Function buildNoteVector

packages/shared/src/maintenance.ts:153–172

* Build a normalized TF-IDF vector for a note's tokens — the same weighting the * semantic engine (`semantic.ts`) uses per chunk, applied here at note level so * a pairwise cosine measures whole-note similarity. A normalized vector means * the dot product *is* the cosine.

(tokens: string[], idf: (term: string) => number)

      return KIND_RANK[a.kind] - KIND_RANK[b.kind];
    }
    const pa = a.paths.join('\u0000');
    const pb = b.paths.join('\u0000');
    return pa.localeCompare(pb) || a.detail.localeCompare(b.detail);
  });
}

/**
 * Scan a corpus of notes for vault-hygiene problems and return a deterministic,
 * stably-ordered list of findings. Pure: no I/O, no model, no mutation — the
 * same documents always yield the same findings.
 */
export function scanVault(
  documents: readonly MaintenanceDocument[],
  options: ScanVaultOptions = {},
): MaintenanceFinding[] {
  const duplicateThreshold = options.duplicateThreshold ?? DEFAULT_DUPLICATE_THRESHOLD;
  const allPaths = documents.map((doc) => doc.path);
  const pathSet = new Set(allPaths);
  const contentByPath = new Map(documents.map((doc) => [doc.path, doc.content]));

  // --- Link analysis: resolved (real) out-targets, inbound sources, broken links.
  const resolvedOut = new Map<string, Set<string>>();
  const inbound = new Map<string, Set<string>>();