@@ -0,0 +1,14 @@
+---
+'@objectstack/service-datasource': patch
+---
+
+fix(service-datasource): a code object bound to a Setup-created managed datasource gets its table when the datasource's pool opens, on create and after a restart (#27701)
+
+Clause-②: no
+Metadata-protocol: none
+
+An author declares an object with `datasource: 'warehouse'`, and an admin creates `warehouse` in Setup → Datasources. Boot's schema sync skipped the object, because no `warehouse` driver existed yet, and the pool the admin door opened synced nothing either. The datasource answered `ok`, the table was never created, and every read of the object was a 500 (Postgres `42P01`), the same after every restart.
+
+- **Now.** Opening a runtime datasource's pool (the admin create, the boot rehydration, and a peer replica's convergence on a new datasource) syncs the schema of the objects already bound to it, as boot does for a declared datasource. A table that is missing is created; a table that exists is left as the sync leaves it. One object's refused sync is logged at `error` and does not fail the connect, as on the declared path.
+- ⛔ **Only a managed datasource.** A datasource whose `schemaMode` is `external` or `validate-only` syncs nothing for its objects, so it never gets DDL. Its mode is read as the driver factory reads it.
+- **Not changed.** An edit that rebuilds the pool (a changed `config`, credential, `schemaMode` or `active`) still syncs nothing.
@@ -0,0 +1,201 @@
+// Copyright (c) 2026 ObjectStack. Licensed under the Apache-2.0 license.
+
+/**
+ * [#27701] A code object bound to a runtime (Setup-created) managed datasource
+ * gets its table when the datasource's pool opens: on create, and on the boot
+ * rehydration after a restart.
+ *
+ * ## The defect this pins closed
+ *
+ * An author declares `wh_note` with `datasource: 'warehouse'`, and an admin
+ * creates `warehouse` in Setup → Datasources. Boot's schema sync skipped the
+ * object, because no `warehouse` driver existed yet. The plugin's
+ * `registerPool` then opened the pool with `connect(record, { context })` and
+ * no `objects`, so the connect's sync loop had nothing to sync. The datasource
+ * answered `ok`, the table was never created, and every read of the object was
+ * a 500 (Postgres `42P01`, SQLite `no such table`), the same after a restart.
+ *
+ * ## What each case pins
+ *
+ *  - on create, the bound managed object's table exists and the object reads;
+ *  - on the restart's rehydration, a table that is missing (the state an
+ *    install created before this fix is in) is created, and the object reads;
+ *  - ⛔ an `external` or `validate-only` datasource gets no schema sync at all:
+ *    its pool opens, and nothing is synced for the objects bound to it.
+ *
+ * The composition is the real one: the real `DatasourceAdminServicePlugin`
+ * through `init()` and `start()`, a real `ObjectQL` engine on the `'objectql'`
+ * and `'data'` slots (one engine, as `ObjectQLPlugin` registers it), and the
+ * real default driver factory opening SQLite files. The engine's `sys_metadata`
+ * door is left off the `'data'` slot, so the plugin takes its in-memory branch:
+ * durability across a restart is the subject of `datasource-admin-plugin.test.ts`
+ * ("runtime datasource durability"). Here the restart's stored row arrives in
+ * the metadata service, where that restore leaves it. The code object is put
+ * in the engine registry before the plugin starts, as `AppPlugin.init()` does
+ * through the `'manifest'` service in Phase 1.
+ */
+
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import { mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { ObjectQL } from '@objectstack/objectql';
+import type { IDataDriver } from '@objectstack/spec/contracts';
+import type { IDatasourceAdminService } from '../contracts/index.js';
+import { DatasourceAdminServicePlugin } from '../datasource-admin-plugin.js';
+import { createDefaultDatasourceDriverFactory } from '../default-datasource-driver-factory.js';
+
+const teardowns: Array<() => Promise<void> | void> = [];
+afterEach(async () => {
+  for (const t of teardowns.splice(0).reverse()) await t();
+});
+
+type Knex = {
+  schema: { hasTable(n: string): Promise<boolean>; dropTableIfExists(n: string): Promise<void> };
+};
+
+/** The author's code object, bound by name to a datasource the stack does not declare. */
+const boundTo = (name: string, datasource: string, extra: Record<string, unknown> = {}) => ({
+  name,
+  label: name,
+  datasource,
+  fields: { title: { type: 'text', label: 'Title' } },
+  ...extra,
+});
+
+/** A fresh directory for this case's SQLite files. */
+function scratch(): string {
+  const dir = mkdtempSync(join(tmpdir(), 'ds-27701-'));
+  teardowns.push(() => rmSync(dir, { recursive: true, force: true }));
+  return dir;
+}
+
+/**
+ * One boot: an engine holding the code objects, then the plugin's `init()` and
+ * `start()`. `stored` is what the restart's restore leaves in the metadata
+ * service.
+ */
+async function boot(opts: { objects: Array<Record<string, unknown>>; stored?: Array<Record<string, unknown>> }) {
+  const ql = new ObjectQL();
+  for (const object of opts.objects) (ql as any).registry.registerObject(object as never, 'com.acme.warehouse');
+  const synced = vi.spyOn(ql, 'syncObjectSchema');
+
+  const records = new Map<string, Map<string, unknown>>([
+    ['datasource', new Map((opts.stored ?? []).map((r) => [r.name as string, r]))],
+  ]);
+  const metadata = {
+    get: async (t: string, n: string) => records.get(t)?.get(n),
+    list: async (t: string) => [...(records.get(t)?.values() ?? [])],
+    register: async (t: string, n: string, d: unknown) => {
+      if (!records.has(t)) records.set(t, new Map());
+      records.get(t)!.set(n, d);
+    },
+    unregister: async (t: string, n: string) => void records.get(t)?.delete(n),
+  };
+  const sysMetadataDoor = new Set(['find', 'findOne', 'insert', 'update', 'delete']);
+  const data = new Proxy(ql, {
+    get: (target, key, receiver) => (sysMetadataDoor.has(key as string) ? undefined : Reflect.get(target, key, receiver)),
+  });
+
+  const services = new Map<string, unknown>([
+    ['objectql', ql],
+    ['data', data],
+    ['metadata', metadata],
+  ]);
+  const ctx = {
+    getService: (n: string) => {
+      if (!services.has(n)) throw new Error(`no service: ${n}`);
+      return services.get(n);
+    },
+    registerService: (n: string, s: unknown) => void services.set(n, s),
+    trigger: async () => {},
+    logger: { info() {}, warn() {}, error() {}, debug() {} },
+  };
+
+  const plugin = new DatasourceAdminServicePlugin({ driverFactory: createDefaultDatasourceDriverFactory({}) });
+  await plugin.init(ctx as never);
+  await plugin.start(ctx as never);
+  teardowns.push(async () => {
+    for (const driver of (ql as any).drivers?.values?.() ?? []) await (driver as IDataDriver).disconnect?.();
+  });
+
+  const knexOf = (datasource: string) => (ql.getDriverByName(datasource) as unknown as { getKnex(): Knex }).getKnex();
+  const read = (object: string) => ql.find(object, { context: { isSystem: true } } as never);
+  return { ql, synced, knexOf, read, admin: services.get('datasource-admin') as IDatasourceAdminService };
+}
+
+describe('a code object bound to a runtime managed datasource gets its table', () => {
+  it('on create: the table exists and the object reads', async () => {
+    const dir = scratch();
+    const { admin, knexOf, read } = await boot({ objects: [boundTo('wh_note', 'warehouse')] });
+
+    const created = await admin.createDatasource({
+      name: 'warehouse',
+      driver: 'sqlite',
+      config: { filename: join(dir, 'warehouse.db') },
+    });
+
+    expect(created.status).toBe('ok');
+    expect(await knexOf('warehouse').schema.hasTable('wh_note')).toBe(true);
+    await expect(read('wh_note')).resolves.toEqual([]);
+  });
+
+  it('after a restart: the rehydrated pool creates a table that is missing, and the object reads', async () => {
+    const dir = scratch();
+    const record = {
+      name: 'warehouse',
+      driver: 'sqlite',
+      origin: 'runtime',
+      active: true,
+      config: { filename: join(dir, 'warehouse.db') },
+    };
+    const first = await boot({ objects: [boundTo('wh_note', 'warehouse')] });
+    await first.admin.createDatasource({ name: record.name, driver: record.driver, config: record.config });
+    // The state an install that created the datasource before this fix is in.
+    await first.knexOf('warehouse').schema.dropTableIfExists('wh_note');
+    await (first.ql.getDriverByName('warehouse') as IDataDriver).disconnect?.();
+
+    const restarted = await boot({ objects: [boundTo('wh_note', 'warehouse')], stored: [record] });
+
+    expect(await restarted.knexOf('warehouse').schema.hasTable('wh_note')).toBe(true);
+    await expect(restarted.read('wh_note')).resolves.toEqual([]);
+  });
+
+  it('control: an object bound to another datasource is not synced on this one', async () => {
+    const dir = scratch();
+    const { admin, synced } = await boot({
+      objects: [boundTo('wh_note', 'warehouse'), boundTo('crm_note', 'crm')],
+    });
+
+    await admin.createDatasource({ name: 'warehouse', driver: 'sqlite', config: { filename: join(dir, 'warehouse.db') } });
+
+    expect(synced.mock.calls.map(([name]) => name)).toEqual(['wh_note']);
+  });
+});
+
+describe('⛔ an external or federated datasource never gets DDL', () => {
+  it.each(['external', 'validate-only'] as const)(
+    'schemaMode %s: the pool opens, and no object bound to it is synced',
+    async (schemaMode) => {
+      const dir = scratch();
+      const { admin, synced, knexOf } = await boot({
+        objects: [
+          boundTo('lake_note', 'lake'),
+          boundTo('lake_orders', 'lake', { external: { remoteName: 'orders' } }),
+        ],
+      });
+
+      const created = await admin.createDatasource({
+        name: 'lake',
+        driver: 'sqlite',
+        schemaMode,
+        external: { validation: { onMismatch: 'warn' } },
+        config: { filename: join(dir, 'lake.db') },
+      } as never);
+
+      expect(created.status).toBe('ok');
+      expect(synced).not.toHaveBeenCalled();
+      expect(await knexOf('lake').schema.hasTable('lake_note')).toBe(false);
+    },
+  );
+});
@@ -20,6 +20,7 @@ import {
   type ConnectionEngineLike,
 } from './datasource-connection-service.js';
 import { datasourceConnectivityChanged } from './datasource-connectivity-change.js';
+import { resolveDatasourceSchemaMode } from './datasource-schema-mode.js';
 import type { DatasourceConnectPolicy } from './contracts/connect-policy.js';
 import type { Logger } from './logger.js';
 import { resolveTursoUrl } from './turso-driver-config.js';
@@ -402,6 +403,26 @@ export class DatasourceAdminServicePlugin implements Plugin {
     });
     ctx.registerService('datasource-connection', this.connection);
 
+    // [ADR-0015 §4.2 · #25277] The objects bound to a datasource, read from
+    // the engine's object registry (the `objectql` service's `registry`), the
+    // registry the metadata save door writes through to: an object saved
+    // through `PUT /api/v1/meta/object/:name` or imported from a remote table
+    // is in it the moment it is saved, and boot restores the stored rows into
+    // it. ⛔ Not the metadata service's `listObjects()`, which holds a copy of
+    // this registry taken once at boot and never sees a runtime save. A host
+    // with no engine registry serves no objects, so none can be bound.
+    //
+    // Resolved where it is used, never at `init()` (AGENTS.md, "Startup
+    // registry reads"). Two readers: the judgement of a datasource write
+    // (`listObjectsBoundTo`) and the pool's connect (`registerPool`, #27701).
+    const objectsBoundTo = (datasource: string): Array<{ name: unknown; external: unknown }> => {
+      const registry = safeGetService<Partial<Pick<IObjectQLEngine, 'registry'>>>(ctx, 'objectql')?.registry;
+      if (typeof registry?.getAllObjects !== 'function') return [];
+      return (registry.getAllObjects() as Array<{ name?: unknown; datasource?: unknown; external?: unknown }>)
+        .filter((o) => o?.datasource === datasource)
+        .map((o) => ({ name: o.name, external: o.external }));
+    };
+
     const config: DatasourceAdminServiceConfig = {
       probe: (input) => this.probe(factory(), input),
 
@@ -503,24 +524,9 @@ export class DatasourceAdminServicePlugin implements Plugin {
       },
 
       // [ADR-0015 §4.2 · #25277] The objects a datasource write is judged
-      // against. Read from the engine's object registry (the `objectql`
-      // service's `registry`), the registry the metadata save door writes
-      // through to: an object saved through `PUT /api/v1/meta/object/:name` or
-      // imported from a remote table is in it the moment it is saved, and boot
-      // restores the stored rows into it. ⛔ Not the metadata service's
-      // `listObjects()`, which holds a copy of this registry taken once at boot
-      // and never sees a runtime save. A host with no engine registry serves
-      // no objects, so none can be bound.
-      //
-      // Resolved where it is used, never at `init()` (AGENTS.md, "Startup
-      // registry reads").
-      listObjectsBoundTo: async (datasource) => {
-        const registry = safeGetService<Partial<Pick<IObjectQLEngine, 'registry'>>>(ctx, 'objectql')?.registry;
-        if (typeof registry?.getAllObjects !== 'function') return [];
-        return (registry.getAllObjects() as Array<{ name?: unknown; datasource?: unknown; external?: unknown }>)
-          .filter((o) => o?.datasource === datasource)
-          .map((o) => ({ name: o.name, external: o.external }));
-      },
+      // against, read from the engine's object registry (`objectsBoundTo`
+      // above says why that registry, and why per call).
+      listObjectsBoundTo: async (datasource) => objectsBoundTo(datasource),
 
       // Hot pool (de)registration converges on the shared
       // DatasourceConnectionService (ADR-0062 D1) — one connect path for code-
@@ -529,9 +535,26 @@ export class DatasourceAdminServicePlugin implements Plugin {
       // opens the connection, and registers the live driver + datasource def.
       // Runtime-admin connects always degrade-with-warning on failure (never
       // fail-fast), preserving the pre-ADR-0062 admin behavior.
+      //
+      // [#27701] The connect carries the objects already bound to a MANAGED
+      // datasource, so its sync loop gives them their tables, as boot's
+      // `connectDeclared` does for a declared datasource. Boot's schema sync
+      // ran before this driver existed and skipped them; without this a code
+      // object bound to a Setup-created datasource had no table, on create and
+      // after every restart, and every read of it was a 500. ⛔ A datasource
+      // whose schema ObjectStack does not own (`external`, `validate-only`)
+      // passes none: it never gets DDL. Its mode is read the way the driver
+      // factory reads it, so the gate and the driver agree on what it is.
       registerPool: async (record) => {
         this.livePoolRecords.set(record.name, record);
+        const managed = (resolveDatasourceSchemaMode({ ...record, config: record.config ?? {} }) ?? 'managed') === 'managed';
+        const objects = managed
+          ? objectsBoundTo(record.name)
+              .map((o) => o.name)
+              .filter((n): n is string => typeof n === 'string')
+          : [];
         await this.connection?.connect(record, {
+          objects,
           context: { origin: record.origin ?? 'runtime', trigger: 'runtime-admin' },
         });
       },