diff --git a/api/server/controllers/agents/__tests__/request.resumeMetadata.spec.js b/api/server/controllers/agents/__tests__/request.resumeMetadata.spec.js
index 1eb5d13a58..08bf86315d 100644
--- a/api/server/controllers/agents/__tests__/request.resumeMetadata.spec.js
+++ b/api/server/controllers/agents/__tests__/request.resumeMetadata.spec.js
@@ -11,6 +11,7 @@ const mockGenerationJobManager = {
   createJob: jest.fn(),
   emitError: jest.fn(),
   completeJob: jest.fn(),
+  abortJob: jest.fn(),
   getResumeState: jest.fn(),
   updateMetadata: jest.fn(),
   claimGeneration: jest.fn(),
@@ -880,6 +881,40 @@ describe('ResumableAgentController resume metadata', () => {
     expect(mockGenerationJobManager.createJob).not.toHaveBeenCalled();
   });
 
+  it('releases the pending slot and idempotency claim when the pre-start fence aborts', async () => {
+    // Manual Run Now is NOT limiter-exempt, so the controller incremented the pending
+    // counter and claimed the clientRequestId. The fence abort skips the whole
+    // generation, so nothing downstream releases either — this exit must, or the
+    // user's next interactive messages 429 until the counter's TTL expires.
+    const schedules = require('~/server/services/Schedules');
+    schedules.isScheduleLive.mockResolvedValueOnce(false);
+    mockGenerationJobManager.claimGeneration.mockResolvedValue({ claimed: true });
+    mockGenerationJobManager.abortJob.mockResolvedValue({ success: true });
+    const req = {
+      user: { id: 'user-123' },
+      _isScheduledFire: true,
+      _isManualScheduledFire: true,
+      body: {
+        text: 'Scheduled prompt',
+        messageId: 'user-msg',
+        clientRequestId: 'req-abc',
+        conversationId: 'conversation-123',
+        scheduleId: 'sched-1',
+        scheduledFor: '2026-07-28T12:00:00.000Z',
+        endpointOption: { endpoint: 'agents', modelOptions: { model: 'gpt-4.1' } },
+      },
+      config: {},
+    };
+    const res = createResumableResponse();
+
+    await AgentController(req, res, jest.fn(), jest.fn(), null);
+
+    expect(res.json).toHaveBeenCalledWith(expect.objectContaining({ status: 'aborted' }));
+    expect(mockCheckAndIncrementPendingRequest).toHaveBeenCalledWith('user-123');
+    expect(mockDecrementPendingRequest).toHaveBeenCalledWith('user-123');
+    expect(mockGenerationJobManager.releaseGeneration).toHaveBeenCalledWith('user-123', 'req-abc');
+  });
+
   it('does not finalize an unscoped generation when job creation rejects before returning', async () => {
     mockGenerationJobManager.claimGeneration.mockResolvedValue({ claimed: true });
     mockGenerationJobManager.createJob.mockRejectedValue(new Error('create failed before return'));
diff --git a/api/server/controllers/agents/request.js b/api/server/controllers/agents/request.js
index 9b60ee1094..f2a265bc17 100644
--- a/api/server/controllers/agents/request.js
+++ b/api/server/controllers/agents/request.js
@@ -477,6 +477,15 @@ const ResumableAgentController = async (req, res, next, initializeClient, addTit
           preserveForReconcile: !outcomeRecorded,
           expectedCreatedAt: jobCreatedAt,
         }).catch(() => undefined);
+        // Mirror the init-error cleanup: this return skips the whole generation, so the
+        // pending-request slot (manual Run Now is NOT limiter-exempt) and the
+        // idempotency claim it acquired must be released here — nothing downstream
+        // will. Leaving them held 429'd the user's next interactive messages until the
+        // counter's TTL expired.
+        if (ownsIdempotencyClaim) {
+          await GenerationJobManager.releaseGeneration(userId, clientRequestId).catch(() => {});
+        }
+        await finishResumableRequest(req, userId);
         startupTelemetry?.end('aborted');
         return res.json({ streamId, conversationId, status: 'aborted' });
       }
diff --git a/api/server/experimental.js b/api/server/experimental.js
index 6c65805dfd..d39d55aac0 100644
--- a/api/server/experimental.js
+++ b/api/server/experimental.js
@@ -253,16 +253,19 @@ if (cluster.isMaster) {
   let expiredFileSweepOptions = null;
   let expiredFileSweepStarted = false;
   let schedulesReady = false;
-  const SCHEDULE_WRITE_READ_ONLY_METHODS = new Set(['GET', 'HEAD', 'OPTIONS']);
+  const SCHEDULE_ENGINE_OPTIONAL_METHODS = new Set(['GET', 'HEAD', 'OPTIONS', 'DELETE']);
 
   /**
    * Refuse schedule WRITES in this clustered entrypoint. v1 supports single-process
    * scheduling only, and this process never arms the engine, so accepting a create or
    * run-now here would persist a schedule nothing will ever fire. Reads stay open so an
-   * operator can still inspect existing schedules.
+   * operator can still inspect existing schedules. DELETE stays open too: erasing a
+   * stored prompt needs no engine (the handler quiesces through the job store and
+   * refuses unsafe cases itself), and a deployment switched to clustered mode must not
+   * strand users with schedules they can see but never remove.
    */
   const rejectScheduleWritesUntilReady = (req, res, next) => {
-    if (schedulesReady || SCHEDULE_WRITE_READ_ONLY_METHODS.has(req.method)) {
+    if (schedulesReady || SCHEDULE_ENGINE_OPTIONAL_METHODS.has(req.method)) {
       return next();
     }
     return res.status(501).json({
diff --git a/api/server/index.js b/api/server/index.js
index 0fea179390..78b435c8b3 100644
--- a/api/server/index.js
+++ b/api/server/index.js
@@ -70,7 +70,7 @@ let schedulesReady = false;
 const SERVER_NOT_READY_CODE = 'SERVER_NOT_READY';
 const SCHEDULES_NOT_READY_CODE = 'SCHEDULES_NOT_READY';
 const CHAT_START_RETRY_AFTER_SECONDS = '1';
-const READ_ONLY_METHODS = new Set(['GET', 'HEAD', 'OPTIONS']);
+const SCHEDULE_ENGINE_OPTIONAL_METHODS = new Set(['GET', 'HEAD', 'OPTIONS', 'DELETE']);
 
 const rejectChatStartsUntilReady = (req, res, next) => {
   if (serverReady || req.method !== 'POST' || req.path === '/abort') {
@@ -89,10 +89,13 @@ const rejectChatStartsUntilReady = (req, res, next) => {
  * indexes) is up. With MONGO_AUTO_INDEX disabled those indexes are created only by
  * initializeScheduleEngine, which runs after the server starts listening — so a
  * create/run-now that lands in that window would persist without duplicate
- * protection. Reads stay open; writers get a 503 + Retry-After.
+ * protection. Reads stay open; writers get a 503 + Retry-After. DELETE stays open
+ * too: erasure needs none of those indexes, and an engine that REFUSES to arm keeps
+ * this gate closed forever — users must still be able to remove schedules (and their
+ * stored prompts) they can see.
  */
 const rejectScheduleWritesUntilReady = (req, res, next) => {
-  if (schedulesReady || READ_ONLY_METHODS.has(req.method)) {
+  if (schedulesReady || SCHEDULE_ENGINE_OPTIONAL_METHODS.has(req.method)) {
     return next();
   }
 
diff --git a/api/server/index.spec.js b/api/server/index.spec.js
index 2843429586..d849b600ef 100644
--- a/api/server/index.spec.js
+++ b/api/server/index.spec.js
@@ -160,6 +160,19 @@ describe('Startup readiness wiring', () => {
     );
   });
 
+  it('keeps schedule DELETE open when the engine never arms (both entrypoints)', () => {
+    // An engine that refuses to arm keeps the gate closed forever; erasure needs no
+    // engine, and users must not be stranded with schedules they can see but never
+    // remove. The set name is asserted so a revert to READ_ONLY_METHODS fails here.
+    const experimental = fs.readFileSync(path.join(__dirname, 'experimental.js'), 'utf8');
+    for (const entrypoint of [source, experimental]) {
+      expect(entrypoint).toMatch(
+        /SCHEDULE_ENGINE_OPTIONAL_METHODS = new Set\(\['GET', 'HEAD', 'OPTIONS', 'DELETE'\]\)/,
+      );
+      expect(entrypoint).toContain('SCHEDULE_ENGINE_OPTIONAL_METHODS.has(req.method)');
+    }
+  });
+
   it('flips schedulesReady only from the engine-init result (indexes must exist first)', () => {
     const engineInitIndex = source.indexOf(
       'const scheduleEngine = await initializeScheduleEngine(',
diff --git a/api/server/routes/agents/__tests__/abort.spec.js b/api/server/routes/agents/__tests__/abort.spec.js
index eb49e56c07..bf365124bf 100644
--- a/api/server/routes/agents/__tests__/abort.spec.js
+++ b/api/server/routes/agents/__tests__/abort.spec.js
@@ -24,6 +24,7 @@ const mockGenerationJobManager = {
 };
 
 const mockSaveMessage = jest.fn();
+const mockDeleteAgentCheckpoint = jest.fn(async () => undefined);
 const mockRecordScheduleOutcome = jest.fn(async () => true);
 const mockClearScheduledJob = jest.fn(async () => undefined);
 
@@ -41,6 +42,7 @@ jest.mock('@librechat/api', () => ({
   ...jest.requireActual('@librechat/api'),
   isEnabled: jest.fn().mockReturnValue(false),
   GenerationJobManager: mockGenerationJobManager,
+  deleteAgentCheckpoint: (...args) => mockDeleteAgentCheckpoint(...args),
 }));
 
 jest.mock('~/models', () => ({
@@ -558,6 +560,32 @@ describe('Agent Abort Endpoint', () => {
         // Nothing to damage, so pressing Stop as a turn finishes stays a quiet success.
         expect(response.status).toBe(200);
       });
+
+      it('fails closed when the replacement check cannot read the store', async () => {
+        // null from getJob means CONFIRMED absent; a THROW means unknown — a
+        // replacement may be live, and the conversation-wide checkpoint prune below
+        // would strip its resume state. With the store unreadable, the handler must
+        // stop before any side effect rather than treat the error as absence.
+        mockGenerationJobManager.getJob
+          .mockResolvedValueOnce({
+            metadata: { userId: 'test-user-123', pendingAction: { payload: {} } },
+            createdAt: 1000,
+          })
+          .mockRejectedValue(new Error('redis gone'));
+        mockGenerationJobManager.abortJob.mockResolvedValue({
+          success: false,
+          jobData: null,
+          content: [],
+        });
+
+        const response = await request(app)
+          .post('/api/agents/chat/abort')
+          .send({ conversationId: 'conv-1' });
+
+        expect(response.status).toBe(503);
+        expect(mockDeleteAgentCheckpoint).not.toHaveBeenCalled();
+        expect(mockSaveMessage).not.toHaveBeenCalled();
+      });
     });
 
     describe('Scheduled runs', () => {
diff --git a/api/server/routes/agents/index.js b/api/server/routes/agents/index.js
index 452e558708..8d8b2c604e 100644
--- a/api/server/routes/agents/index.js
+++ b/api/server/routes/agents/index.js
@@ -391,7 +391,20 @@ router.post('/chat/abort', configMiddleware, async (req, res) => {
     // is nothing to damage in that case, so it keeps its previous behaviour rather than
     // turning a routine stop into an error.
     if (!abortResult.success && abortResult.jobData == null) {
-      const liveJob = await GenerationJobManager.getJob(jobStreamId).catch(() => null);
+      // FAIL CLOSED on an unreadable store: null means "confirmed absent" (benign),
+      // but a thrown read means UNKNOWN — a replacement may be live, and everything
+      // below acts on the conversation as a whole (the checkpoint prune would strip
+      // its resume state). Nothing of ours was stopped (the abort already failed), so
+      // refusing here has no side effects and the client simply retries.
+      let liveJob;
+      try {
+        liveJob = await GenerationJobManager.getJob(jobStreamId);
+      } catch (err) {
+        logger.error(`[AgentStream] Could not verify abort state: ${jobStreamId}`, err);
+        return res
+          .status(503)
+          .json({ error: 'Could not verify the generation state. Please retry.', aborted: null });
+      }
       if (liveJob != null && liveJob.createdAt !== job.createdAt) {
         logger.debug(
           `[AgentStream] Abort refused: generation was replaced before it landed: ${jobStreamId}`,
diff --git a/config/delete-user.js b/config/delete-user.js
index 2b3f03ff08..55afa2e144 100644
--- a/config/delete-user.js
+++ b/config/delete-user.js
@@ -86,11 +86,13 @@ async function gracefulExit(code = 0) {
   // boundary (fireSchedule's isOwnerDeleting probe) from this point on, so anything the
   // count then misses cannot have started after it.
   //
-  // Through markUserDeleting, never a raw update: the barrier is only in force once no
-  // CACHED pre-barrier user document can still populate req.user on a live server, and
-  // that method is what drops the auth cache entry. It fails closed, so a cache it
-  // cannot reach aborts the deletion rather than proceeding behind a barrier that was
-  // never actually raised.
+  // Through markUserDeleting, never a raw update: that method also drops the auth
+  // user-doc cache entry, and fails closed — a cache it cannot reach aborts the
+  // deletion rather than proceeding behind a barrier that was never actually raised.
+  // With a shared (Redis) cache this invalidation reaches the live server. With the
+  // in-process cache no external script can, but the residual window is bounded by the
+  // 5s AUTH_USER_DOC_CACHE_TTL_MS, and the schedule-fire guard (isOwnerDeleting) reads
+  // Mongo directly rather than trusting req.user.
   try {
     await markUserDeleting(uid);
   } catch (err) {
diff --git a/packages/api/src/admin/users.spec.ts b/packages/api/src/admin/users.spec.ts
index 83ea1ca018..f1627016fa 100644
--- a/packages/api/src/admin/users.spec.ts
+++ b/packages/api/src/admin/users.spec.ts
@@ -580,6 +580,22 @@ describe('createAdminUsersHandlers', () => {
       expect(deps.deleteUserById).not.toHaveBeenCalled();
     });
 
+    it('refuses the delete when removing the schedule rows fails', async () => {
+      const deps = createDeps({
+        deleteSchedulesByUser: jest.fn().mockRejectedValue(new Error('mongo down')),
+      });
+      const handlers = createAdminUsersHandlers(deps);
+      const { req, res, status } = createReqRes({ params: { id: validUserId } });
+
+      await handlers.deleteUser(req, res);
+
+      // Deleting the user doc first would make this endpoint 404 on retry, so the
+      // leftover schedule rows (the deleted user's prompt text, no TTL) would become
+      // unretryable — permanent retention in the clustered topology with no sweep.
+      expect(status).toHaveBeenCalledWith(503);
+      expect(deps.deleteUserById).not.toHaveBeenCalled();
+    });
+
     it('returns 500 on error', async () => {
       const deps = createDeps({
         deleteUserById: jest.fn().mockRejectedValue(new Error('db crash')),
diff --git a/packages/api/src/admin/users.ts b/packages/api/src/admin/users.ts
index 9bdb1f770e..6dc1cae018 100644
--- a/packages/api/src/admin/users.ts
+++ b/packages/api/src/admin/users.ts
@@ -206,10 +206,24 @@ export function createAdminUsersHandlers(deps: AdminUsersDeps): {
       // Hard-delete the schedule rows rather than relying on the reconciler's
       // `deleting` sweep: the clustered `experimental.js` entrypoint never arms the
       // engine, so in that topology nothing would ever erase them and the deleted
-      // user's prompt text would persist indefinitely.
-      await deleteSchedulesByUser(id).catch((error) => {
-        logger.error('[adminUsers] Failed to delete schedules for the removed user', error);
-      });
+      // user's prompt text would persist indefinitely. REFUSE on failure, before the
+      // user document goes: once it is deleted this endpoint 404s on retry, making the
+      // leftover rows unretryable — the exact retention this hard delete exists to
+      // prevent. The barrier is up and quiesce confirmed, so refusing here is safe and
+      // the admin simply retries. Mirrors the self-service controller's ordering.
+      const schedulesDeleted = await deleteSchedulesByUser(id).then(
+        () => true,
+        (error) => {
+          logger.error('[adminUsers] Failed to delete schedules for the removed user', error);
+          return false;
+        },
+      );
+      if (!schedulesDeleted) {
+        res.set('Retry-After', '30');
+        return res.status(503).json({
+          error: 'Could not remove scheduled chats for this user. Please retry shortly.',
+        });
+      }
 
       const result = await deleteUserById(id);
 
diff --git a/packages/api/src/schedules/fire.spec.ts b/packages/api/src/schedules/fire.spec.ts
index 161464a707..f55b8725a5 100644
--- a/packages/api/src/schedules/fire.spec.ts
+++ b/packages/api/src/schedules/fire.spec.ts
@@ -287,6 +287,11 @@ describe('fireSchedule', () => {
     expect(calls.recordOutcome).toHaveLength(0);
     // The occurrence is done, so the schedule still advances past it.
     expect(calls.advance).toBe(1);
+    // The lease is handed back explicitly: an owner EDIT rotates the claim token (so the
+    // token-fenced advance no-ops) but never touches the lease fields — without this
+    // release, Run Now and the next claim of the recomputed occurrence stay blocked for
+    // the full 5-minute lease TTL.
+    expect(methods.releaseLeaseByHolder).toHaveBeenCalledWith('sched-1', 'inst-1');
   });
 
   it('treats a message-limiter 429 as a skip, not a schedule failure', async () => {
@@ -484,6 +489,35 @@ describe('fireSchedule', () => {
     expect(methods.releaseLeaseByHolder).toHaveBeenCalledWith('sched-1', 'inst-1');
   });
 
+  /**
+   * A failed revalidation must NEVER advance. In every true supersession (takeover,
+   * owner edit) the claim token rotated, so an advance would no-op anyway — the only
+   * case where it can land is a PURE lease expiry with no takeover, where the token
+   * never rotated. advanceSchedule checks no lease, so that advance moved nextRunAt
+   * past an occurrence nothing had fired: a slow preflight silently DROPPED it
+   * instead of leaving it due for the next claim to retry.
+   */
+  it('leaves the occurrence due when only the lease expired (no takeover)', async () => {
+    const { methods } = makeMethods();
+    // Same claim token, merely past leaseUntil: revalidateClaim fails on the lease
+    // predicate alone while the token-fenced advance filter would still MATCH.
+    (methods.revalidateClaim as jest.Mock).mockResolvedValue(false);
+    mockFetch(async () => okResponse());
+    const result = await fireSchedule(makeDeps(methods), makeSchedule(), LIMITS, dueAt());
+    expect(result.skipped).toBe('superseded');
+    expect(methods.advanceSchedule).not.toHaveBeenCalled();
+    expect(global.fetch).not.toHaveBeenCalled();
+  });
+
+  it('never advances from any superseded branch, including post-reserve', async () => {
+    const { methods } = makeMethods();
+    (methods.revalidateClaim as jest.Mock).mockResolvedValueOnce(true).mockResolvedValue(false);
+    mockFetch(async () => okResponse());
+    const result = await fireSchedule(makeDeps(methods), makeSchedule(), LIMITS, dueAt());
+    expect(result.skipped).toBe('superseded');
+    expect(methods.advanceSchedule).not.toHaveBeenCalled();
+  });
+
   /**
    * The lease can expire between the pre-reserve revalidation and the reservation
    * itself: the capacity allocator reads occupancy in between, and the preflight
diff --git a/packages/api/src/schedules/fire.ts b/packages/api/src/schedules/fire.ts
index 5c0e7fc88c..2505883ec3 100644
--- a/packages/api/src/schedules/fire.ts
+++ b/packages/api/src/schedules/fire.ts
@@ -268,6 +268,25 @@ export async function fireSchedule(
     }
   };
 
+  /**
+   * Steps aside after a failed revalidation WITHOUT advancing. `advance()` here is a
+   * designed no-op in every true supersession (takeover, owner edit — both rotate the
+   * claim token, so its token-fenced filter misses), which means the only case where it
+   * actually LANDS is the one it must not: a PURE lease expiry with no takeover, where
+   * the token never rotated. advanceSchedule checks no lease, so the advance moved
+   * nextRunAt past an occurrence nothing had fired — a preflight outlasting the lease
+   * silently dropped it instead of leaving it due for the next claim to retry.
+   * Manual run-now still releases its serialization lease (release-only, no advance)
+   * so a repeat click isn't met with a stale "already in progress".
+   */
+  const stepAsideSuperseded = async () => {
+    await releaseSupersededLease();
+    if (options?.manual) {
+      await methods.releaseLease(schedule.id, claimToken);
+    }
+    return { fired: false, skipped: 'superseded' as const };
+  };
+
   if (nextRunAt == null) {
     await methods.disableSchedule(schedule.id, 'invalid_schedule', claimToken);
     await advance();
@@ -354,9 +373,7 @@ export async function fireSchedule(
         claimToken != null &&
         !(await methods.revalidateClaim(schedule.id, claimToken, !options?.manual))
       ) {
-        await releaseSupersededLease();
-        await advance();
-        return { fired: false, skipped: 'superseded' as const };
+        return stepAsideSuperseded();
       }
       // Skip rows carry no `bookkept:false` marker, so the reconciler has no path to
       // repair a half-applied skip. If the schedule-side bookkeeping throws, do NOT
@@ -411,9 +428,7 @@ export async function fireSchedule(
       claimToken != null &&
       !(await methods.revalidateClaim(schedule.id, claimToken, !options?.manual))
     ) {
-      await releaseSupersededLease();
-      await advance();
-      return { fired: false, skipped: 'superseded' as const };
+      return stepAsideSuperseded();
     }
 
     // Pre-generate the conversation id and reserve the run row up front. The
@@ -532,9 +547,7 @@ export async function fireSchedule(
       !(await methods.revalidateClaim(schedule.id, claimToken, !options?.manual))
     ) {
       await rollbackReservation(conversationId);
-      await releaseSupersededLease();
-      await advance();
-      return { fired: false, skipped: 'superseded' as const };
+      return stepAsideSuperseded();
     }
 
     try {
@@ -565,10 +578,14 @@ export async function fireSchedule(
       }
       if (error instanceof ScheduleFireError && error.preStartAbort) {
         // The controller refused this fire at its own liveness/revision fence and has
-        // already recorded the occurrence's outcome. Advancing is all that's left:
-        // recording `error` here would count a delete/edit as a schedule FAULT and walk
-        // it toward auto-disable.
+        // already recorded the occurrence's outcome. Recording `error` here would count
+        // a delete/edit as a schedule FAULT and walk it toward auto-disable.
+        // Release OUR lease explicitly: the fence tripping on an owner EDIT rotated the
+        // claim token, so the token-fenced advance() below no-ops — but the edit never
+        // touched the lease fields, and leaving them held blocks Run Now and the next
+        // claim of the recomputed occurrence for the full 5-minute TTL.
         logger.info(`[schedules] fire aborted pre-start for ${schedule.id} (superseded)`);
+        await releaseSupersededLease();
         await advance();
         return { fired: false, skipped: 'superseded' as const };
       }
diff --git a/packages/api/src/schedules/service.spec.ts b/packages/api/src/schedules/service.spec.ts
index db308989f9..9079727d40 100644
--- a/packages/api/src/schedules/service.spec.ts
+++ b/packages/api/src/schedules/service.spec.ts
@@ -431,6 +431,26 @@ describe('global kill switch', () => {
     const service = makeService(noRuns(), getAppConfig);
     expect(await service.engineDeps.isGloballyDisabled()).toBe(false);
   });
+
+  it('trips on the object form `{ use: false }` exactly like the boolean stop', async () => {
+    // Both stop shapes must FREEZE occurrences (engine stops claiming, nothing
+    // advances). A shape-blind gate left the engine claiming while getLimits refused
+    // fires, so the disabled path ADVANCED each occurrence — a short maintenance stop
+    // silently dropped everything it covered instead of leaving it due.
+    const getAppConfig = jest.fn(async () => ({
+      interfaceConfig: { schedules: { use: false, maxPerUser: 5 } },
+    })) as unknown as SchedulesServiceDeps['getAppConfig'];
+    const service = makeService(noRuns(), getAppConfig);
+    expect(await service.engineDeps.isGloballyDisabled()).toBe(true);
+  });
+
+  it('does not trip on the object form while `use` stays enabled', async () => {
+    const getAppConfig = jest.fn(async () => ({
+      interfaceConfig: { schedules: { use: true, maxPerUser: 5 } },
+    })) as unknown as SchedulesServiceDeps['getAppConfig'];
+    const service = makeService(noRuns(), getAppConfig);
+    expect(await service.engineDeps.isGloballyDisabled()).toBe(false);
+  });
 });
 
 describe('deployment-wide limits', () => {
diff --git a/packages/api/src/schedules/service.ts b/packages/api/src/schedules/service.ts
index eb1b8856b5..8bb35844bc 100644
--- a/packages/api/src/schedules/service.ts
+++ b/packages/api/src/schedules/service.ts
@@ -1,4 +1,4 @@
-import { logger, runAsSystem, tenantStorage } from '@librechat/data-schemas';
+import { logger, runAsSystem, tenantStorage, isRuntimeDisabled } from '@librechat/data-schemas';
 import { getRefillEligibilityDate, Permissions, PermissionTypes } from 'librechat-data-provider';
 import type { ScheduleMethods, AppConfig, IBalance } from '@librechat/data-schemas';
 import type { TCheckpointerConfig } from 'librechat-data-provider';
@@ -463,8 +463,14 @@ export function createSchedulesService(deps: SchedulesServiceDeps): SchedulesSer
       // BASE config only: DB principal overrides can narrow availability but must never
       // widen past an operator's global stop, so `schedules: false` in librechat.yaml is
       // genuinely non-overridable rather than emergent from the override filters.
+      // isRuntimeDisabled reads BOTH stop shapes (`false` and `{ use: false }`) — the
+      // same predicate the override merge preserves base stops with. A shape-blind
+      // check here made the object form disable getLimits but not this gate, so the
+      // engine kept claiming and fireSchedule ADVANCED each occurrence: a short
+      // maintenance stop silently dropped every occurrence it covered instead of
+      // leaving them due.
       const base = await deps.getAppConfig({ baseOnly: true });
-      return base?.interfaceConfig?.schedules === false;
+      return isRuntimeDisabled(base?.interfaceConfig?.schedules);
     },
     // Occupancy is read in SYSTEM scope so the cap is global across tenants (the
     // owner's tenant context would only see its own runs); the claim itself stays in
diff --git a/packages/data-schemas/src/app/resolution.ts b/packages/data-schemas/src/app/resolution.ts
index 23af163488..372921a7a7 100644
--- a/packages/data-schemas/src/app/resolution.ts
+++ b/packages/data-schemas/src/app/resolution.ts
@@ -280,8 +280,10 @@ export function mergeConfigOverrides(baseConfig: AppConfig, configs: IConfig[]):
   return preserveRuntimeStops(baseConfig, merged);
 }
 
-/** Whether a runtime-config interface field reads as OFF in either of its two shapes. */
-function isRuntimeDisabled(value: unknown): boolean {
+/** Whether a runtime-config interface field reads as OFF in either of its two shapes:
+ *  boolean `false`, or the object form with `use: false`. Exported so runtime gates
+ *  (e.g. the schedule engine's global stop) apply the same semantics as the merge. */
+export function isRuntimeDisabled(value: unknown): boolean {
   if (value === false) {
     return true;
   }