config, baseDir: string): FileStat[] {\n  const coverage = new Map<string, Set<string>>();\n  const posixNeeded = sep === '\\\\';\n  for (const name of presetNames(cfg)) {\n    const preset = cfg.presets[name];\n    for (const matched of globSync(preset.include, { cwd: baseDir, exclude: preset.exclude })) {\n      const relPath = posixNeeded ? toPosixPath(matched) : matched;\n      const set = coverage.get(relPath) ?? new Set<string>();\n      set.add(name);\n      coverage.set(relPath, set);\n    }\n  }\n\n  // Which presets want vectors is a property of the config, not of any file.\n  const embedding = embedEnabled(cfg);\n  const semanticPresets = embedding ? new Set(presetNames(cfg).filter((name) => presetSemanticEnabled(cfg, name))) : null;\n\n  const files: FileStat[] = [];\n  for (const relPath of [...coverage.keys()].sort()) {\n    const absPath = join(baseDir, relPath); // join re-applies the platform separator for fs calls\n    // node:fs glob matches directories and dangling symlinks; fast-glob returned neither, so\n    // one stat filters both back out (throwIfNoEntry keeps a dangling link from throwing).\n    const st = statSync(absPath, NO_THROW);\n    if (!st?.isFile()) continue;\n    const presets = [...(coverage.get(relPath) as Set<string>)].sort();\n    const embed = semanticPresets !== null && presets.some((name) => semanticPresets.has(name));\n    files.push({ relPath, absPath, mtimeMs: st.mtimeMs, size: st.size, presets, embed });\n  }\n  return files;\n}\n\nexport interface ParsedDoc {\n  relPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[];\n  data: Record<string, string | number | bigint | null>;\n  // NULL when the frontmatter parsed, the first YAML message otherwise. In the row rather than\n  // a side table so `SELECT *` and any `IS NULL` investigation trip over it without being asked.\n  parseError: string | null;\n  // title/summary are duplicated from frontmatter so bm25() can weight them above the body text.\n  search: { title: string; summary: string; text: string };\n  // Per-feature extraction results, keyed by feature name; features store them at reconcile.\n  extracted: Record<string, unknown>;\n}\n\n// SQLite's datetime() rejects a colonless offset (`-0800`) and a space separator, which ISO 8601\n// allows and producers emit. A rejected date is invisible, not excluded: every comparison is NULL.\nconst ISO_DATETIME = /^(\\d{4}-\\d{2}-\\d{2})[T ](\\d{2}:\\d{2}(?::\\d{2})?(?:\\.\\d+)?)(Z|[+-]\\d{2}(?::?\\d{2})?)?$/;\n\n// A value opening `YYYY-MM-DDT` was meant to be a datetime; prose never is. Reported when it\n// cannot be normalized, so a typo surfaces on the next crawl instead of at some later audit.\nconst MEANT_AS_DATETIME = /^\\d{4}-\\d{2}-\\d{2}[T ]\\d/;\n\nexport function looksLikeDatetime(value: string): boolean {\n  return MEANT_AS_DATETIME.test(value);\n}\n\n// Punctuation only, never a timezone conversion: the offset survives, so substr(d,1,10) is still\n// the local date. A shape that is not a real instant is left as written, and stays auditable.\nexport function normalizeDate(value: string): string {\n  const m = ISO_DATETIME.exec(value);\n  if (m === null) return value;\n  const [, date, time, zone] = m;\n  const digits = zone === undefined || zone === 'Z' ? '' : zone.replace(':', '');\n  const offset = digits === '' ? (zone ?? '') : digits.length === 3 ? `${digits}:00` : `${digits.slice(0, 3)}:${digits.slice(3)}`;\n  const normalized = `${date}T${time}${offset}`;\n  return Number.isNaN(Date.parse(normalized)) ? value : normalized;\n}\n\n// Storage class follows the YAML scalar. Booleans store as 1/0, so `WHERE flag = 1` matches\n// and `WHERE flag = 'true'` cannot; `map` prints observed types so the mismatch is visible.\nfunction mapValue(value: unknown): string | number | bigint | null {\n  if (value === null || value === undefined) return null;\n  if (typeof value === 'boolean') return BigInt(value ? 1 : 0);\n  if (typeof value === 'number') return Number.isSafeInteger(value) ? BigInt(value) : value;\n  if (typeof value === 'string') return normalizeDate(value);\n  return JSON.stringify(value);\n}\n\n// The delimiter split is all this package used gray-matter for.\nfunction splitFrontmatter(raw: string): { fm: string | null; body: string } {\n  const open = raw.match(/^---\\r?\\n/);\n  if (!open) return { fm: null, body: raw };\n  const rest = raw.slice(open[0].length);\n  const close = rest.match(/^---\\r?(\\n|$)/m);\n  if (!close || close.index === undefined) return { fm: null, body: raw };\n  return { fm: rest.slice(0, close.index), body: rest.slice(close.index + close[0].length) };\n}\n\n// A well-formed document can still hold a value nobody meant: `created: {{date}}` is valid\n// YAML for a flow map used as a mapping key, so it raises no error and stores\n// {\"{ date }\": null}. No error code can catch that, but yaml notices the stringified key, so\n// this reports it with the path instead (yaml's own warning has none, fires once per document,\n// and is what trains readers to discard stderr).\nfunction warnStringifiedKeys(relPath: string, doc: ReturnType<typeof parseDocument>, warnings: string[]): void {\n  let found = false;\n  // Nested, not top level: `created: {{date}}` puts the collection key one level down, inside\n  // the flow map that `{{...}}` parses as.\n  visit(doc, {\n    Pair(_key, pair) {\n      if (!isCollection(pair.key)) return undefined;\n      found = true;\n      return visit.BREAK;\n    },\n  });\n  // One per file: a template repeats the same mistake on every field it stamps.\n  if (found) warnings.push(`warning: ${relPath} frontmatter has a key that is itself a list or mapping, stored as text; this is usually an unrendered template placeholder like {{date}}`);\n}\n\n// Accept a clean parse, and one whose every error is unambiguous (ACCEPTED_YAML_CODES).\n// Anything else is quarantined: no frontmatter columns at all, and `_parse_error` carries the\n// reason. Recovering it would write values nobody wrote, which is worse than absence because\n// no query can see it. The file is still indexed -- content, links and sections never touch\n// frontmatter -- so a broken note stays searchable while it is being hunted for.\n// yaml's message continues onto a source excerpt, so the first line is the sentence -- minus\n// the colon that introduced the part being dropped.\nfunction firstLine(message: string): string {\n  return message.split('\\n')[0].replace(/:\\s*$/, '');\n}\n\nfunction parseFrontmatter(relPath: string, fm: string, warnings: string[]): { data: Record<string, unknown>; parseError: string | null } {\n  // logLevel silences yaml's own pathless warnings; warnStringifiedKeys re-reports the one\n  // that carries information, with the file it came from.\n  const doc = parseDocument(fm, { logLevel: 'silent' });\n  const refused = doc.errors.filter((err) => !ACCEPTED_YAML_CODES.has(err.code));\n  if (refused.length > 0) {\n    const detail = refused.length > 1 ? ` (and ${refused.length - 1} more)` : '';\n    const parseError = `${firstLine(refused[0].message)}${detail}`;\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  let data: unknown;\n  try {\n    data = doc.toJS();\n  } catch (err) {\n    // Reaches here with doc.errors empty: `title: **Bold**` parses, then opens an alias on\n    // materialisation. An empty error list is not a successful parse.\n    const parseError = firstLine((err as Error).message);\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  if (data === null || data === undefined) return { data: {}, parseError: null };\n  if (typeof data !== 'object' || Array.isArray(data)) {\n    const parseError = 'frontmatter is not a key-value mapping';\n    warnings.push(`warning: ${relPath} ${parseError}; none of it is indexed`);\n    return { data: {}, parseError };\n  }\n  warnStringifiedKeys(relPath, doc, warnings);\n  return { data: data as Record<string, unknown>, parseError: null };\n}\n\nexport function parseFile(file: FileStat, extractors: Feature[] = []): { doc: ParsedDoc; warnings: string[] } {\n  const raw = readFileSync(file.absPath, 'utf8');\n  const warnings: string[] = [];\n\n  const { fm, body: content } = splitFrontmatter(raw);\n  const { data, parseError } = fm === null ? { data: {} as Record<string, unknown>, parseError: null } : parseFrontmatter(file.relPath, fm, warnings);\n  const mapped: Record<string, string | number | bigint | null> = {};\n\n  for (const key of Object.keys(data)) {\n    if (RESERVED_COLUMNS.has(key)) {\n      warnings.push(`warning: ${file.relPath} has a frontmatter key named \"${key}\", which is reserved; ignoring it`);\n      continue;\n    }\n    const value = mapValue(data[key]);\n    if (typeof value === 'string' && looksLikeDatetime(value) && Number.isNaN(Date.parse(value))) {\n      warnings.push(`warning: ${file.relPath}: ${key} is not a valid date (${value}), so it is invisible to every date comparison`);\n    }\n    mapped[key] = value;\n  }\n\n  // title/summary are plain YAML strings -- whitespace-collapse only;\n  // the prose gets the full markdown strip.\n  const search = { title: normalizeText(data.title), summary: normalizeText(data.summary), text: stripText(content) };\n\n  return {\n    doc: {\n      relPath: file.relPath,\n      mtimeMs: file.mtimeMs,\n      size: file.size,\n      presets: file.presets,\n      data: mapped,\n      parseError,\n      search,\n      extracted: Object.fromEntries(extractors.filter((f) => f.extract).map((f) => [f.name, f.extract?.(raw, content, search)])),\n    },\n    warnings,\n  };\n}\n"],"names":["RESERVED_COLUMNS","listFiles","looksLikeDatetime","normalizeDate","parseFile","toPosixPath","Set","ACCEPTED_YAML_CODES","normalizeText","value","undefined","String","replace","trim","stripText","withoutWikilinks","withoutMarkdown","removeMarkdown","withoutTables","relPath","separator","sep","split","join","NO_THROW","throwIfNoEntry","cfg","baseDir","coverage","Map","posixNeeded","presetNames","name","preset","presets","globSync","include","cwd","exclude","matched","set","get","add","embedding","embedEnabled","semanticPresets","filter","presetSemanticEnabled","files","keys","sort","absPath","st","statSync","isFile","embed","some","has","push","mtimeMs","size","ISO_DATETIME","MEANT_AS_DATETIME","test","m","exec","date","time","zone","digits","offset","length","slice","normalized","Number","isNaN","Date","parse","mapValue","BigInt","isSafeInteger","JSON","stringify","splitFrontmatter","raw","open","match","fm","body","rest","close","index","warnStringifiedKeys","doc","warnings","found","visit","Pair","_key","pair","isCollection","key","BREAK","firstLine","message","parseFrontmatter","parseDocument","logLevel","refused","errors","err","code","detail","parseError","data","toJS","Array","isArray","file","extractors","readFileSync","content","mapped","Object","search","title","summary","text","extracted","fromEntries","f","extract","map"],"mappings":";;;;;;;;;;;QAYaA;eAAAA;;QA4CGC;eAAAA;;QAsDAC;eAAAA;;QAMAC;eAAAA;;QA8FAC;eAAAA;;QAnKAC;eAAAA;;;sBA/CiC;wBACvB;qEACC;oBACwB;wBAEc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAO1D,IAAML,mBAAmB,IAAIM,IAAI;IAAC;IAAQ;IAAU;IAAS;IAAS;IAAgB;IAAW;IAAS;CAAW;AAE5H,uFAAuF;AACvF,6FAA6F;AAC7F,4FAA4F;AAC5F,8FAA8F;AAC9F,8FAA8F;AAC9F,mGAAmG;AACnG,IAAMC,sBAAsB,IAAID,IAAI;IAAC;CAAmB;AAExD,SAASE,cAAcC,KAAc;IACnC,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,OAAOC,OAAOF,OAAOG,OAAO,CAAC,QAAQ,KAAKC,IAAI;AAChD;AAEA,6FAA6F;AAC7F,sFAAsF;AACtF,SAASC,UAAUL,KAAa;IAC9B,IAAMM,mBAAmBN,MAAMG,OAAO,CAAC,gCAAgC,MAAMA,OAAO,CAAC,qBAAqB;IAC1G,IAAMI,kBAAkBC,IAAAA,uBAAc,EAACF;IACvC,IAAMG,gBAAgBF,gBAAgBJ,OAAO,CAAC,2BAA2B,IAAIA,OAAO,CAAC,OAAO;IAC5F,OAAOJ,cAAcU;AACvB;AAaO,SAASb,YAAYc,OAAe;QAAEC,YAAAA,iEAAoBC,aAAG;IAClE,OAAOD,cAAc,OAAOD,QAAQG,KAAK,CAACF,WAAWG,IAAI,CAAC,OAAOJ;AACnE;AAEA,2FAA2F;AAC3F,8FAA8F;AAC9F,0CAA0C;AAC1C,IAAMK,WAAW;IAAEC,gBAAgB;AAAM;AAElC,SAASxB,UAAUyB,GAAW,EAAEC,OAAe;IACpD,IAAMC,WAAW,IAAIC;IACrB,IAAMC,cAAcT,aAAG,KAAK;QACvB,kCAAA,2BAAA;;QAAL,QAAK,YAAcU,IAAAA,qBAAW,EAACL,yBAA1B,SAAA,6BAAA,QAAA,yBAAA,iCAAgC;YAAhC,IAAMM,OAAN;YACH,IAAMC,SAASP,IAAIQ,OAAO,CAACF,KAAK;gBAC3B,mCAAA,4BAAA;;gBAAL,QAAK,aAAiBG,IAAAA,gBAAQ,EAACF,OAAOG,OAAO,EAAE;oBAAEC,KAAKV;oBAASW,SAASL,OAAOK,OAAO;gBAAC,uBAAlF,UAAA,8BAAA,SAAA,0BAAA,kCAAsF;oBAAtF,IAAMC,UAAN;wBAESX;oBADZ,IAAMT,UAAUW,cAAczB,YAAYkC,WAAWA;oBACrD,IAAMC,OAAMZ,gBAAAA,SAASa,GAAG,CAACtB,sBAAbS,2BAAAA,gBAAyB,IAAItB;oBACzCkC,IAAIE,GAAG,CAACV;oBACRJ,SAASY,GAAG,CAACrB,SAASqB;gBACxB;;gBALK;gBAAA;;;yBAAA,8BAAA;wBAAA;;;wBAAA;8BAAA;;;;QAMP;;QARK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAUL,2EAA2E;IAC3E,IAAMG,YAAYC,IAAAA,sBAAY,EAAClB;IAC/B,IAAMmB,kBAAkBF,YAAY,IAAIrC,IAAIyB,IAAAA,qBAAW,EAACL,KAAKoB,MAAM,CAAC,SAACd;eAASe,IAAAA,+BAAqB,EAACrB,KAAKM;UAAU;IAEnH,IAAMgB,QAAoB,EAAE;QACvB,mCAAA,4BAAA;;QAAL,QAAK,aAAiB,AAAC,qBAAGpB,SAASqB,IAAI,IAAIC,IAAI,uBAA1C,UAAA,8BAAA,SAAA,0BAAA,kCAA8C;YAA9C,IAAM/B,WAAN;YACH,IAAMgC,UAAU5B,IAAAA,cAAI,EAACI,SAASR,WAAU,sDAAsD;YAC9F,yFAAyF;YACzF,uFAAuF;YACvF,IAAMiC,KAAKC,IAAAA,gBAAQ,EAACF,SAAS3B;YAC7B,IAAI,EAAC4B,eAAAA,yBAAAA,GAAIE,MAAM,KAAI;YACnB,IAAMpB,UAAU,AAAC,qBAAIN,SAASa,GAAG,CAACtB,WAA0B+B,IAAI;YAChE,IAAMK,QAAQV,oBAAoB,QAAQX,QAAQsB,IAAI,CAAC,SAACxB;uBAASa,gBAAgBY,GAAG,CAACzB;;YACrFgB,MAAMU,IAAI,CAAC;gBAAEvC,SAAAA;gBAASgC,SAAAA;gBAASQ,SAASP,GAAGO,OAAO;gBAAEC,MAAMR,GAAGQ,IAAI;gBAAE1B,SAAAA;gBAASqB,OAAAA;YAAM;QACpF;;QATK;QAAA;;;iBAAA,8BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAUL,OAAOP;AACT;AAiBA,iGAAiG;AACjG,mGAAmG;AACnG,IAAMa,eAAe;AAErB,6FAA6F;AAC7F,6FAA6F;AAC7F,IAAMC,oBAAoB;AAEnB,SAAS5D,kBAAkBO,KAAa;IAC7C,OAAOqD,kBAAkBC,IAAI,CAACtD;AAChC;AAIO,SAASN,cAAcM,KAAa;IACzC,IAAMuD,IAAIH,aAAaI,IAAI,CAACxD;IAC5B,IAAIuD,MAAM,MAAM,OAAOvD;IACvB,IAA6BuD,sBAAAA,OAApBE,OAAoBF,OAAdG,OAAcH,OAARI,OAAQJ;IAC7B,IAAMK,SAASD,SAAS1D,aAAa0D,SAAS,MAAM,KAAKA,KAAKxD,OAAO,CAAC,KAAK;IAC3E,IAAM0D,SAASD,WAAW,KAAMD,iBAAAA,kBAAAA,OAAQ,KAAMC,OAAOE,MAAM,KAAK,IAAI,AAAC,GAAS,OAAPF,QAAO,SAAO,AAAC,GAAwBA,OAAtBA,OAAOG,KAAK,CAAC,GAAG,IAAG,KAAmB,OAAhBH,OAAOG,KAAK,CAAC;IAC3H,IAAMC,aAAa,AAAC,GAAUN,OAARD,MAAK,KAAUI,OAAPH,MAAc,OAAPG;IACrC,OAAOI,OAAOC,KAAK,CAACC,KAAKC,KAAK,CAACJ,eAAehE,QAAQgE;AACxD;AAEA,4FAA4F;AAC5F,4FAA4F;AAC5F,SAASK,SAASrE,KAAc;IAC9B,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,IAAI,OAAOD,UAAU,WAAW,OAAOsE,OAAOtE,QAAQ,IAAI;IAC1D,IAAI,OAAOA,UAAU,UAAU,OAAOiE,OAAOM,aAAa,CAACvE,SAASsE,OAAOtE,SAASA;IACpF,IAAI,OAAOA,UAAU,UAAU,OAAON,cAAcM;IACpD,OAAOwE,KAAKC,SAAS,CAACzE;AACxB;AAEA,gEAAgE;AAChE,SAAS0E,iBAAiBC,GAAW;IACnC,IAAMC,OAAOD,IAAIE,KAAK,CAAC;IACvB,IAAI,CAACD,MAAM,OAAO;QAAEE,IAAI;QAAMC,MAAMJ;IAAI;IACxC,IAAMK,OAAOL,IAAIZ,KAAK,CAACa,IAAI,CAAC,EAAE,CAACd,MAAM;IACrC,IAAMmB,QAAQD,KAAKH,KAAK,CAAC;IACzB,IAAI,CAACI,SAASA,MAAMC,KAAK,KAAKjF,WAAW,OAAO;QAAE6E,IAAI;QAAMC,MAAMJ;IAAI;IACtE,OAAO;QAAEG,IAAIE,KAAKjB,KAAK,CAAC,GAAGkB,MAAMC,KAAK;QAAGH,MAAMC,KAAKjB,KAAK,CAACkB,MAAMC,KAAK,GAAGD,KAAK,CAAC,EAAE,CAACnB,MAAM;IAAE;AAC3F;AAEA,2FAA2F;AAC3F,8EAA8E;AAC9E,6FAA6F;AAC7F,+FAA+F;AAC/F,iDAAiD;AACjD,SAASqB,oBAAoBzE,OAAe,EAAE0E,GAAqC,EAAEC,QAAkB;IACrG,IAAIC,QAAQ;IACZ,4FAA4F;IAC5F,yCAAyC;IACzCC,IAAAA,WAAK,EAACH,KAAK;QACTI,MAAAA,SAAAA,KAAKC,IAAI,EAAEC,IAAI;YACb,IAAI,CAACC,IAAAA,kBAAY,EAACD,KAAKE,GAAG,GAAG,OAAO3F;YACpCqF,QAAQ;YACR,OAAOC,WAAK,CAACM,KAAK;QACpB;IACF;IACA,8EAA8E;IAC9E,IAAIP,OAAOD,SAASpC,IAAI,CAAC,AAAC,YAAmB,OAARvC,SAAQ;AAC/C;AAEA,wFAAwF;AACxF,8FAA8F;AAC9F,6FAA6F;AAC7F,4FAA4F;AAC5F,iFAAiF;AACjF,6FAA6F;AAC7F,oDAAoD;AACpD,SAASoF,UAAUC,OAAe;IAChC,OAAOA,QAAQlF,KAAK,CAAC,KAAK,CAAC,EAAE,CAACV,OAAO,CAAC,SAAS;AACjD;AAEA,SAAS6F,iBAAiBtF,OAAe,EAAEoE,EAAU,EAAEO,QAAkB;IACvE,yFAAyF;IACzF,wDAAwD;IACxD,IAAMD,MAAMa,IAAAA,mBAAa,EAACnB,IAAI;QAAEoB,UAAU;IAAS;IACnD,IAAMC,UAAUf,IAAIgB,MAAM,CAAC/D,MAAM,CAAC,SAACgE;eAAQ,CAACvG,oBAAoBkD,GAAG,CAACqD,IAAIC,IAAI;;IAC5E,IAAIH,QAAQrC,MAAM,GAAG,GAAG;QACtB,IAAMyC,SAASJ,QAAQrC,MAAM,GAAG,IAAI,AAAC,SAA2B,OAAnBqC,QAAQrC,MAAM,GAAG,GAAE,YAAU;QAC1E,IAAM0C,aAAa,AAAC,GAAkCD,OAAhCT,UAAUK,OAAO,CAAC,EAAE,CAACJ,OAAO,GAAW,OAAPQ;QACtDlB,SAASpC,IAAI,CAAC,AAAC,YAA2EuD,OAAhE9F,SAAQ,0DAAmE,OAAX8F;QAC1F,OAAO;YAAEC,MAAM,CAAC;YAAGD,YAAAA;QAAW;IAChC;IAEA,IAAIC;IACJ,IAAI;QACFA,OAAOrB,IAAIsB,IAAI;IACjB,EAAE,OAAOL,KAAK;QACZ,uFAAuF;QACvF,kEAAkE;QAClE,IAAMG,cAAaV,UAAU,AAACO,IAAcN,OAAO;QACnDV,SAASpC,IAAI,CAAC,AAAC,YAA2EuD,OAAhE9F,SAAQ,0DAAmE,OAAX8F;QAC1F,OAAO;YAAEC,MAAM,CAAC;YAAGD,YAAAA;QAAW;IAChC;IAEA,IAAIC,SAAS,QAAQA,SAASxG,WAAW,OAAO;QAAEwG,MAAM,CAAC;QAAGD,YAAY;IAAK;IAC7E,IAAI,CAAA,OAAOC,qCAAP,SAAOA,KAAG,MAAM,YAAYE,MAAMC,OAAO,CAACH,OAAO;QACnD,IAAMD,cAAa;QACnBnB,SAASpC,IAAI,CAAC,AAAC,YAAsBuD,OAAX9F,SAAQ,KAAc,OAAX8F,aAAW;QAChD,OAAO;YAAEC,MAAM,CAAC;YAAGD,YAAAA;QAAW;IAChC;IACArB,oBAAoBzE,SAAS0E,KAAKC;IAClC,OAAO;QAAEoB,MAAMA;QAAiCD,YAAY;IAAK;AACnE;AAEO,SAAS7G,UAAUkH,IAAc;QAAEC,aAAAA,iEAAwB,EAAE;IAClE,IAAMnC,MAAMoC,IAAAA,oBAAY,EAACF,KAAKnE,OAAO,EAAE;IACvC,IAAM2C,WAAqB,EAAE;IAE7B,IAA8BX,oBAAAA,iBAAiBC,MAAvCG,KAAsBJ,kBAAtBI,IAAIC,AAAMiC,UAAYtC,kBAAlBK;IACZ,IAA6BD,OAAAA,OAAO,OAAO;QAAE2B,MAAM,CAAC;QAA8BD,YAAY;IAAK,IAAIR,iBAAiBa,KAAKnG,OAAO,EAAEoE,IAAIO,WAAlIoB,OAAqB3B,KAArB2B,MAAMD,aAAe1B,KAAf0B;IACd,IAAMS,SAA0D,CAAC;QAE5D,kCAAA,2BAAA;;QAAL,QAAK,YAAaC,OAAO1E,IAAI,CAACiE,0BAAzB,SAAA,6BAAA,QAAA,yBAAA,iCAAgC;YAAhC,IAAMb,MAAN;YACH,IAAIrG,iBAAiByD,GAAG,CAAC4C,MAAM;gBAC7BP,SAASpC,IAAI,CAAC,AAAC,YAAwD2C,OAA7CiB,KAAKnG,OAAO,EAAC,kCAAoC,OAAJkF,KAAI;gBAC3E;YACF;YACA,IAAM5F,QAAQqE,SAASoC,IAAI,CAACb,IAAI;YAChC,IAAI,OAAO5F,UAAU,YAAYP,kBAAkBO,UAAUiE,OAAOC,KAAK,CAACC,KAAKC,KAAK,CAACpE,SAAS;gBAC5FqF,SAASpC,IAAI,CAAC,AAAC,YAA4B2C,OAAjBiB,KAAKnG,OAAO,EAAC,MAAgCV,OAA5B4F,KAAI,0BAA8B,OAAN5F,OAAM;YAC/E;YACAiH,MAAM,CAACrB,IAAI,GAAG5F;QAChB;;QAVK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAYL,oEAAoE;IACpE,0CAA0C;IAC1C,IAAMmH,SAAS;QAAEC,OAAOrH,cAAc0G,KAAKW,KAAK;QAAGC,SAAStH,cAAc0G,KAAKY,OAAO;QAAGC,MAAMjH,UAAU2G;IAAS;IAElH,OAAO;QACL5B,KAAK;YACH1E,SAASmG,KAAKnG,OAAO;YACrBwC,SAAS2D,KAAK3D,OAAO;YACrBC,MAAM0D,KAAK1D,IAAI;YACf1B,SAASoF,KAAKpF,OAAO;YACrBgF,MAAMQ;YACNT,YAAAA;YACAW,QAAAA;YACAI,WAAWL,OAAOM,WAAW,CAACV,WAAWzE,MAAM,CAAC,SAACoF;uBAAMA,EAAEC,OAAO;eAAEC,GAAG,CAAC,SAACF;oBAAeA;uBAAT;oBAACA,EAAElG,IAAI;qBAAEkG,aAAAA,EAAEC,OAAO,cAATD,iCAAAA,gBAAAA,GAAY9C,KAAKqC,SAASG;iBAAQ;;QAC1H;QACA9B,UAAAA;IACF;AACF"}

package/dist/esm/scan.d.ts

@@ -25,6 +25,7 @@ export interface ParsedDoc {
     };
     extracted: Record<string, unknown>;
 }
+export declare function looksLikeDatetime(value: string): boolean;
 export declare function normalizeDate(value: string): string;
 export declare function parseFile(file: FileStat, extractors?: Feature[]): {
     doc: ParsedDoc;

package/dist/esm/scan.js

@@ -93,14 +93,21 @@ export function listFiles(cfg, baseDir) {
 }
 // SQLite's datetime() rejects a colonless offset (`-0800`) and a space separator, which ISO 8601
 // allows and producers emit. A rejected date is invisible, not excluded: every comparison is NULL.
-const ISO_DATETIME = /^(\d{4}-\d{2}-\d{2})[T ](\d{2}:\d{2}(?::\d{2})?(?:\.\d+)?)(Z|[+-]\d{2}:?\d{2})?$/;
+const ISO_DATETIME = /^(\d{4}-\d{2}-\d{2})[T ](\d{2}:\d{2}(?::\d{2})?(?:\.\d+)?)(Z|[+-]\d{2}(?::?\d{2})?)?$/;
+// A value opening `YYYY-MM-DDT` was meant to be a datetime; prose never is. Reported when it
+// cannot be normalized, so a typo surfaces on the next crawl instead of at some later audit.
+const MEANT_AS_DATETIME = /^\d{4}-\d{2}-\d{2}[T ]\d/;
+export function looksLikeDatetime(value) {
+    return MEANT_AS_DATETIME.test(value);
+}
 // Punctuation only, never a timezone conversion: the offset survives, so substr(d,1,10) is still
 // the local date. A shape that is not a real instant is left as written, and stays auditable.
 export function normalizeDate(value) {
     const m = ISO_DATETIME.exec(value);
     if (m === null) return value;
     const [, date, time, zone] = m;
-    const offset = zone !== undefined && zone !== 'Z' && !zone.includes(':') ? `${zone.slice(0, 3)}:${zone.slice(3)}` : zone !== null && zone !== void 0 ? zone : '';
+    const digits = zone === undefined || zone === 'Z' ? '' : zone.replace(':', '');
+    const offset = digits === '' ? zone !== null && zone !== void 0 ? zone : '' : digits.length === 3 ? `${digits}:00` : `${digits.slice(0, 3)}:${digits.slice(3)}`;
     const normalized = `${date}T${time}${offset}`;
     return Number.isNaN(Date.parse(normalized)) ? value : normalized;
 }
@@ -221,7 +228,11 @@ export function parseFile(file, extractors = []) {
             warnings.push(`warning: ${file.relPath} has a frontmatter key named "${key}", which is reserved; ignoring it`);
             continue;
         }
-        mapped[key] = mapValue(data[key]);
+        const value = mapValue(data[key]);
+        if (typeof value === 'string' && looksLikeDatetime(value) && Number.isNaN(Date.parse(value))) {
+            warnings.push(`warning: ${file.relPath}: ${key} is not a valid date (${value}), so it is invisible to every date comparison`);
+        }
+        mapped[key] = value;
     }
     // title/summary are plain YAML strings -- whitespace-collapse only;
     // the prose gets the full markdown strip.

package/dist/esm/scan.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/scan.ts"],"sourcesContent":["import { globSync, readFileSync, statSync } from 'node:fs';\nimport { join, sep } from 'node:path';\nimport removeMarkdown from 'remove-markdown';\nimport { isCollection, parseDocument, visit } from 'yaml';\nimport type { Config } from './config.ts';\nimport { embedEnabled, presetNames, presetSemanticEnabled } from './config.ts';\nimport type { Feature } from './features/types.ts';\n\n// Filesystem -> rows. Pure data in, data + warnings out; db.ts does the SQL.\n\n// Frontmatter keys that would collide with table columns. Exported so db.ts's upsert can tell\n// a feature-owned column (`_rank`) from a parsed one and leave it alone on reparse.\nexport const RESERVED_COLUMNS = new Set(['path', '_mtime', '_size', '_rank', '_parse_error', 'content', 'links', 'sections']);\n\n// YAML error codes whose recovery is unambiguous, so the parse is accepted rather than\n// quarantined. Only one qualifies: YAML 1.2 reserves `@` and `` ` `` at the start of a plain\n// scalar for future use, so they can never be valid and the text can only be what was typed\n// (`aliases: [@handle]` -> [\"@handle\"]). Every other code has a second reading -- an unquoted\n// `:` swallows the keys after it, an unquoted `[..](..)` drops the URL, a duplicate key picks\n// one value in silence -- so it writes values nobody wrote. See plans/frontmatter-parse-policy.md.\nconst ACCEPTED_YAML_CODES = new Set(['BAD_SCALAR_START']);\n\nfunction normalizeText(value: unknown): string {\n  if (value === null || value === undefined) return '';\n  return String(value).replace(/\\s+/g, ' ').trim();\n}\n\n// Keeps URL query strings, asset filenames, and HTML attributes out of the index: rare terms\n// carry high IDF, so they outrank prose. remove-markdown misses wikilinks and tables.\nfunction stripText(value: string): string {\n  const withoutWikilinks = value.replace(/\\[\\[([^\\]|]+)\\|([^\\]]+)\\]\\]/g, '$2').replace(/\\[\\[([^\\]]+)\\]\\]/g, '$1');\n  const withoutMarkdown = removeMarkdown(withoutWikilinks);\n  const withoutTables = withoutMarkdown.replace(/^\\s*\\|?[-\\s|:]+\\|\\s*$/gm, '').replace(/\\|/g, ' ');\n  return normalizeText(withoutTables);\n}\n\nexport interface FileStat {\n  relPath: string;\n  absPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[]; // every declared preset covering this file (>= 1; union, overlap allowed)\n  embed: boolean; // true iff a model is named and some covering preset has semantic on\n}\n\n// Presets are views, not partitions: they overlap freely, and a file's covering set (not one\n// owner) drives indexing. Globs resolve relative to baseDir; unmatched files are not indexed.\nexport function toPosixPath(relPath: string, separator: string = sep): string {\n  return separator === '\\\\' ? relPath.split(separator).join('/') : relPath;\n}\n\n// Every command pays listFiles before it answers (the freshness check stats each file), so\n// per-file work here is the hottest path in the package. Everything derivable from the config\n// alone is computed once, above the loop.\nconst NO_THROW = { throwIfNoEntry: false } as const;\n\nexport function listFiles(cfg: Config, baseDir: string): FileStat[] {\n  const coverage = new Map<string, Set<string>>();\n  const posixNeeded = sep === '\\\\';\n  for (const name of presetNames(cfg)) {\n    const preset = cfg.presets[name];\n    for (const matched of globSync(preset.include, { cwd: baseDir, exclude: preset.exclude })) {\n      const relPath = posixNeeded ? toPosixPath(matched) : matched;\n      const set = coverage.get(relPath) ?? new Set<string>();\n      set.add(name);\n      coverage.set(relPath, set);\n    }\n  }\n\n  // Which presets want vectors is a property of the config, not of any file.\n  const embedding = embedEnabled(cfg);\n  const semanticPresets = embedding ? new Set(presetNames(cfg).filter((name) => presetSemanticEnabled(cfg, name))) : null;\n\n  const files: FileStat[] = [];\n  for (const relPath of [...coverage.keys()].sort()) {\n    const absPath = join(baseDir, relPath); // join re-applies the platform separator for fs calls\n    // node:fs glob matches directories and dangling symlinks; fast-glob returned neither, so\n    // one stat filters both back out (throwIfNoEntry keeps a dangling link from throwing).\n    const st = statSync(absPath, NO_THROW);\n    if (!st?.isFile()) continue;\n    const presets = [...(coverage.get(relPath) as Set<string>)].sort();\n    const embed = semanticPresets !== null && presets.some((name) => semanticPresets.has(name));\n    files.push({ relPath, absPath, mtimeMs: st.mtimeMs, size: st.size, presets, embed });\n  }\n  return files;\n}\n\nexport interface ParsedDoc {\n  relPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[];\n  data: Record<string, string | number | bigint | null>;\n  // NULL when the frontmatter parsed, the first YAML message otherwise. In the row rather than\n  // a side table so `SELECT *` and any `IS NULL` investigation trip over it without being asked.\n  parseError: string | null;\n  // title/summary are duplicated from frontmatter so bm25() can weight them above the body text.\n  search: { title: string; summary: string; text: string };\n  // Per-feature extraction results, keyed by feature name; features store them at reconcile.\n  extracted: Record<string, unknown>;\n}\n\n// SQLite's datetime() rejects a colonless offset (`-0800`) and a space separator, which ISO 8601\n// allows and producers emit. A rejected date is invisible, not excluded: every comparison is NULL.\nconst ISO_DATETIME = /^(\\d{4}-\\d{2}-\\d{2})[T ](\\d{2}:\\d{2}(?::\\d{2})?(?:\\.\\d+)?)(Z|[+-]\\d{2}:?\\d{2})?$/;\n\n// Punctuation only, never a timezone conversion: the offset survives, so substr(d,1,10) is still\n// the local date. A shape that is not a real instant is left as written, and stays auditable.\nexport function normalizeDate(value: string): string {\n  const m = ISO_DATETIME.exec(value);\n  if (m === null) return value;\n  const [, date, time, zone] = m;\n  const offset = zone !== undefined && zone !== 'Z' && !zone.includes(':') ? `${zone.slice(0, 3)}:${zone.slice(3)}` : (zone ?? '');\n  const normalized = `${date}T${time}${offset}`;\n  return Number.isNaN(Date.parse(normalized)) ? value : normalized;\n}\n\n// Storage class follows the YAML scalar. Booleans store as 1/0, so `WHERE flag = 1` matches\n// and `WHERE flag = 'true'` cannot; `map` prints observed types so the mismatch is visible.\nfunction mapValue(value: unknown): string | number | bigint | null {\n  if (value === null || value === undefined) return null;\n  if (typeof value === 'boolean') return BigInt(value ? 1 : 0);\n  if (typeof value === 'number') return Number.isSafeInteger(value) ? BigInt(value) : value;\n  if (typeof value === 'string') return normalizeDate(value);\n  return JSON.stringify(value);\n}\n\n// The delimiter split is all this package used gray-matter for.\nfunction splitFrontmatter(raw: string): { fm: string | null; body: string } {\n  const open = raw.match(/^---\\r?\\n/);\n  if (!open) return { fm: null, body: raw };\n  const rest = raw.slice(open[0].length);\n  const close = rest.match(/^---\\r?(\\n|$)/m);\n  if (!close || close.index === undefined) return { fm: null, body: raw };\n  return { fm: rest.slice(0, close.index), body: rest.slice(close.index + close[0].length) };\n}\n\n// A well-formed document can still hold a value nobody meant: `created: {{date}}` is valid\n// YAML for a flow map used as a mapping key, so it raises no error and stores\n// {\"{ date }\": null}. No error code can catch that, but yaml notices the stringified key, so\n// this reports it with the path instead (yaml's own warning has none, fires once per document,\n// and is what trains readers to discard stderr).\nfunction warnStringifiedKeys(relPath: string, doc: ReturnType<typeof parseDocument>, warnings: string[]): void {\n  let found = false;\n  // Nested, not top level: `created: {{date}}` puts the collection key one level down, inside\n  // the flow map that `{{...}}` parses as.\n  visit(doc, {\n    Pair(_key, pair) {\n      if (!isCollection(pair.key)) return undefined;\n      found = true;\n      return visit.BREAK;\n    },\n  });\n  // One per file: a template repeats the same mistake on every field it stamps.\n  if (found) warnings.push(`warning: ${relPath} frontmatter has a key that is itself a list or mapping, stored as text; this is usually an unrendered template placeholder like {{date}}`);\n}\n\n// Accept a clean parse, and one whose every error is unambiguous (ACCEPTED_YAML_CODES).\n// Anything else is quarantined: no frontmatter columns at all, and `_parse_error` carries the\n// reason. Recovering it would write values nobody wrote, which is worse than absence because\n// no query can see it. The file is still indexed -- content, links and sections never touch\n// frontmatter -- so a broken note stays searchable while it is being hunted for.\n// yaml's message continues onto a source excerpt, so the first line is the sentence -- minus\n// the colon that introduced the part being dropped.\nfunction firstLine(message: string): string {\n  return message.split('\\n')[0].replace(/:\\s*$/, '');\n}\n\nfunction parseFrontmatter(relPath: string, fm: string, warnings: string[]): { data: Record<string, unknown>; parseError: string | null } {\n  // logLevel silences yaml's own pathless warnings; warnStringifiedKeys re-reports the one\n  // that carries information, with the file it came from.\n  const doc = parseDocument(fm, { logLevel: 'silent' });\n  const refused = doc.errors.filter((err) => !ACCEPTED_YAML_CODES.has(err.code));\n  if (refused.length > 0) {\n    const detail = refused.length > 1 ? ` (and ${refused.length - 1} more)` : '';\n    const parseError = `${firstLine(refused[0].message)}${detail}`;\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  let data: unknown;\n  try {\n    data = doc.toJS();\n  } catch (err) {\n    // Reaches here with doc.errors empty: `title: **Bold**` parses, then opens an alias on\n    // materialisation. An empty error list is not a successful parse.\n    const parseError = firstLine((err as Error).message);\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  if (data === null || data === undefined) return { data: {}, parseError: null };\n  if (typeof data !== 'object' || Array.isArray(data)) {\n    const parseError = 'frontmatter is not a key-value mapping';\n    warnings.push(`warning: ${relPath} ${parseError}; none of it is indexed`);\n    return { data: {}, parseError };\n  }\n  warnStringifiedKeys(relPath, doc, warnings);\n  return { data: data as Record<string, unknown>, parseError: null };\n}\n\nexport function parseFile(file: FileStat, extractors: Feature[] = []): { doc: ParsedDoc; warnings: string[] } {\n  const raw = readFileSync(file.absPath, 'utf8');\n  const warnings: string[] = [];\n\n  const { fm, body: content } = splitFrontmatter(raw);\n  const { data, parseError } = fm === null ? { data: {} as Record<string, unknown>, parseError: null } : parseFrontmatter(file.relPath, fm, warnings);\n  const mapped: Record<string, string | number | bigint | null> = {};\n\n  for (const key of Object.keys(data)) {\n    if (RESERVED_COLUMNS.has(key)) {\n      warnings.push(`warning: ${file.relPath} has a frontmatter key named \"${key}\", which is reserved; ignoring it`);\n      continue;\n    }\n    mapped[key] = mapValue(data[key]);\n  }\n\n  // title/summary are plain YAML strings -- whitespace-collapse only;\n  // the prose gets the full markdown strip.\n  const search = { title: normalizeText(data.title), summary: normalizeText(data.summary), text: stripText(content) };\n\n  return {\n    doc: {\n      relPath: file.relPath,\n      mtimeMs: file.mtimeMs,\n      size: file.size,\n      presets: file.presets,\n      data: mapped,\n      parseError,\n      search,\n      extracted: Object.fromEntries(extractors.filter((f) => f.extract).map((f) => [f.name, f.extract?.(raw, content, search)])),\n    },\n    warnings,\n  };\n}\n"],"names":["globSync","readFileSync","statSync","join","sep","removeMarkdown","isCollection","parseDocument","visit","embedEnabled","presetNames","presetSemanticEnabled","RESERVED_COLUMNS","Set","ACCEPTED_YAML_CODES","normalizeText","value","undefined","String","replace","trim","stripText","withoutWikilinks","withoutMarkdown","withoutTables","toPosixPath","relPath","separator","split","NO_THROW","throwIfNoEntry","listFiles","cfg","baseDir","coverage","Map","posixNeeded","name","preset","presets","matched","include","cwd","exclude","set","get","add","embedding","semanticPresets","filter","files","keys","sort","absPath","st","isFile","embed","some","has","push","mtimeMs","size","ISO_DATETIME","normalizeDate","m","exec","date","time","zone","offset","includes","slice","normalized","Number","isNaN","Date","parse","mapValue","BigInt","isSafeInteger","JSON","stringify","splitFrontmatter","raw","open","match","fm","body","rest","length","close","index","warnStringifiedKeys","doc","warnings","found","Pair","_key","pair","key","BREAK","firstLine","message","parseFrontmatter","logLevel","refused","errors","err","code","detail","parseError","data","toJS","Array","isArray","parseFile","file","extractors","content","mapped","Object","search","title","summary","text","extracted","fromEntries","f","extract","map"],"mappings":"AAAA,SAASA,QAAQ,EAAEC,YAAY,EAAEC,QAAQ,QAAQ,UAAU;AAC3D,SAASC,IAAI,EAAEC,GAAG,QAAQ,YAAY;AACtC,OAAOC,oBAAoB,kBAAkB;AAC7C,SAASC,YAAY,EAAEC,aAAa,EAAEC,KAAK,QAAQ,OAAO;AAE1D,SAASC,YAAY,EAAEC,WAAW,EAAEC,qBAAqB,QAAQ,cAAc;AAG/E,6EAA6E;AAE7E,8FAA8F;AAC9F,oFAAoF;AACpF,OAAO,MAAMC,mBAAmB,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAS;IAAS;IAAgB;IAAW;IAAS;CAAW,EAAE;AAE9H,uFAAuF;AACvF,6FAA6F;AAC7F,4FAA4F;AAC5F,8FAA8F;AAC9F,8FAA8F;AAC9F,mGAAmG;AACnG,MAAMC,sBAAsB,IAAID,IAAI;IAAC;CAAmB;AAExD,SAASE,cAAcC,KAAc;IACnC,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,OAAOC,OAAOF,OAAOG,OAAO,CAAC,QAAQ,KAAKC,IAAI;AAChD;AAEA,6FAA6F;AAC7F,sFAAsF;AACtF,SAASC,UAAUL,KAAa;IAC9B,MAAMM,mBAAmBN,MAAMG,OAAO,CAAC,gCAAgC,MAAMA,OAAO,CAAC,qBAAqB;IAC1G,MAAMI,kBAAkBlB,eAAeiB;IACvC,MAAME,gBAAgBD,gBAAgBJ,OAAO,CAAC,2BAA2B,IAAIA,OAAO,CAAC,OAAO;IAC5F,OAAOJ,cAAcS;AACvB;AAWA,6FAA6F;AAC7F,8FAA8F;AAC9F,OAAO,SAASC,YAAYC,OAAe,EAAEC,YAAoBvB,GAAG;IAClE,OAAOuB,cAAc,OAAOD,QAAQE,KAAK,CAACD,WAAWxB,IAAI,CAAC,OAAOuB;AACnE;AAEA,2FAA2F;AAC3F,8FAA8F;AAC9F,0CAA0C;AAC1C,MAAMG,WAAW;IAAEC,gBAAgB;AAAM;AAEzC,OAAO,SAASC,UAAUC,GAAW,EAAEC,OAAe;IACpD,MAAMC,WAAW,IAAIC;IACrB,MAAMC,cAAchC,QAAQ;IAC5B,KAAK,MAAMiC,QAAQ3B,YAAYsB,KAAM;QACnC,MAAMM,SAASN,IAAIO,OAAO,CAACF,KAAK;QAChC,KAAK,MAAMG,WAAWxC,SAASsC,OAAOG,OAAO,EAAE;YAAEC,KAAKT;YAASU,SAASL,OAAOK,OAAO;QAAC,GAAI;gBAE7ET;YADZ,MAAMR,UAAUU,cAAcX,YAAYe,WAAWA;YACrD,MAAMI,OAAMV,gBAAAA,SAASW,GAAG,CAACnB,sBAAbQ,2BAAAA,gBAAyB,IAAIrB;YACzC+B,IAAIE,GAAG,CAACT;YACRH,SAASU,GAAG,CAAClB,SAASkB;QACxB;IACF;IAEA,2EAA2E;IAC3E,MAAMG,YAAYtC,aAAauB;IAC/B,MAAMgB,kBAAkBD,YAAY,IAAIlC,IAAIH,YAAYsB,KAAKiB,MAAM,CAAC,CAACZ,OAAS1B,sBAAsBqB,KAAKK,UAAU;IAEnH,MAAMa,QAAoB,EAAE;IAC5B,KAAK,MAAMxB,WAAW;WAAIQ,SAASiB,IAAI;KAAG,CAACC,IAAI,GAAI;QACjD,MAAMC,UAAUlD,KAAK8B,SAASP,UAAU,sDAAsD;QAC9F,yFAAyF;QACzF,uFAAuF;QACvF,MAAM4B,KAAKpD,SAASmD,SAASxB;QAC7B,IAAI,EAACyB,eAAAA,yBAAAA,GAAIC,MAAM,KAAI;QACnB,MAAMhB,UAAU;eAAKL,SAASW,GAAG,CAACnB;SAAyB,CAAC0B,IAAI;QAChE,MAAMI,QAAQR,oBAAoB,QAAQT,QAAQkB,IAAI,CAAC,CAACpB,OAASW,gBAAgBU,GAAG,CAACrB;QACrFa,MAAMS,IAAI,CAAC;YAAEjC;YAAS2B;YAASO,SAASN,GAAGM,OAAO;YAAEC,MAAMP,GAAGO,IAAI;YAAEtB;YAASiB;QAAM;IACpF;IACA,OAAON;AACT;AAiBA,iGAAiG;AACjG,mGAAmG;AACnG,MAAMY,eAAe;AAErB,iGAAiG;AACjG,8FAA8F;AAC9F,OAAO,SAASC,cAAc/C,KAAa;IACzC,MAAMgD,IAAIF,aAAaG,IAAI,CAACjD;IAC5B,IAAIgD,MAAM,MAAM,OAAOhD;IACvB,MAAM,GAAGkD,MAAMC,MAAMC,KAAK,GAAGJ;IAC7B,MAAMK,SAASD,SAASnD,aAAamD,SAAS,OAAO,CAACA,KAAKE,QAAQ,CAAC,OAAO,GAAGF,KAAKG,KAAK,CAAC,GAAG,GAAG,CAAC,EAAEH,KAAKG,KAAK,CAAC,IAAI,GAAIH,iBAAAA,kBAAAA,OAAQ;IAC7H,MAAMI,aAAa,GAAGN,KAAK,CAAC,EAAEC,OAAOE,QAAQ;IAC7C,OAAOI,OAAOC,KAAK,CAACC,KAAKC,KAAK,CAACJ,eAAexD,QAAQwD;AACxD;AAEA,4FAA4F;AAC5F,4FAA4F;AAC5F,SAASK,SAAS7D,KAAc;IAC9B,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,IAAI,OAAOD,UAAU,WAAW,OAAO8D,OAAO9D,QAAQ,IAAI;IAC1D,IAAI,OAAOA,UAAU,UAAU,OAAOyD,OAAOM,aAAa,CAAC/D,SAAS8D,OAAO9D,SAASA;IACpF,IAAI,OAAOA,UAAU,UAAU,OAAO+C,cAAc/C;IACpD,OAAOgE,KAAKC,SAAS,CAACjE;AACxB;AAEA,gEAAgE;AAChE,SAASkE,iBAAiBC,GAAW;IACnC,MAAMC,OAAOD,IAAIE,KAAK,CAAC;IACvB,IAAI,CAACD,MAAM,OAAO;QAAEE,IAAI;QAAMC,MAAMJ;IAAI;IACxC,MAAMK,OAAOL,IAAIZ,KAAK,CAACa,IAAI,CAAC,EAAE,CAACK,MAAM;IACrC,MAAMC,QAAQF,KAAKH,KAAK,CAAC;IACzB,IAAI,CAACK,SAASA,MAAMC,KAAK,KAAK1E,WAAW,OAAO;QAAEqE,IAAI;QAAMC,MAAMJ;IAAI;IACtE,OAAO;QAAEG,IAAIE,KAAKjB,KAAK,CAAC,GAAGmB,MAAMC,KAAK;QAAGJ,MAAMC,KAAKjB,KAAK,CAACmB,MAAMC,KAAK,GAAGD,KAAK,CAAC,EAAE,CAACD,MAAM;IAAE;AAC3F;AAEA,2FAA2F;AAC3F,8EAA8E;AAC9E,6FAA6F;AAC7F,+FAA+F;AAC/F,iDAAiD;AACjD,SAASG,oBAAoBlE,OAAe,EAAEmE,GAAqC,EAAEC,QAAkB;IACrG,IAAIC,QAAQ;IACZ,4FAA4F;IAC5F,yCAAyC;IACzCvF,MAAMqF,KAAK;QACTG,MAAKC,IAAI,EAAEC,IAAI;YACb,IAAI,CAAC5F,aAAa4F,KAAKC,GAAG,GAAG,OAAOlF;YACpC8E,QAAQ;YACR,OAAOvF,MAAM4F,KAAK;QACpB;IACF;IACA,8EAA8E;IAC9E,IAAIL,OAAOD,SAASnC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,yIAAyI,CAAC;AACzL;AAEA,wFAAwF;AACxF,8FAA8F;AAC9F,6FAA6F;AAC7F,4FAA4F;AAC5F,iFAAiF;AACjF,6FAA6F;AAC7F,oDAAoD;AACpD,SAAS2E,UAAUC,OAAe;IAChC,OAAOA,QAAQ1E,KAAK,CAAC,KAAK,CAAC,EAAE,CAACT,OAAO,CAAC,SAAS;AACjD;AAEA,SAASoF,iBAAiB7E,OAAe,EAAE4D,EAAU,EAAEQ,QAAkB;IACvE,yFAAyF;IACzF,wDAAwD;IACxD,MAAMD,MAAMtF,cAAc+E,IAAI;QAAEkB,UAAU;IAAS;IACnD,MAAMC,UAAUZ,IAAIa,MAAM,CAACzD,MAAM,CAAC,CAAC0D,MAAQ,CAAC7F,oBAAoB4C,GAAG,CAACiD,IAAIC,IAAI;IAC5E,IAAIH,QAAQhB,MAAM,GAAG,GAAG;QACtB,MAAMoB,SAASJ,QAAQhB,MAAM,GAAG,IAAI,CAAC,MAAM,EAAEgB,QAAQhB,MAAM,GAAG,EAAE,MAAM,CAAC,GAAG;QAC1E,MAAMqB,aAAa,GAAGT,UAAUI,OAAO,CAAC,EAAE,CAACH,OAAO,IAAIO,QAAQ;QAC9Df,SAASnC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,sDAAsD,EAAEoF,YAAY;QACtG,OAAO;YAAEC,MAAM,CAAC;YAAGD;QAAW;IAChC;IAEA,IAAIC;IACJ,IAAI;QACFA,OAAOlB,IAAImB,IAAI;IACjB,EAAE,OAAOL,KAAK;QACZ,uFAAuF;QACvF,kEAAkE;QAClE,MAAMG,aAAaT,UAAU,AAACM,IAAcL,OAAO;QACnDR,SAASnC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,sDAAsD,EAAEoF,YAAY;QACtG,OAAO;YAAEC,MAAM,CAAC;YAAGD;QAAW;IAChC;IAEA,IAAIC,SAAS,QAAQA,SAAS9F,WAAW,OAAO;QAAE8F,MAAM,CAAC;QAAGD,YAAY;IAAK;IAC7E,IAAI,OAAOC,SAAS,YAAYE,MAAMC,OAAO,CAACH,OAAO;QACnD,MAAMD,aAAa;QACnBhB,SAASnC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,CAAC,EAAEoF,WAAW,uBAAuB,CAAC;QACxE,OAAO;YAAEC,MAAM,CAAC;YAAGD;QAAW;IAChC;IACAlB,oBAAoBlE,SAASmE,KAAKC;IAClC,OAAO;QAAEiB,MAAMA;QAAiCD,YAAY;IAAK;AACnE;AAEA,OAAO,SAASK,UAAUC,IAAc,EAAEC,aAAwB,EAAE;IAClE,MAAMlC,MAAMlF,aAAamH,KAAK/D,OAAO,EAAE;IACvC,MAAMyC,WAAqB,EAAE;IAE7B,MAAM,EAAER,EAAE,EAAEC,MAAM+B,OAAO,EAAE,GAAGpC,iBAAiBC;IAC/C,MAAM,EAAE4B,IAAI,EAAED,UAAU,EAAE,GAAGxB,OAAO,OAAO;QAAEyB,MAAM,CAAC;QAA8BD,YAAY;IAAK,IAAIP,iBAAiBa,KAAK1F,OAAO,EAAE4D,IAAIQ;IAC1I,MAAMyB,SAA0D,CAAC;IAEjE,KAAK,MAAMpB,OAAOqB,OAAOrE,IAAI,CAAC4D,MAAO;QACnC,IAAInG,iBAAiB8C,GAAG,CAACyC,MAAM;YAC7BL,SAASnC,IAAI,CAAC,CAAC,SAAS,EAAEyD,KAAK1F,OAAO,CAAC,8BAA8B,EAAEyE,IAAI,iCAAiC,CAAC;YAC7G;QACF;QACAoB,MAAM,CAACpB,IAAI,GAAGtB,SAASkC,IAAI,CAACZ,IAAI;IAClC;IAEA,oEAAoE;IACpE,0CAA0C;IAC1C,MAAMsB,SAAS;QAAEC,OAAO3G,cAAcgG,KAAKW,KAAK;QAAGC,SAAS5G,cAAcgG,KAAKY,OAAO;QAAGC,MAAMvG,UAAUiG;IAAS;IAElH,OAAO;QACLzB,KAAK;YACHnE,SAAS0F,KAAK1F,OAAO;YACrBkC,SAASwD,KAAKxD,OAAO;YACrBC,MAAMuD,KAAKvD,IAAI;YACftB,SAAS6E,KAAK7E,OAAO;YACrBwE,MAAMQ;YACNT;YACAW;YACAI,WAAWL,OAAOM,WAAW,CAACT,WAAWpE,MAAM,CAAC,CAAC8E,IAAMA,EAAEC,OAAO,EAAEC,GAAG,CAAC,CAACF;oBAAeA;uBAAT;oBAACA,EAAE1F,IAAI;qBAAE0F,aAAAA,EAAEC,OAAO,cAATD,iCAAAA,gBAAAA,GAAY5C,KAAKmC,SAASG;iBAAQ;;QAC1H;QACA3B;IACF;AACF"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/scan.ts"],"sourcesContent":["import { globSync, readFileSync, statSync } from 'node:fs';\nimport { join, sep } from 'node:path';\nimport removeMarkdown from 'remove-markdown';\nimport { isCollection, parseDocument, visit } from 'yaml';\nimport type { Config } from './config.ts';\nimport { embedEnabled, presetNames, presetSemanticEnabled } from './config.ts';\nimport type { Feature } from './features/types.ts';\n\n// Filesystem -> rows. Pure data in, data + warnings out; db.ts does the SQL.\n\n// Frontmatter keys that would collide with table columns. Exported so db.ts's upsert can tell\n// a feature-owned column (`_rank`) from a parsed one and leave it alone on reparse.\nexport const RESERVED_COLUMNS = new Set(['path', '_mtime', '_size', '_rank', '_parse_error', 'content', 'links', 'sections']);\n\n// YAML error codes whose recovery is unambiguous, so the parse is accepted rather than\n// quarantined. Only one qualifies: YAML 1.2 reserves `@` and `` ` `` at the start of a plain\n// scalar for future use, so they can never be valid and the text can only be what was typed\n// (`aliases: [@handle]` -> [\"@handle\"]). Every other code has a second reading -- an unquoted\n// `:` swallows the keys after it, an unquoted `[..](..)` drops the URL, a duplicate key picks\n// one value in silence -- so it writes values nobody wrote. See plans/frontmatter-parse-policy.md.\nconst ACCEPTED_YAML_CODES = new Set(['BAD_SCALAR_START']);\n\nfunction normalizeText(value: unknown): string {\n  if (value === null || value === undefined) return '';\n  return String(value).replace(/\\s+/g, ' ').trim();\n}\n\n// Keeps URL query strings, asset filenames, and HTML attributes out of the index: rare terms\n// carry high IDF, so they outrank prose. remove-markdown misses wikilinks and tables.\nfunction stripText(value: string): string {\n  const withoutWikilinks = value.replace(/\\[\\[([^\\]|]+)\\|([^\\]]+)\\]\\]/g, '$2').replace(/\\[\\[([^\\]]+)\\]\\]/g, '$1');\n  const withoutMarkdown = removeMarkdown(withoutWikilinks);\n  const withoutTables = withoutMarkdown.replace(/^\\s*\\|?[-\\s|:]+\\|\\s*$/gm, '').replace(/\\|/g, ' ');\n  return normalizeText(withoutTables);\n}\n\nexport interface FileStat {\n  relPath: string;\n  absPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[]; // every declared preset covering this file (>= 1; union, overlap allowed)\n  embed: boolean; // true iff a model is named and some covering preset has semantic on\n}\n\n// Presets are views, not partitions: they overlap freely, and a file's covering set (not one\n// owner) drives indexing. Globs resolve relative to baseDir; unmatched files are not indexed.\nexport function toPosixPath(relPath: string, separator: string = sep): string {\n  return separator === '\\\\' ? relPath.split(separator).join('/') : relPath;\n}\n\n// Every command pays listFiles before it answers (the freshness check stats each file), so\n// per-file work here is the hottest path in the package. Everything derivable from the config\n// alone is computed once, above the loop.\nconst NO_THROW = { throwIfNoEntry: false } as const;\n\nexport function listFiles(cfg: Config, baseDir: string): FileStat[] {\n  const coverage = new Map<string, Set<string>>();\n  const posixNeeded = sep === '\\\\';\n  for (const name of presetNames(cfg)) {\n    const preset = cfg.presets[name];\n    for (const matched of globSync(preset.include, { cwd: baseDir, exclude: preset.exclude })) {\n      const relPath = posixNeeded ? toPosixPath(matched) : matched;\n      const set = coverage.get(relPath) ?? new Set<string>();\n      set.add(name);\n      coverage.set(relPath, set);\n    }\n  }\n\n  // Which presets want vectors is a property of the config, not of any file.\n  const embedding = embedEnabled(cfg);\n  const semanticPresets = embedding ? new Set(presetNames(cfg).filter((name) => presetSemanticEnabled(cfg, name))) : null;\n\n  const files: FileStat[] = [];\n  for (const relPath of [...coverage.keys()].sort()) {\n    const absPath = join(baseDir, relPath); // join re-applies the platform separator for fs calls\n    // node:fs glob matches directories and dangling symlinks; fast-glob returned neither, so\n    // one stat filters both back out (throwIfNoEntry keeps a dangling link from throwing).\n    const st = statSync(absPath, NO_THROW);\n    if (!st?.isFile()) continue;\n    const presets = [...(coverage.get(relPath) as Set<string>)].sort();\n    const embed = semanticPresets !== null && presets.some((name) => semanticPresets.has(name));\n    files.push({ relPath, absPath, mtimeMs: st.mtimeMs, size: st.size, presets, embed });\n  }\n  return files;\n}\n\nexport interface ParsedDoc {\n  relPath: string;\n  mtimeMs: number;\n  size: number;\n  presets: string[];\n  data: Record<string, string | number | bigint | null>;\n  // NULL when the frontmatter parsed, the first YAML message otherwise. In the row rather than\n  // a side table so `SELECT *` and any `IS NULL` investigation trip over it without being asked.\n  parseError: string | null;\n  // title/summary are duplicated from frontmatter so bm25() can weight them above the body text.\n  search: { title: string; summary: string; text: string };\n  // Per-feature extraction results, keyed by feature name; features store them at reconcile.\n  extracted: Record<string, unknown>;\n}\n\n// SQLite's datetime() rejects a colonless offset (`-0800`) and a space separator, which ISO 8601\n// allows and producers emit. A rejected date is invisible, not excluded: every comparison is NULL.\nconst ISO_DATETIME = /^(\\d{4}-\\d{2}-\\d{2})[T ](\\d{2}:\\d{2}(?::\\d{2})?(?:\\.\\d+)?)(Z|[+-]\\d{2}(?::?\\d{2})?)?$/;\n\n// A value opening `YYYY-MM-DDT` was meant to be a datetime; prose never is. Reported when it\n// cannot be normalized, so a typo surfaces on the next crawl instead of at some later audit.\nconst MEANT_AS_DATETIME = /^\\d{4}-\\d{2}-\\d{2}[T ]\\d/;\n\nexport function looksLikeDatetime(value: string): boolean {\n  return MEANT_AS_DATETIME.test(value);\n}\n\n// Punctuation only, never a timezone conversion: the offset survives, so substr(d,1,10) is still\n// the local date. A shape that is not a real instant is left as written, and stays auditable.\nexport function normalizeDate(value: string): string {\n  const m = ISO_DATETIME.exec(value);\n  if (m === null) return value;\n  const [, date, time, zone] = m;\n  const digits = zone === undefined || zone === 'Z' ? '' : zone.replace(':', '');\n  const offset = digits === '' ? (zone ?? '') : digits.length === 3 ? `${digits}:00` : `${digits.slice(0, 3)}:${digits.slice(3)}`;\n  const normalized = `${date}T${time}${offset}`;\n  return Number.isNaN(Date.parse(normalized)) ? value : normalized;\n}\n\n// Storage class follows the YAML scalar. Booleans store as 1/0, so `WHERE flag = 1` matches\n// and `WHERE flag = 'true'` cannot; `map` prints observed types so the mismatch is visible.\nfunction mapValue(value: unknown): string | number | bigint | null {\n  if (value === null || value === undefined) return null;\n  if (typeof value === 'boolean') return BigInt(value ? 1 : 0);\n  if (typeof value === 'number') return Number.isSafeInteger(value) ? BigInt(value) : value;\n  if (typeof value === 'string') return normalizeDate(value);\n  return JSON.stringify(value);\n}\n\n// The delimiter split is all this package used gray-matter for.\nfunction splitFrontmatter(raw: string): { fm: string | null; body: string } {\n  const open = raw.match(/^---\\r?\\n/);\n  if (!open) return { fm: null, body: raw };\n  const rest = raw.slice(open[0].length);\n  const close = rest.match(/^---\\r?(\\n|$)/m);\n  if (!close || close.index === undefined) return { fm: null, body: raw };\n  return { fm: rest.slice(0, close.index), body: rest.slice(close.index + close[0].length) };\n}\n\n// A well-formed document can still hold a value nobody meant: `created: {{date}}` is valid\n// YAML for a flow map used as a mapping key, so it raises no error and stores\n// {\"{ date }\": null}. No error code can catch that, but yaml notices the stringified key, so\n// this reports it with the path instead (yaml's own warning has none, fires once per document,\n// and is what trains readers to discard stderr).\nfunction warnStringifiedKeys(relPath: string, doc: ReturnType<typeof parseDocument>, warnings: string[]): void {\n  let found = false;\n  // Nested, not top level: `created: {{date}}` puts the collection key one level down, inside\n  // the flow map that `{{...}}` parses as.\n  visit(doc, {\n    Pair(_key, pair) {\n      if (!isCollection(pair.key)) return undefined;\n      found = true;\n      return visit.BREAK;\n    },\n  });\n  // One per file: a template repeats the same mistake on every field it stamps.\n  if (found) warnings.push(`warning: ${relPath} frontmatter has a key that is itself a list or mapping, stored as text; this is usually an unrendered template placeholder like {{date}}`);\n}\n\n// Accept a clean parse, and one whose every error is unambiguous (ACCEPTED_YAML_CODES).\n// Anything else is quarantined: no frontmatter columns at all, and `_parse_error` carries the\n// reason. Recovering it would write values nobody wrote, which is worse than absence because\n// no query can see it. The file is still indexed -- content, links and sections never touch\n// frontmatter -- so a broken note stays searchable while it is being hunted for.\n// yaml's message continues onto a source excerpt, so the first line is the sentence -- minus\n// the colon that introduced the part being dropped.\nfunction firstLine(message: string): string {\n  return message.split('\\n')[0].replace(/:\\s*$/, '');\n}\n\nfunction parseFrontmatter(relPath: string, fm: string, warnings: string[]): { data: Record<string, unknown>; parseError: string | null } {\n  // logLevel silences yaml's own pathless warnings; warnStringifiedKeys re-reports the one\n  // that carries information, with the file it came from.\n  const doc = parseDocument(fm, { logLevel: 'silent' });\n  const refused = doc.errors.filter((err) => !ACCEPTED_YAML_CODES.has(err.code));\n  if (refused.length > 0) {\n    const detail = refused.length > 1 ? ` (and ${refused.length - 1} more)` : '';\n    const parseError = `${firstLine(refused[0].message)}${detail}`;\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  let data: unknown;\n  try {\n    data = doc.toJS();\n  } catch (err) {\n    // Reaches here with doc.errors empty: `title: **Bold**` parses, then opens an alias on\n    // materialisation. An empty error list is not a successful parse.\n    const parseError = firstLine((err as Error).message);\n    warnings.push(`warning: ${relPath} frontmatter did not parse, so none of it is indexed: ${parseError}`);\n    return { data: {}, parseError };\n  }\n\n  if (data === null || data === undefined) return { data: {}, parseError: null };\n  if (typeof data !== 'object' || Array.isArray(data)) {\n    const parseError = 'frontmatter is not a key-value mapping';\n    warnings.push(`warning: ${relPath} ${parseError}; none of it is indexed`);\n    return { data: {}, parseError };\n  }\n  warnStringifiedKeys(relPath, doc, warnings);\n  return { data: data as Record<string, unknown>, parseError: null };\n}\n\nexport function parseFile(file: FileStat, extractors: Feature[] = []): { doc: ParsedDoc; warnings: string[] } {\n  const raw = readFileSync(file.absPath, 'utf8');\n  const warnings: string[] = [];\n\n  const { fm, body: content } = splitFrontmatter(raw);\n  const { data, parseError } = fm === null ? { data: {} as Record<string, unknown>, parseError: null } : parseFrontmatter(file.relPath, fm, warnings);\n  const mapped: Record<string, string | number | bigint | null> = {};\n\n  for (const key of Object.keys(data)) {\n    if (RESERVED_COLUMNS.has(key)) {\n      warnings.push(`warning: ${file.relPath} has a frontmatter key named \"${key}\", which is reserved; ignoring it`);\n      continue;\n    }\n    const value = mapValue(data[key]);\n    if (typeof value === 'string' && looksLikeDatetime(value) && Number.isNaN(Date.parse(value))) {\n      warnings.push(`warning: ${file.relPath}: ${key} is not a valid date (${value}), so it is invisible to every date comparison`);\n    }\n    mapped[key] = value;\n  }\n\n  // title/summary are plain YAML strings -- whitespace-collapse only;\n  // the prose gets the full markdown strip.\n  const search = { title: normalizeText(data.title), summary: normalizeText(data.summary), text: stripText(content) };\n\n  return {\n    doc: {\n      relPath: file.relPath,\n      mtimeMs: file.mtimeMs,\n      size: file.size,\n      presets: file.presets,\n      data: mapped,\n      parseError,\n      search,\n      extracted: Object.fromEntries(extractors.filter((f) => f.extract).map((f) => [f.name, f.extract?.(raw, content, search)])),\n    },\n    warnings,\n  };\n}\n"],"names":["globSync","readFileSync","statSync","join","sep","removeMarkdown","isCollection","parseDocument","visit","embedEnabled","presetNames","presetSemanticEnabled","RESERVED_COLUMNS","Set","ACCEPTED_YAML_CODES","normalizeText","value","undefined","String","replace","trim","stripText","withoutWikilinks","withoutMarkdown","withoutTables","toPosixPath","relPath","separator","split","NO_THROW","throwIfNoEntry","listFiles","cfg","baseDir","coverage","Map","posixNeeded","name","preset","presets","matched","include","cwd","exclude","set","get","add","embedding","semanticPresets","filter","files","keys","sort","absPath","st","isFile","embed","some","has","push","mtimeMs","size","ISO_DATETIME","MEANT_AS_DATETIME","looksLikeDatetime","test","normalizeDate","m","exec","date","time","zone","digits","offset","length","slice","normalized","Number","isNaN","Date","parse","mapValue","BigInt","isSafeInteger","JSON","stringify","splitFrontmatter","raw","open","match","fm","body","rest","close","index","warnStringifiedKeys","doc","warnings","found","Pair","_key","pair","key","BREAK","firstLine","message","parseFrontmatter","logLevel","refused","errors","err","code","detail","parseError","data","toJS","Array","isArray","parseFile","file","extractors","content","mapped","Object","search","title","summary","text","extracted","fromEntries","f","extract","map"],"mappings":"AAAA,SAASA,QAAQ,EAAEC,YAAY,EAAEC,QAAQ,QAAQ,UAAU;AAC3D,SAASC,IAAI,EAAEC,GAAG,QAAQ,YAAY;AACtC,OAAOC,oBAAoB,kBAAkB;AAC7C,SAASC,YAAY,EAAEC,aAAa,EAAEC,KAAK,QAAQ,OAAO;AAE1D,SAASC,YAAY,EAAEC,WAAW,EAAEC,qBAAqB,QAAQ,cAAc;AAG/E,6EAA6E;AAE7E,8FAA8F;AAC9F,oFAAoF;AACpF,OAAO,MAAMC,mBAAmB,IAAIC,IAAI;IAAC;IAAQ;IAAU;IAAS;IAAS;IAAgB;IAAW;IAAS;CAAW,EAAE;AAE9H,uFAAuF;AACvF,6FAA6F;AAC7F,4FAA4F;AAC5F,8FAA8F;AAC9F,8FAA8F;AAC9F,mGAAmG;AACnG,MAAMC,sBAAsB,IAAID,IAAI;IAAC;CAAmB;AAExD,SAASE,cAAcC,KAAc;IACnC,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,OAAOC,OAAOF,OAAOG,OAAO,CAAC,QAAQ,KAAKC,IAAI;AAChD;AAEA,6FAA6F;AAC7F,sFAAsF;AACtF,SAASC,UAAUL,KAAa;IAC9B,MAAMM,mBAAmBN,MAAMG,OAAO,CAAC,gCAAgC,MAAMA,OAAO,CAAC,qBAAqB;IAC1G,MAAMI,kBAAkBlB,eAAeiB;IACvC,MAAME,gBAAgBD,gBAAgBJ,OAAO,CAAC,2BAA2B,IAAIA,OAAO,CAAC,OAAO;IAC5F,OAAOJ,cAAcS;AACvB;AAWA,6FAA6F;AAC7F,8FAA8F;AAC9F,OAAO,SAASC,YAAYC,OAAe,EAAEC,YAAoBvB,GAAG;IAClE,OAAOuB,cAAc,OAAOD,QAAQE,KAAK,CAACD,WAAWxB,IAAI,CAAC,OAAOuB;AACnE;AAEA,2FAA2F;AAC3F,8FAA8F;AAC9F,0CAA0C;AAC1C,MAAMG,WAAW;IAAEC,gBAAgB;AAAM;AAEzC,OAAO,SAASC,UAAUC,GAAW,EAAEC,OAAe;IACpD,MAAMC,WAAW,IAAIC;IACrB,MAAMC,cAAchC,QAAQ;IAC5B,KAAK,MAAMiC,QAAQ3B,YAAYsB,KAAM;QACnC,MAAMM,SAASN,IAAIO,OAAO,CAACF,KAAK;QAChC,KAAK,MAAMG,WAAWxC,SAASsC,OAAOG,OAAO,EAAE;YAAEC,KAAKT;YAASU,SAASL,OAAOK,OAAO;QAAC,GAAI;gBAE7ET;YADZ,MAAMR,UAAUU,cAAcX,YAAYe,WAAWA;YACrD,MAAMI,OAAMV,gBAAAA,SAASW,GAAG,CAACnB,sBAAbQ,2BAAAA,gBAAyB,IAAIrB;YACzC+B,IAAIE,GAAG,CAACT;YACRH,SAASU,GAAG,CAAClB,SAASkB;QACxB;IACF;IAEA,2EAA2E;IAC3E,MAAMG,YAAYtC,aAAauB;IAC/B,MAAMgB,kBAAkBD,YAAY,IAAIlC,IAAIH,YAAYsB,KAAKiB,MAAM,CAAC,CAACZ,OAAS1B,sBAAsBqB,KAAKK,UAAU;IAEnH,MAAMa,QAAoB,EAAE;IAC5B,KAAK,MAAMxB,WAAW;WAAIQ,SAASiB,IAAI;KAAG,CAACC,IAAI,GAAI;QACjD,MAAMC,UAAUlD,KAAK8B,SAASP,UAAU,sDAAsD;QAC9F,yFAAyF;QACzF,uFAAuF;QACvF,MAAM4B,KAAKpD,SAASmD,SAASxB;QAC7B,IAAI,EAACyB,eAAAA,yBAAAA,GAAIC,MAAM,KAAI;QACnB,MAAMhB,UAAU;eAAKL,SAASW,GAAG,CAACnB;SAAyB,CAAC0B,IAAI;QAChE,MAAMI,QAAQR,oBAAoB,QAAQT,QAAQkB,IAAI,CAAC,CAACpB,OAASW,gBAAgBU,GAAG,CAACrB;QACrFa,MAAMS,IAAI,CAAC;YAAEjC;YAAS2B;YAASO,SAASN,GAAGM,OAAO;YAAEC,MAAMP,GAAGO,IAAI;YAAEtB;YAASiB;QAAM;IACpF;IACA,OAAON;AACT;AAiBA,iGAAiG;AACjG,mGAAmG;AACnG,MAAMY,eAAe;AAErB,6FAA6F;AAC7F,6FAA6F;AAC7F,MAAMC,oBAAoB;AAE1B,OAAO,SAASC,kBAAkBhD,KAAa;IAC7C,OAAO+C,kBAAkBE,IAAI,CAACjD;AAChC;AAEA,iGAAiG;AACjG,8FAA8F;AAC9F,OAAO,SAASkD,cAAclD,KAAa;IACzC,MAAMmD,IAAIL,aAAaM,IAAI,CAACpD;IAC5B,IAAImD,MAAM,MAAM,OAAOnD;IACvB,MAAM,GAAGqD,MAAMC,MAAMC,KAAK,GAAGJ;IAC7B,MAAMK,SAASD,SAAStD,aAAasD,SAAS,MAAM,KAAKA,KAAKpD,OAAO,CAAC,KAAK;IAC3E,MAAMsD,SAASD,WAAW,KAAMD,iBAAAA,kBAAAA,OAAQ,KAAMC,OAAOE,MAAM,KAAK,IAAI,GAAGF,OAAO,GAAG,CAAC,GAAG,GAAGA,OAAOG,KAAK,CAAC,GAAG,GAAG,CAAC,EAAEH,OAAOG,KAAK,CAAC,IAAI;IAC/H,MAAMC,aAAa,GAAGP,KAAK,CAAC,EAAEC,OAAOG,QAAQ;IAC7C,OAAOI,OAAOC,KAAK,CAACC,KAAKC,KAAK,CAACJ,eAAe5D,QAAQ4D;AACxD;AAEA,4FAA4F;AAC5F,4FAA4F;AAC5F,SAASK,SAASjE,KAAc;IAC9B,IAAIA,UAAU,QAAQA,UAAUC,WAAW,OAAO;IAClD,IAAI,OAAOD,UAAU,WAAW,OAAOkE,OAAOlE,QAAQ,IAAI;IAC1D,IAAI,OAAOA,UAAU,UAAU,OAAO6D,OAAOM,aAAa,CAACnE,SAASkE,OAAOlE,SAASA;IACpF,IAAI,OAAOA,UAAU,UAAU,OAAOkD,cAAclD;IACpD,OAAOoE,KAAKC,SAAS,CAACrE;AACxB;AAEA,gEAAgE;AAChE,SAASsE,iBAAiBC,GAAW;IACnC,MAAMC,OAAOD,IAAIE,KAAK,CAAC;IACvB,IAAI,CAACD,MAAM,OAAO;QAAEE,IAAI;QAAMC,MAAMJ;IAAI;IACxC,MAAMK,OAAOL,IAAIZ,KAAK,CAACa,IAAI,CAAC,EAAE,CAACd,MAAM;IACrC,MAAMmB,QAAQD,KAAKH,KAAK,CAAC;IACzB,IAAI,CAACI,SAASA,MAAMC,KAAK,KAAK7E,WAAW,OAAO;QAAEyE,IAAI;QAAMC,MAAMJ;IAAI;IACtE,OAAO;QAAEG,IAAIE,KAAKjB,KAAK,CAAC,GAAGkB,MAAMC,KAAK;QAAGH,MAAMC,KAAKjB,KAAK,CAACkB,MAAMC,KAAK,GAAGD,KAAK,CAAC,EAAE,CAACnB,MAAM;IAAE;AAC3F;AAEA,2FAA2F;AAC3F,8EAA8E;AAC9E,6FAA6F;AAC7F,+FAA+F;AAC/F,iDAAiD;AACjD,SAASqB,oBAAoBrE,OAAe,EAAEsE,GAAqC,EAAEC,QAAkB;IACrG,IAAIC,QAAQ;IACZ,4FAA4F;IAC5F,yCAAyC;IACzC1F,MAAMwF,KAAK;QACTG,MAAKC,IAAI,EAAEC,IAAI;YACb,IAAI,CAAC/F,aAAa+F,KAAKC,GAAG,GAAG,OAAOrF;YACpCiF,QAAQ;YACR,OAAO1F,MAAM+F,KAAK;QACpB;IACF;IACA,8EAA8E;IAC9E,IAAIL,OAAOD,SAAStC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,yIAAyI,CAAC;AACzL;AAEA,wFAAwF;AACxF,8FAA8F;AAC9F,6FAA6F;AAC7F,4FAA4F;AAC5F,iFAAiF;AACjF,6FAA6F;AAC7F,oDAAoD;AACpD,SAAS8E,UAAUC,OAAe;IAChC,OAAOA,QAAQ7E,KAAK,CAAC,KAAK,CAAC,EAAE,CAACT,OAAO,CAAC,SAAS;AACjD;AAEA,SAASuF,iBAAiBhF,OAAe,EAAEgE,EAAU,EAAEO,QAAkB;IACvE,yFAAyF;IACzF,wDAAwD;IACxD,MAAMD,MAAMzF,cAAcmF,IAAI;QAAEiB,UAAU;IAAS;IACnD,MAAMC,UAAUZ,IAAIa,MAAM,CAAC5D,MAAM,CAAC,CAAC6D,MAAQ,CAAChG,oBAAoB4C,GAAG,CAACoD,IAAIC,IAAI;IAC5E,IAAIH,QAAQlC,MAAM,GAAG,GAAG;QACtB,MAAMsC,SAASJ,QAAQlC,MAAM,GAAG,IAAI,CAAC,MAAM,EAAEkC,QAAQlC,MAAM,GAAG,EAAE,MAAM,CAAC,GAAG;QAC1E,MAAMuC,aAAa,GAAGT,UAAUI,OAAO,CAAC,EAAE,CAACH,OAAO,IAAIO,QAAQ;QAC9Df,SAAStC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,sDAAsD,EAAEuF,YAAY;QACtG,OAAO;YAAEC,MAAM,CAAC;YAAGD;QAAW;IAChC;IAEA,IAAIC;IACJ,IAAI;QACFA,OAAOlB,IAAImB,IAAI;IACjB,EAAE,OAAOL,KAAK;QACZ,uFAAuF;QACvF,kEAAkE;QAClE,MAAMG,aAAaT,UAAU,AAACM,IAAcL,OAAO;QACnDR,SAAStC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,sDAAsD,EAAEuF,YAAY;QACtG,OAAO;YAAEC,MAAM,CAAC;YAAGD;QAAW;IAChC;IAEA,IAAIC,SAAS,QAAQA,SAASjG,WAAW,OAAO;QAAEiG,MAAM,CAAC;QAAGD,YAAY;IAAK;IAC7E,IAAI,OAAOC,SAAS,YAAYE,MAAMC,OAAO,CAACH,OAAO;QACnD,MAAMD,aAAa;QACnBhB,SAAStC,IAAI,CAAC,CAAC,SAAS,EAAEjC,QAAQ,CAAC,EAAEuF,WAAW,uBAAuB,CAAC;QACxE,OAAO;YAAEC,MAAM,CAAC;YAAGD;QAAW;IAChC;IACAlB,oBAAoBrE,SAASsE,KAAKC;IAClC,OAAO;QAAEiB,MAAMA;QAAiCD,YAAY;IAAK;AACnE;AAEA,OAAO,SAASK,UAAUC,IAAc,EAAEC,aAAwB,EAAE;IAClE,MAAMjC,MAAMtF,aAAasH,KAAKlE,OAAO,EAAE;IACvC,MAAM4C,WAAqB,EAAE;IAE7B,MAAM,EAAEP,EAAE,EAAEC,MAAM8B,OAAO,EAAE,GAAGnC,iBAAiBC;IAC/C,MAAM,EAAE2B,IAAI,EAAED,UAAU,EAAE,GAAGvB,OAAO,OAAO;QAAEwB,MAAM,CAAC;QAA8BD,YAAY;IAAK,IAAIP,iBAAiBa,KAAK7F,OAAO,EAAEgE,IAAIO;IAC1I,MAAMyB,SAA0D,CAAC;IAEjE,KAAK,MAAMpB,OAAOqB,OAAOxE,IAAI,CAAC+D,MAAO;QACnC,IAAItG,iBAAiB8C,GAAG,CAAC4C,MAAM;YAC7BL,SAAStC,IAAI,CAAC,CAAC,SAAS,EAAE4D,KAAK7F,OAAO,CAAC,8BAA8B,EAAE4E,IAAI,iCAAiC,CAAC;YAC7G;QACF;QACA,MAAMtF,QAAQiE,SAASiC,IAAI,CAACZ,IAAI;QAChC,IAAI,OAAOtF,UAAU,YAAYgD,kBAAkBhD,UAAU6D,OAAOC,KAAK,CAACC,KAAKC,KAAK,CAAChE,SAAS;YAC5FiF,SAAStC,IAAI,CAAC,CAAC,SAAS,EAAE4D,KAAK7F,OAAO,CAAC,EAAE,EAAE4E,IAAI,sBAAsB,EAAEtF,MAAM,8CAA8C,CAAC;QAC9H;QACA0G,MAAM,CAACpB,IAAI,GAAGtF;IAChB;IAEA,oEAAoE;IACpE,0CAA0C;IAC1C,MAAM4G,SAAS;QAAEC,OAAO9G,cAAcmG,KAAKW,KAAK;QAAGC,SAAS/G,cAAcmG,KAAKY,OAAO;QAAGC,MAAM1G,UAAUoG;IAAS;IAElH,OAAO;QACLzB,KAAK;YACHtE,SAAS6F,KAAK7F,OAAO;YACrBkC,SAAS2D,KAAK3D,OAAO;YACrBC,MAAM0D,KAAK1D,IAAI;YACftB,SAASgF,KAAKhF,OAAO;YACrB2E,MAAMQ;YACNT;YACAW;YACAI,WAAWL,OAAOM,WAAW,CAACT,WAAWvE,MAAM,CAAC,CAACiF,IAAMA,EAAEC,OAAO,EAAEC,GAAG,CAAC,CAACF;oBAAeA;uBAAT;oBAACA,EAAE7F,IAAI;qBAAE6F,aAAAA,EAAEC,OAAO,cAATD,iCAAAA,gBAAAA,GAAY3C,KAAKkC,SAASG;iBAAQ;;QAC1H;QACA3B;IACF;AACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sensemaking",
-  "version": "0.11.3",
+  "version": "0.11.4",
   "description": "Query and search your markdown notes with context-aware progressive disclosure: SQL over frontmatter, links, and text, plus semantic search and link-graph ranking. No server, no build step",
   "keywords": [
     "markdown",

package/skills/sense/SKILL.md

@@ -97,8 +97,7 @@ WHERE a.dst = ? AND b.dst IS NOT NULL AND b.dst <> a.dst;
 - **Dead links need the attachment filter.** `dst IS NULL` alone is not "broken link": a wikilink to anything that is not markdown (`[[Board.base]]`, `![[Pasted image.png]]`, `[[spec.pdf]]`) can never resolve, because sense indexes markdown and resolution only tries the exact path or `+.md`. Those are out of the index's universe, not broken. On a 1,400-note Obsidian vault the unfiltered query returns 143 rows where 14 are real. Exclude anything carrying a file extension, as in the recipe above, and widen the exclusion if your notes have dotted titles (`[[Node.js]]` carries one too, so a stricter list -- `'*.png'`, `'*.pdf'`, `'*.base'`, and whatever else your vault attaches -- is safer on a tree whose titles use dots). Scope it with `preset_files` as well: template and skill files are full of `[[Note Name]]` examples that are deliberately unresolved.
 - `has(field, value)`: array membership on JSON-array fields, substring on strings, false on NULL. This is the `includes()` convention. Substring means `has(f.status, 'active')` also matches `inactive`; exact scalar match is `f.status = ?`, deliberate substring is `LIKE`, exact array membership is `EXISTS (SELECT 1 FROM json_each(f.tags) WHERE value = ?)`. To aggregate per member instead, use `json_each(frontmatter.<field>)` (above) -- GROUP BY on the raw column splits `["a","b"]` and `["b","a"]` into separate buckets.
 - Date fields are stored as written. Compare through `datetime()`, which normalizes ISO 8601 timezone offsets to UTC: `WHERE datetime(created) >= datetime(?)`. Bare string comparison is only safe when every note uses the same offset.
-- SQLite's `datetime()` accepts a narrower spelling than ISO 8601 does, so sense normalizes the difference at index time: a colonless offset (`-0800`) and a space date/time separator are rewritten to the form SQLite parses. Punctuation only, never a timezone conversion, so `substr(d,1,10)` is still the local date.
-- **What normalization cannot fix stays invisible rather than excluded, so audit before trusting a range.** A value that is not a real instant (`T10:4:00`, unpadded) is stored as written, and `datetime()` returns NULL on it, which makes `datetime(d) < ?` NULL: the row drops out of the result *and* out of its negation, while `d IS NULL` stays false because the string is there. `SELECT COUNT(*) FROM frontmatter WHERE d IS NOT NULL AND datetime(d) IS NULL` must be 0; when it is not, write `datetime(d) IS NULL OR datetime(d) < ?` so the unparseable rows land somewhere.
+- Date spellings SQLite rejects (`-0800`, `-08`, a space separator) are normalized at index time, offset preserved. One it cannot fix is left as written and warned about by path: `datetime()` returns NULL there, so the row is invisible to a date comparison rather than excluded by it. List them with `WHERE d IS NOT NULL AND datetime(d) IS NULL`.
 - **SQLite's `now` is UTC, so any query about "today" needs `'localtime'`.** `date('now')` reads as tomorrow from mid-afternoon onward in the Americas, which silently flips "scheduled today" into "overdue" every evening: write `date('now','localtime')` and `datetime('now','start of day','localtime')`. This only matters where the boundary carries the meaning; a `'-90 day'` window is unaffected by a few hours of skew.
 - To bound what a query puts into context: `snippet()` excerpts just the matching text, `LIMIT` caps row counts, and selecting `path`/`title`/`summary` keeps rows small. `SELECT text FROM content` returns the tree's entire prose (sense warns past 50 KB). Aggregates (`COUNT`, `GROUP BY`) are already bounded. `SELECT * FROM frontmatter` is always safe: prose is not a frontmatter column.