@@ -0,0 +1,14 @@
+---
+'@objectstack/rest': patch
+---
+
+fix(rest): a refused `GET /data/:object/export?format=xlsx` answers the 400 envelope, not a 200 workbook stream
+
+Clause-②: no
+Metadata-protocol: none
+
+The xlsx export opened its streaming workbook writer before its first read. Opening the writer commits the response to a streaming 200, so a query that read refused (for example `?search=x&searchFields=nope`, a search over a field the object does not have) reached the client as a 200 workbook download, while `format=csv` and `format=json` answered the 400 envelope.
+
+- **Read first.** The workbook writer now opens after the first page is read, as the csv and json exports write their first byte. A refusal on that read answers the same status and body as the csv export.
+- **A successful export is unchanged.** One page, several pages and no rows produce the same status, headers and workbook cells.
+- A refusal on a later page, after rows have been written, still ends the stream; that is not changed here.
@@ -1457,7 +1457,7 @@ export function registerDataActionEndpoints(this: DataActionRouteHost, basePath:
                 let exported = 0;
                 let firstChunk = true;
                 let skip = 0;
-                const xlsx = format === 'xlsx' ? await createXlsxStream(res, styled) : null;
+                let xlsx: Awaited<ReturnType<typeof createXlsxStream>> | null = null;
 
                 while (exported < limit) {
                     const take = Math.min(chunkSize, limit - exported);
@@ -1507,6 +1507,12 @@ export function registerDataActionEndpoints(this: DataActionRouteHost, basePath:
                     // runs at least once (`limit` is at least 1), so the
                     // closing `]` below always has its opener.
                     if (format === 'json' && firstChunk) res.write('[');
+                    // [#27755] The xlsx workbook writer opens after the first
+                    // read for the same reason: opening it commits the stream
+                    // too, so a refused query answered 200 instead of the 400
+                    // envelope. Opened before the zero-row break, so an empty
+                    // export is still a readable workbook.
+                    if (format === 'xlsx' && firstChunk) xlsx = await createXlsxStream(res, styled);
 
                     if (rows.length === 0) break;
 
 
@@ -0,0 +1,159 @@
+// Copyright (c) 2026 ObjectStack. Licensed under the Apache-2.0 license.
+
+/**
+ * [#27755] `GET /data/:object/export?format=xlsx` reads its first page before
+ * it opens the workbook writer, so a query the read refuses answers the `400`
+ * envelope the csv and json exports answer — not a `200` carrying a workbook
+ * stream. The xlsx half of #27682.
+ *
+ * The defect: the xlsx branch opened `createXlsxStream(res, …)` before the
+ * first `findData`. Opening the writer commits the adapter to a streaming
+ * `200`, so a refusal raised by that read (`searchFields` naming a field that
+ * does not exist) reached the client as a `200`, while `format=csv` answered
+ * `400 INVALID_FIELD`.
+ *
+ * Both halves are pinned through the real `HonoHttpServer` (the adapter
+ * `os serve` mounts), because the `200` was decided by the adapter's streaming
+ * switch, which a hand-rolled response double does not have:
+ *
+ *  - refusal: over a real `SqlDriver` (SQLite), the refused query answers the
+ *    same status and body for xlsx as for csv;
+ *  - preservation: a successful xlsx export still answers `200`, the workbook
+ *    content type and a readable workbook with the same cells — one page,
+ *    several pages, and no rows at all.
+ *
+ * A refusal on a LATER page (after rows have been written) is out of this pin,
+ * as it is in #27682's (triage 6510461172).
+ */
+
+import { describe, it, expect, vi, beforeAll, afterAll } from 'vitest';
+import { HonoHttpServer } from '@objectstack/plugin-hono-server';
+import { ObjectQL } from '@objectstack/objectql';
+import { SqlDriver } from '@objectstack/driver-sql';
+import { ObjectStackProtocolImplementation } from '@objectstack/metadata-protocol';
+import { RestServer } from './rest-server';
+import { installDefaultOrganization } from './default-organization.testkit.js';
+import { loadXlsxWorkbook } from './xlsx-test-loader.js';
+
+const OBJECT = 'rest_export_xlsx_27755';
+
+const XLSX_TYPE = 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet';
+
+const ITEMS = {
+    name: OBJECT,
+    label: 'Items 27755',
+    fields: {
+        label: { name: 'label', type: 'text' as const },
+        note: { name: 'note', type: 'text' as const },
+    },
+};
+
+/**
+ * code · querystring: two refusals `findData` raises on the first read — a
+ * search over a field the object does not have, and #27484's empty override.
+ */
+const REFUSED: ReadonlyArray<readonly [string, string]> = [
+    ['INVALID_FIELD', '?search=x&searchFields=nope'],
+    ['VALIDATION_ERROR', '?search=x&searchFields=,'],
+];
+
+describe('[#27755] export: a refused first read answers the envelope for xlsx as for csv — sqlite', () => {
+    let engine: ObjectQL;
+    let app: { fetch: (req: Request) => Promise<Response> };
+
+    beforeAll(async () => {
+        const driver: any = new SqlDriver({ client: 'better-sqlite3', connection: { filename: ':memory:' }, useNullAsDefault: true } as any);
+        engine = new ObjectQL();
+        engine.registerDriver(driver, true);
+        await engine.init();
+        engine.registry.registerObject(ITEMS as any);
+        await engine.syncSchemas();
+        await installDefaultOrganization(engine);
+        for (const label of ['a', 'b', 'c']) await engine.insert(OBJECT, { id: `r${label}`, label, note: `n-${label}` } as any);
+
+        const protocol = new ObjectStackProtocolImplementation(engine as any);
+        const server = new HonoHttpServer(0);
+        const rest = new RestServer(server as any, protocol as any, { api: { requireAuth: false } } as any);
+        (rest as any).resolveExecCtx = async () => ({ userId: 'u_export_xlsx' });
+        rest.registerRoutes();
+        app = server.getRawApp() as any;
+    });
+
+    afterAll(async () => {
+        try { await engine?.destroy(); } catch { /* noop */ }
+    });
+
+    async function exportAs(format: string, query: string) {
+        const res = await app.fetch(new Request(`http://local/api/v1/data/${OBJECT}/export${query}&format=${format}`));
+        // The status is read before the body: a committed `200` stream is the
+        // defect, and its body need not ever finish.
+        const status = res.status;
+        if (status !== 400) {
+            try { await res.body?.cancel(); } catch { /* noop */ }
+            return { status, body: undefined as any };
+        }
+        return { status, body: JSON.parse(await res.text()) };
+    }
+
+    it.each(REFUSED)('the control: csv answers 400 %s for %s', async (code, query) => {
+        const csv = await exportAs('csv', query);
+        expect(csv.status).toBe(400);
+        expect(csv.body?.code).toBe(code);
+    });
+
+    it.each(REFUSED)('xlsx answers the same 400 %s as csv for %s — no workbook 200', async (_code, query) => {
+        const csv = await exportAs('csv', query);
+        const xlsx = await exportAs('xlsx', query);
+        expect(xlsx.status, 'xlsx must not commit a 200 before its first read').toBe(400);
+        expect(xlsx.body).toEqual(csv.body);
+    });
+});
+
+describe('[#27755] export: a successful xlsx export is unchanged', () => {
+    const TASK = {
+        name: 'task',
+        label: 'Task',
+        fields: {
+            id: { type: 'text', label: 'ID' },
+            title: { type: 'text', label: 'Title' },
+        },
+    };
+
+    function boot(total: number) {
+        const all = Array.from({ length: total }, (_, i) => ({ id: String(i + 1), title: `t${i + 1}` }));
+        const findData = vi.fn(async (req: any) => {
+            const { offset = 0, limit = all.length } = req.query ?? {};
+            return { object: 'task', records: all.slice(offset, offset + limit) };
+        });
+        const protocol: any = {
+            getMetaItems: vi.fn().mockResolvedValue({ items: [TASK] }),
+            getMetaItem: vi.fn().mockResolvedValue({ type: 'object', name: 'task', item: TASK }),
+            findData,
+        };
+        const server = new HonoHttpServer(0);
+        const rest = new RestServer(server as any, protocol, { api: { requireAuth: false } } as any);
+        (rest as any).resolveExecCtx = async () => ({ userId: 'u_export_xlsx' });
+        rest.registerRoutes();
+        return { app: server.getRawApp(), findData, all };
+    }
+
+    it.each([
+        ['one page', 3, '', 1],
+        ['several pages (three reads of 50, 50, 20)', 120, '&page=50', 3],
+        ['no rows', 0, '', 1],
+    ])('%s', async (_label, total, query, reads) => {
+        const { app, findData, all } = boot(total);
+        const res: Response = await app.fetch(new Request(`http://local/api/v1/data/task/export?format=xlsx${query}`));
+        expect(res.status).toBe(200);
+        expect(res.headers.get('content-type')).toBe(XLSX_TYPE);
+        expect(res.headers.get('x-export-styles')).toBe('applied');
+
+        const ws = (await loadXlsxWorkbook(Buffer.from(await res.arrayBuffer()))).worksheets[0];
+        const cells: unknown[][] = [];
+        ws.eachRow((row) => { cells.push((row.values as unknown[]).slice(1)); });
+        // No rows: the schema-derived header is not authoritative without a
+        // security projection (#3391), so the workbook is empty but readable.
+        expect(cells).toEqual(total === 0 ? [] : [['ID', 'Title'], ...all.map((r) => [r.id, r.title])]);
+        expect(findData).toHaveBeenCalledTimes(reads);
+    });
+});