diff --git a/src/tools/documents/__fixtures__/sample.doc b/src/tools/documents/__fixtures__/sample.doc
new file mode 100644
index 0000000..643838d
Binary files /dev/null and b/src/tools/documents/__fixtures__/sample.doc differ
diff --git a/src/tools/documents/__fixtures__/table.doc b/src/tools/documents/__fixtures__/table.doc
new file mode 100644
index 0000000..ddaba51
Binary files /dev/null and b/src/tools/documents/__fixtures__/table.doc differ
diff --git a/src/tools/documents/doc.fixtures.test.ts b/src/tools/documents/doc.fixtures.test.ts
new file mode 100644
index 0000000..5c6321e
--- /dev/null
+++ b/src/tools/documents/doc.fixtures.test.ts
@@ -0,0 +1,31 @@
+import { describe, it, expect } from 'vitest';
+import { readFileSync } from 'node:fs';
+import { resolve } from 'node:path';
+import { extractDocText } from './doc.lib';
+
+// Real Word 97-2003 binary .doc files (generated with macOS `textutil`, neutral
+// synthetic content) — regression coverage for the OLE + piece-table extractor.
+const load = (name: string) =>
+  new Uint8Array(readFileSync(resolve(process.cwd(), 'src/tools/documents/__fixtures__', name)));
+
+describe('extractDocText on real .doc fixtures', () => {
+  it('extracts plain and Unicode text from a simple .doc', () => {
+    const text = extractDocText(load('sample.doc'));
+    expect(text).toContain('Hello World');
+    expect(text).toContain('This is a test document.');
+    expect(text).toContain('Café résumé'); // accented characters
+    expect(text).toContain('90% done');
+    expect(text).toContain('Line A');
+    expect(text).toContain('Line B');
+  });
+
+  it('extracts a table as rows (tab-separated cells) plus a bulleted list', () => {
+    const text = extractDocText(load('table.doc'));
+    expect(text).toContain('Quarterly Report');
+    expect(text).toContain('Item\tQ1\tQ2');
+    expect(text).toContain('Revenue\t100\t150');
+    expect(text).toContain('Costs\t40\t55');
+    expect(text).toContain('First bullet');
+    expect(text).toContain('Closing line.');
+  });
+});
diff --git a/src/tools/documents/doc.lib.ts b/src/tools/documents/doc.lib.ts
index b90dc83..8b44dd6 100644
--- a/src/tools/documents/doc.lib.ts
+++ b/src/tools/documents/doc.lib.ts
@@ -106,12 +106,18 @@ function decodePieces(wd: Uint8Array, table: Uint8Array, pcdt: Pcdt, ccpText: nu
   return out;
 }
 
-/** Tidy extracted text: collapse runs of blank lines and trailing spaces. */
+/**
+ * Tidy extracted text. Table cells are separated by single tab marks; a run of
+ * two or more marks (a cell mark plus the row-end mark) is a row boundary, so
+ * those become line breaks. Leading tabs and blank-line runs are trimmed.
+ */
 export function cleanDocText(s: string): string {
   return s
     .replace(/\r\n?/g, '\n')
-    .replace(/[ \t]+\n/g, '\n')
     .replace(/\uFEFF/g, '')
+    .replace(/\t{2,}/g, '\n') // consecutive cell/row marks \u2192 new row
+    .replace(/[ \t]+\n/g, '\n')
+    .replace(/\n[ \t]+/g, '\n') // strip leading tabs/spaces on a line
     .replace(/\n{3,}/g, '\n\n')
     .trim();
 }