diff --git a/CHANGELOG.md b/CHANGELOG.md
index 43e1d15..3d2e622 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -16,6 +16,7 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 - A test manifest or test sequence saved with a UTF-8 byte-order mark is read correctly again. Until now such a file was rejected outright, so every test it registered quietly vanished from the graph with nothing but a warning to show for it — and on Windows that mark is what PowerShell writes by default. (#316)
 - Saved Access queries now appear in the graph no matter what order Dysflow writes them. The `queries.json` manifest is what marks the `.sql` files beside it as saved queries, and a manifest arriving after its queries left them unindexed until the next manual sync. Editing only the manifest — renaming a query, adding an entry — now re-indexes too. (#318)
 - Agents are now told about three Access surfaces the index already carried but never advertised: saved queries and the tables they touch, Dysflow test sequences alongside the manifests, and the table-and-column structure from the Access ERD export. An agent that was not told these exist fell back to reading the files by hand. (#320)
+- Table and column names from an Access structure export keep their accents. The export is read with the same encoding rules as the rest of an Access project, so a table like `TbSituación` is no longer recorded under a corrupted name — which had also stopped it from lining up with the queries and code that reference it, leaving its column list attached to nothing. (#322)
 
 ## [1.17.1] - 2026-09-12
 
diff --git a/__tests__/index-orphan-watchdog.test.ts b/__tests__/index-orphan-watchdog.test.ts
index a3d49e9..f728310 100644
--- a/__tests__/index-orphan-watchdog.test.ts
+++ b/__tests__/index-orphan-watchdog.test.ts
@@ -56,7 +56,12 @@ vi.mock('../src/project-config', () => ({
 vi.mock('../src/resolution/frameworks', () => ({ detectFrameworks: () => [] }));
 vi.mock('../src/extraction/tree-sitter', () => ({ extractFromSource: vi.fn() }));
 
-vi.mock('../src/extraction/grammars', () => ({
+// Spread the real module and override only what this test needs to control.
+// A from-scratch object silently returns `undefined` for every grammars helper
+// it forgot, so the orchestrator importing one more of them (e.g.
+// `isAccessErdFile`, #322) failed here as an unrelated-looking index error.
+vi.mock('../src/extraction/grammars', async (importOriginal) => ({
+  ...(await importOriginal<typeof import('../src/extraction/grammars')>()),
   detectLanguage: () => 'typescript',
   isSourceFile: (file: string) => file.endsWith('.ts'),
   isLanguageSupported: () => true,
diff --git a/__tests__/vba-erd-encoding.test.ts b/__tests__/vba-erd-encoding.test.ts
new file mode 100644
index 0000000..6311525
--- /dev/null
+++ b/__tests__/vba-erd-encoding.test.ts
@@ -0,0 +1,166 @@
+/**
+ * Issue #322: a CP1252 Access ERD export must decode like the rest of the
+ * export, so its table names still join the graph.
+ *
+ * #53 gave `.bas` / `.cls` / `.form.txt` / `.report.txt` / `.sql` an
+ * encoding-robust read because Access writes Windows-1252 and accented
+ * identifiers are everywhere in Spanish-language Access projects. The Access
+ * structure export (`ERD/*.md`) was left on the plain UTF-8 path.
+ *
+ * The damage was not cosmetic. The `.sql` naming a table DID get the CP1252
+ * fallback, so it emitted a placeholder named `TbSituación`; the ERD declaring
+ * that same table did NOT, so it declared `TbSituaci<U+FFFD>n`. The two never
+ * matched: the declaration never joined the placeholder, and the table's
+ * fields hung off an orphan node. This suite pins both halves — the decode,
+ * and the join it exists to make possible.
+ */
+import { describe, it, expect, afterEach } from 'vitest';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import * as os from 'node:os';
+import { readVbaSource } from '../src/extraction/vba-source';
+import { usesAccessEncoding } from '../src/extraction';
+import CodeGraph from '../src/index';
+
+/** The generated export's real byte shape: BOM, CRLF, `# Estructura de Datos:`. */
+const ERD_DOC = [
+  '# Estructura de Datos: Datos.accdb',
+  '',
+  '## Tabla: TbSituación',
+  '| Campo | Tipo | Longitud |',
+  '| :--- | :--- | :--- |',
+  '| Código | 4 | 4 |',
+  '| Descripción | 10 | 255 |',
+  '',
+].join('\r\n');
+
+/** Same document, CP1252 on disk — one byte per accented character. */
+const cp1252 = (text: string) => Buffer.from(text, 'latin1');
+const utf8 = (text: string) => Buffer.from(text, 'utf8');
+
+const openProjects: Array<{ cg: CodeGraph; dir: string }> = [];
+
+afterEach(async () => {
+  while (openProjects.length > 0) {
+    const { cg, dir } = openProjects.pop()!;
+    try {
+      await cg.close();
+    } catch {
+      // ignore close errors
+    }
+    await new Promise((r) => setTimeout(r, 50));
+    try {
+      fs.rmSync(dir, { recursive: true, force: true });
+    } catch {
+      // ignore cleanup EPERM on Windows
+    }
+  }
+});
+
+describe('Access ERD export — encoding (#322)', () => {
+  it('routes the ERD export by path shape, and nothing else markdown', () => {
+    // `.md` must never become a family EXTENSION — that would pull every
+    // markdown file in every repo through the CP1252 fallback. Only a document
+    // directly inside an `ERD/` directory qualifies.
+    for (const p of [
+      'ERD/Estructura_Datos.md',
+      'src/backend/erd/Estructura_Datos.md',
+      'C:\\proj\\ERD\\Estructura_Datos.md',
+    ]) {
+      expect(usesAccessEncoding(p), p).toBe(true);
+    }
+    for (const p of [
+      'docs/architecture.md',
+      'README.md',
+      'ERD/notes/deep.md',
+      'index.ts',
+    ]) {
+      expect(usesAccessEncoding(p), p).toBe(false);
+    }
+    // The extension half still routes on its own.
+    expect(usesAccessEncoding('src/modules/constantes.bas')).toBe(true);
+  });
+
+  it('decodes a CP1252 export to the same names as its UTF-8 twin', () => {
+    const asUtf8 = readVbaSource('ERD/Estructura_Datos.md', {
+      readFile: () => utf8(ERD_DOC),
+    });
+    let fallbacks = 0;
+    const asCp1252 = readVbaSource('ERD/Estructura_Datos.md', {
+      readFile: () => cp1252(ERD_DOC),
+      onFallback: () => { fallbacks++; },
+    });
+
+    expect(asCp1252.text).toBe(asUtf8.text);
+    expect(asCp1252.text).toContain('TbSituación');
+    expect(asCp1252.text).toContain('Código');
+    // The fallback is what did the work; a silent pass would mean the bytes
+    // happened to be valid UTF-8 and the test proves nothing.
+    expect(fallbacks).toBe(1);
+  });
+
+  it('leaves a UTF-8 export with a BOM byte-identical', () => {
+    const withBom = Buffer.concat([
+      Buffer.from([0xef, 0xbb, 0xbf]),
+      utf8(ERD_DOC),
+    ]);
+    let fallbacks = 0;
+    const r = readVbaSource('ERD/Estructura_Datos.md', {
+      readFile: () => withBom,
+      onFallback: () => { fallbacks++; },
+    });
+
+    expect(r.text).toBe(ERD_DOC);
+    expect(r.bomStripped).toBe(true);
+    expect(fallbacks).toBe(0);
+  });
+
+  it('joins a CP1252 ERD table to the query that references it', async () => {
+    // The failure this issue is really about: both files are CP1252 on disk,
+    // but only one of them used to be decoded as such, so the two spellings of
+    // the same table name never met.
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'cg-erd-enc-'));
+    fs.mkdirSync(path.join(dir, 'ERD'), { recursive: true });
+    fs.mkdirSync(path.join(dir, 'queries'), { recursive: true });
+
+    fs.writeFileSync(path.join(dir, 'ERD', 'Estructura_Datos.md'), cp1252(ERD_DOC));
+    fs.writeFileSync(
+      path.join(dir, 'queries', 'Q_Situaciones.sql'),
+      cp1252('SELECT * FROM TbSituación;\r\n'),
+    );
+    fs.writeFileSync(
+      path.join(dir, 'queries', 'queries.json'),
+      '[{ "name": "Q_Situaciones", "file": "Q_Situaciones.sql" }]',
+    );
+
+    const cg = await CodeGraph.init(dir, { index: false });
+    openProjects.push({ cg, dir });
+    await cg.indexAll();
+
+    // The ERD declared the table and its fields under their real names. A
+    // replacement character anywhere here is the exact signature of the bug.
+    const erdNodes = cg.getNodesInFile('ERD/Estructura_Datos.md');
+    const names = erdNodes.map((n) => n.name);
+    expect(names).toContain('TbSituación');
+    expect(names).toContain('Código');
+    expect(names).toContain('Descripción');
+    expect(names.filter((n) => (n ?? '').includes('�'))).toEqual([]);
+
+    // And the two sides meet: the saved query's table reference resolves onto
+    // the ERD's own declaration rather than a second, differently-spelled node.
+    const table = erdNodes.find(
+      (n) => n.kind === 'class' && n.name === 'TbSituación',
+    );
+    expect(table).toBeDefined();
+
+    const query = cg
+      .getNodesInFile('queries/Q_Situaciones.sql')
+      .find((n) => n.kind === 'query');
+    expect(query).toBeDefined();
+
+    const toTable = cg
+      .getOutgoingEdges(query!.id)
+      .filter((e) => e.target === table!.id);
+    expect(toTable.length).toBeGreaterThan(0);
+  });
+});
diff --git a/__tests__/vba-source.test.ts b/__tests__/vba-source.test.ts
index b3d9c78..4e03ccb 100644
--- a/__tests__/vba-source.test.ts
+++ b/__tests__/vba-source.test.ts
@@ -134,6 +134,11 @@ describe('isVbaFamilyFile - extension routing (Issue #53)', () => {
       'Weird.clsss',
       'MyModule.txt', // NOT `.form.txt` / `.report.txt`
       'notes.form.md',
+      // `.md` is never a family EXTENSION. The Access structure export shares
+      // the decoding but is matched by path shape at the read site instead —
+      // see `usesAccessEncoding` and its tests (#322).
+      'ERD/Estructura_Datos.md',
+      'docs/architecture.md',
     ]) {
       expect(isVbaFamilyFile(p)).toBe(false);
     }
diff --git a/src/extraction/index.ts b/src/extraction/index.ts
index 8d76dab..af799d1 100644
--- a/src/extraction/index.ts
+++ b/src/extraction/index.ts
@@ -24,7 +24,7 @@ import { extractFromSource } from './tree-sitter';
 import type { VbaExtractionOptions } from './vba/options';
 import { readVbaSource, isVbaFamilyFile } from './vba-source';
 import { ParseWorkerPool, resolveParsePoolSize, resolveParseTimeoutMs } from './parse-pool';
-import { detectLanguage, isSourceFile, isLanguageSupported, isFileLevelOnlyLanguage, initGrammars, loadGrammarsForLanguages, readGrammarWasmBytes } from './grammars';
+import { detectLanguage, isSourceFile, isLanguageSupported, isFileLevelOnlyLanguage, initGrammars, loadGrammarsForLanguages, readGrammarWasmBytes, isAccessErdFile } from './grammars';
 import { loadExtensionOverrides, loadIncludeIgnoredPatterns, loadExcludePatterns, loadVbaConfig, loadIncludePatterns, loadDysflowExportConfig } from '../project-config';
 import { isCodeGraphDataDir } from '../directory';
 import { logDebug, logWarn } from '../errors';
@@ -130,15 +130,38 @@ export function hashContent(content: string): string {
  * change-detection hashing so a BOM-carrying or CP1252 VBA file doesn't hash
  * differently at detect-time than what was stored (perpetual "modified").
  */
+/**
+ * Files that must be decoded with Access encoding rules (BOM strip + CP1252
+ * fallback) rather than as plain UTF-8.
+ *
+ * Two predicates because the family is matched two ways. `isVbaFamilyFile`
+ * owns the extensions (`.bas`, `.cls`, `.form.txt`, `.sql`, …). The Access
+ * structure export is `ERD/*.md`, so it is matched by PATH SHAPE — `.md`
+ * alone must never route every markdown file in a repo through the CP1252
+ * fallback — and that shape is owned by `grammars.ts` (issue #322).
+ *
+ * Combining them here rather than inside `isVbaFamilyFile` keeps
+ * `vba-source.ts` a leaf module: this is the only place that needs both.
+ *
+ * Without the ERD half, an export naming a table with accents decoded to
+ * replacement characters while the `.sql` and `.bas` referencing that same
+ * table decoded it correctly, so the ERD declaration never joined the
+ * placeholder those scanners emit and the table's field list hung off an
+ * orphan node.
+ */
+export function usesAccessEncoding(filePath: string): boolean {
+  return isVbaFamilyFile(filePath) || isAccessErdFile(filePath);
+}
+
 async function readForExtraction(fullPath: string): Promise<string> {
-  return isVbaFamilyFile(fullPath)
+  return usesAccessEncoding(fullPath)
     ? readVbaSource(fullPath).text
     : await fsp.readFile(fullPath, 'utf-8');
 }
 
 /** Synchronous counterpart of {@link readForExtraction} for the sync read sites. */
 function readForExtractionSync(fullPath: string): string {
-  return isVbaFamilyFile(fullPath)
+  return usesAccessEncoding(fullPath)
     ? readVbaSource(fullPath).text
     : fs.readFileSync(fullPath, 'utf-8');
 }
diff --git a/src/extraction/vba-source.ts b/src/extraction/vba-source.ts
index 2dba000..ac36633 100644
--- a/src/extraction/vba-source.ts
+++ b/src/extraction/vba-source.ts
@@ -15,7 +15,10 @@
  * Scope (do NOT change decoding for tree-sitter languages): only
  * `.bas`/`.cls`/`.frm`/`.dsr`/`.form.txt`/`.report.txt`/`.sql` files get
  * the BOM strip + CP1252 fallback; everything else stays byte-identical to
- * the existing `fs.readFile(path, 'utf-8')` path.
+ * the existing `fs.readFile(path, 'utf-8')` path. The Access structure export
+ * (`ERD/*.md`) shares that decoding but is matched by path shape, so the
+ * combined predicate lives at the read site — see `usesAccessEncoding` in
+ * `./index.ts` (issue #322).
  */
 import * as fs from 'fs';