Keep failed chat history available for retry (#8278)

* fix(app): keep failed chat history until retry

Automatic retries repeatedly reopen provider sessions after persistent failures. Keep the error visible until the user retries, while retaining reconnect recovery.

* fix(app): catch up chats after subscription failure

A failed replacement subscription leaves chats unsubscribed, so discard its old catch-up state and fetch missed history after retry.

* fix(app): recover gaps during history retry

Keep gap recovery active while a user-initiated retry is running so the chat is not marked current before its follow-up fetch.

* fix(app): show history errors in archived chats

Cached archived chats need the same manual retry affordance when history synchronization fails.

packages/app/src/contexts/session-context.tsx

@@ -455,10 +455,6 @@ function SessionProviderInternal({ children, serverId, client }: SessionProvider
       reportError: (error) => {
         console.warn("[Session] viewed timeline synchronization failed", { serverId, error });
       },
-      schedule: (task, delayMs) => {
-        const timeout = setTimeout(task, delayMs);
-        return () => clearTimeout(timeout);
-      },
     });
     viewedTimelineSyncRef.current = sync;
     setViewedTimelineSync(serverId, sync);

packages/app/src/hooks/use-agent-screen-state-machine.test.ts

@@ -422,6 +422,23 @@ describe("deriveAgentScreenViewState", () => {
     expect(result.memory.lastReadyAgent).toBeNull();
   });
 
+  it.each(["error", "retrying"] as const)("shows archived history sync %s", (status) => {
+    const result = deriveAgentScreenViewState({
+      input: {
+        ...createBaseInput(),
+        agent: createAgent("agent-1"),
+        isArchived: true,
+        hasHydratedHistoryBefore: true,
+        visibilityCatchUpStatus: status,
+      },
+      memory: createBaseMemory(),
+    });
+    expect(expectReadyState(result.state).sync).toEqual({
+      status: "sync_error",
+      isRetrying: status === "retrying",
+    });
+  });
+
   it("renders an archived agent before provider history is initialized", () => {
     const result = deriveAgentScreenViewState({
       input: {

packages/app/src/hooks/use-agent-screen-state-machine.ts

@@ -175,10 +175,14 @@ function resolveAgentScreenSync(args: {
   hadInitialSyncFailure: boolean;
 }): AgentScreenReadySyncState {
   const { input, hadInitialSyncFailure } = args;
-  if (input.isArchived) {
+  if (
+    input.isArchived &&
+    input.visibilityCatchUpStatus !== "error" &&
+    input.visibilityCatchUpStatus !== "retrying"
+  ) {
     return { status: "idle" };
   }
-  if (!input.isConnected) {
+  if (!input.isConnected && !input.isArchived) {
     return { status: "reconnecting" };
   }
   if (input.missingAgentState.kind === "error") {

packages/app/src/stores/workspace-layout-store.test.ts

@@ -174,7 +174,6 @@ it("feeds restored layout to timeline sync as a release signal, not a subscripti
     fetchLatestTail: async () => ({ hasNewer: false, endCursor: null }),
     onCatchUpEnded: () => undefined,
     reportError: () => undefined,
-    schedule: () => () => undefined,
   });
   // The edge session-context.tsx owns. Pointing it at subscription instead would resume every
   // agent behind a restored tab on the daemon.

packages/app/src/timeline/replica.test.ts

@@ -57,7 +57,6 @@ function createOwner(storage: TimelineReplicaStorage): ViewedTimelineOwner {
       fetchPage: async () => ({ hasNewer: false, endCursor: null }),
       fetchLatestTail: async () => ({ hasNewer: false, endCursor: null }),
       reportError: () => undefined,
-      schedule: () => () => undefined,
     },
   });
 }
@@ -617,7 +616,6 @@ describe("create handoff lifetime", () => {
         },
         fetchLatestTail: async () => ({ hasNewer: false, endCursor: null }),
         reportError: () => undefined,
-        schedule: () => () => undefined,
       },
     });
 

packages/app/src/timeline/viewed-timeline-sync.test.ts

@@ -90,7 +90,7 @@ test("visible catch-up settles before hidden open chats start fetching", async (
   world.sync.dispose();
 });
 
-test("a failed visible catch-up releases hidden chats without bypassing its retry delay", async () => {
+test("a failed visible catch-up releases hidden chats without bypassing its error", async () => {
   const world = new TimelineWorld();
   world.openedEarlier(["a-hidden"]);
   world.sync.replaceVisibleAgentIds("pane", ["z-visible"]);
@@ -103,7 +103,7 @@ test("a failed visible catch-up releases hidden chats without bypassing its retr
   hidden.respond({ hasNewer: false });
   await world.nextError();
   world.expectNoPendingFetch();
-  (await world.nextRetry())();
+  world.sync.retryVisibleAgentTimeline("z-visible");
   (await world.nextFetch("z-visible")).respond({ hasNewer: false });
   world.sync.dispose();
 });
@@ -200,20 +200,6 @@ class TimelineWorld {
       const waiter = this.errorWaiters.shift();
       if (waiter) waiter(this.errors.at(-1) ?? "");
     },
-    schedule: (task, delayMs) => {
-      const scheduled = { task, delayMs };
-      this.scheduled.push(scheduled);
-      const waiterIndex = this.retryWaiters.findIndex((waiter) => waiter.delayMs === delayMs);
-      if (waiterIndex >= 0) {
-        const [waiter] = this.retryWaiters.splice(waiterIndex, 1);
-        this.scheduled.splice(this.scheduled.indexOf(scheduled), 1);
-        waiter.resolve(task);
-      }
-      return () => {
-        const index = this.scheduled.indexOf(scheduled);
-        if (index >= 0) this.scheduled.splice(index, 1);
-      };
-    },
   });
 
   show(sourceId: string, agentIds: string[]): void {
@@ -235,12 +221,6 @@ class TimelineWorld {
   }> = [];
   private readonly errorWaiters: Array<(message: string) => void> = [];
   private readonly cacheRequestWaiters: Array<(agentId: string) => void> = [];
-  private readonly scheduled: Array<{ task: () => void; delayMs: number }> = [];
-  private readonly retryWaiters: Array<{
-    delayMs: number;
-    resolve(retry: () => void): void;
-  }> = [];
-
   get pendingFetchCount(): number {
     return this.fetches.length;
   }
@@ -300,22 +280,6 @@ class TimelineWorld {
     return new Promise((resolve) => this.errorWaiters.push(resolve));
   }
 
-  nextRetry(delayMs = 1_000): Promise<() => void> {
-    const index = this.scheduled.findIndex((entry) => entry.delayMs === delayMs);
-    if (index >= 0) return Promise.resolve(this.scheduled.splice(index, 1)[0].task);
-    return new Promise((resolve) => this.retryWaiters.push({ delayMs, resolve }));
-  }
-
-  elapse(elapsedMs: number): void {
-    const due = this.scheduled.filter((entry) => entry.delayMs <= elapsedMs);
-    this.scheduled.splice(
-      0,
-      this.scheduled.length,
-      ...this.scheduled.filter((entry) => entry.delayMs > elapsedMs),
-    );
-    for (const entry of due) entry.task();
-  }
-
   private releaseMembershipWaiter(): void {
     const waiter = this.membershipWaiters.shift();
     if (!waiter) return;
@@ -483,6 +447,28 @@ test("reports a chat current only once its parked follow-up page completes", asy
   world.sync.dispose();
 });
 
+test("a gap during manual retry is recovered before reporting ready", async () => {
+  const world = new TimelineWorld();
+  world.sync.setConnected(true);
+  world.show("workspace", ["agent-a"]);
+  (await world.nextMembership()).succeed();
+  (await world.nextFetch("agent-a")).fail("unavailable");
+  await world.nextError();
+  world.sync.retryVisibleAgentTimeline("agent-a");
+  const retry = await world.nextFetch("agent-a");
+  world.sync.recoverGap("agent-a", { epoch: "epoch-agent-a", endSeq: 9 });
+  retry.respond({ hasNewer: false });
+  await vi.waitFor(() => expect(world.pendingFetchCount).toBe(1));
+  const recovery = await world.nextFetch("agent-a");
+  expect(recovery.request).toMatchObject({
+    direction: "after",
+    cursor: { epoch: "epoch-agent-a", seq: 9 },
+  });
+  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("retrying");
+  recovery.respond({ hasNewer: false });
+  await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
+});
+
 test("reports a chat current once the latest-tail fallback replaces an overflowing catch-up", async () => {
   const world = new TimelineWorld();
   world.cursors.set("agent-a", { epoch: "epoch-agent-a", endSeq: 42 });
@@ -780,6 +766,21 @@ test("overlapping sources deduplicate membership and retain open hidden chats",
   world.expectNoPendingFetch();
 });
 
+test("failed timeline fetch does not schedule a retry", async () => {
+  vi.useFakeTimers();
+  const world = new TimelineWorld();
+  world.sync.setConnected(true);
+  world.show("workspace", ["agent-a"]);
+  (await world.nextMembership()).succeed();
+  (await world.nextFetch("agent-a")).fail("unavailable");
+  await world.nextError();
+  expect(vi.getTimerCount()).toBe(0);
+  await vi.advanceTimersByTimeAsync(120_000);
+  world.expectNoPendingFetch();
+  world.expectNoPendingMembership();
+  vi.useRealTimers();
+});
+
 test("a failed catch-up reports once and retries through the explicit retry policy", async () => {
   const world = new TimelineWorld();
   world.sync.setConnected(true);
@@ -788,12 +789,12 @@ test("a failed catch-up reports once and retries through the explicit retry poli
   membership.succeed();
   const failed = await world.nextFetch("agent-a");
   failed.fail("timeline unavailable");
-  const [error, retryCatchUp] = await Promise.all([world.nextError(), world.nextRetry()]);
+  const error = await world.nextError();
   expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
 
-  retryCatchUp();
+  world.sync.retryVisibleAgentTimeline("agent-a");
   const retry = await world.nextFetch("agent-a");
-  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
+  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("retrying");
   retry.respond({ hasNewer: false });
   await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
 
@@ -804,35 +805,6 @@ test("a failed catch-up reports once and retries through the explicit retry poli
   world.expectNoPendingMembership();
 });
 
-test("a failed catch-up retries with exponential backoff", async () => {
-  const world = new TimelineWorld();
-  world.sync.setConnected(true);
-  world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
-  const membership = await world.nextMembership();
-  membership.succeed();
-
-  const first = await world.nextFetch("agent-a");
-  first.fail("timeline unavailable");
-  const [firstError, retryAfterFirstFailure] = await Promise.all([
-    world.nextError(),
-    world.nextRetry(),
-  ]);
-  expect(firstError).toBe("timeline unavailable");
-  retryAfterFirstFailure();
-
-  const second = await world.nextFetch("agent-a");
-  second.fail("timeline unavailable");
-  await world.nextError();
-
-  const retryAfterSecondFailure = await world.nextRetry(2_000);
-  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
-
-  retryAfterSecondFailure();
-  const third = await world.nextFetch("agent-a");
-  third.respond({ hasNewer: false });
-  await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
-});
-
 test("manual retries can immediately re-attempt a failed catch-up", async () => {
   const world = new TimelineWorld();
   world.sync.setConnected(true);
@@ -842,7 +814,7 @@ test("manual retries can immediately re-attempt a failed catch-up", async () =>
 
   const failed = await world.nextFetch("agent-a");
   failed.fail("timeline unavailable");
-  await world.nextRetry();
+  await world.nextError();
   expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
 
   world.sync.retryVisibleAgentTimeline("agent-a");
@@ -853,7 +825,7 @@ test("manual retries can immediately re-attempt a failed catch-up", async () =>
   await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
 });
 
-test("redeclaring unchanged visibility does not bypass catch-up backoff", async () => {
+test("redeclaring unchanged visibility does not bypass catch-up error", async () => {
   const world = new TimelineWorld();
   world.sync.setConnected(true);
   world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
@@ -862,9 +834,11 @@ test("redeclaring unchanged visibility does not bypass catch-up backoff", async
 
   const failed = await world.nextFetch("agent-a");
   failed.fail("timeline unavailable");
-  await world.nextRetry();
+  await world.nextError();
 
   world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
+  world.sync.setActive(false);
+  world.sync.setActive(true);
 
   world.expectNoPendingMembership();
   world.expectNoPendingFetch();
@@ -880,7 +854,7 @@ test("a manual retry that fails returns to the error state", async () => {
 
   const failed = await world.nextFetch("agent-a");
   failed.fail("timeline unavailable");
-  await world.nextRetry();
+  await world.nextError();
 
   world.sync.retryVisibleAgentTimeline("agent-a");
   expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("retrying");
@@ -943,20 +917,34 @@ test("repeated recovery for the same running gap reuses the in-flight fetch", as
   gapPage.respond({ hasNewer: false });
 });
 
-test("membership failure autonomously retries without another visibility declaration", async () => {
+test("failed membership does not schedule a retry", async () => {
+  vi.useFakeTimers();
+  const world = new TimelineWorld();
+  world.sync.setConnected(true);
+  world.show("workspace", ["agent-a"]);
+  (await world.nextMembership()).fail("unavailable");
+  await world.nextError();
+  expect(vi.getTimerCount()).toBe(0);
+  await vi.advanceTimersByTimeAsync(120_000);
+  world.expectNoPendingFetch();
+  world.expectNoPendingMembership();
+  vi.useRealTimers();
+});
+
+test("membership failure waits for manual retry", async () => {
   const world = new TimelineWorld();
   world.sync.setConnected(true);
   world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
   const failed = await world.nextMembership();
   failed.fail("subscription unavailable");
-  const [error, retryMembership] = await Promise.all([world.nextError(), world.nextRetry()]);
+  const error = await world.nextError();
   expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
 
-  retryMembership();
+  world.sync.retryVisibleAgentTimeline("agent-a");
   const retry = await world.nextMembership();
   retry.succeed();
   const catchUp = await world.nextFetch("agent-a");
-  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
+  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("retrying");
   catchUp.respond({ hasNewer: false });
   await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
 
@@ -967,44 +955,68 @@ test("membership failure autonomously retries without another visibility declara
   });
 });
 
-test("membership failures retry with exponential backoff", async () => {
+test("retrying membership restores a previously loaded chat", async () => {
   const world = new TimelineWorld();
   world.sync.setConnected(true);
-  world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
-
-  const first = await world.nextMembership();
-  first.fail("subscription unavailable");
-  const [firstError, retryAfterFirstFailure] = await Promise.all([
-    world.nextError(),
-    world.nextRetry(),
-  ]);
-  expect(firstError).toBe("subscription unavailable");
-
-  retryAfterFirstFailure();
-  const second = await world.nextMembership();
-  second.fail("subscription unavailable again");
-
-  const retryAfterSecondFailure = await world.nextRetry(2_000);
-  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
-
-  retryAfterSecondFailure();
-  const retry = await world.nextMembership();
-  retry.succeed();
-  const catchUp = await world.nextFetch("agent-a");
-  catchUp.respond({ hasNewer: false });
+  world.show("workspace", ["agent-a"]);
+  (await world.nextMembership()).succeed();
+  (await world.nextFetch("agent-a")).respond({ hasNewer: false });
+  await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
+  world.cursors.set("agent-a", { epoch: "epoch-agent-a", endSeq: 10 });
+  world.show("workspace", ["agent-b"]);
+  (await world.nextMembership()).fail("subscription unavailable");
+  await world.nextError();
+  world.sync.retryVisibleAgentTimeline("agent-a");
+  (await world.nextMembership()).succeed();
+  (await world.nextFetch("agent-b")).respond({ hasNewer: false });
+  const resumed = await world.nextFetch("agent-a");
+  expect(resumed.request).toMatchObject({
+    direction: "after",
+    cursor: { epoch: "epoch-agent-a", seq: 10 },
+  });
+  resumed.respond({ hasNewer: false });
   await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("ready"));
 });
 
-test("redeclaring unchanged visibility does not bypass membership backoff", async () => {
+test("a fetch that finishes after membership failure cannot clear the error", async () => {
+  const world = new TimelineWorld();
+  world.sync.setConnected(true);
+  world.show("workspace", ["agent-a"]);
+  (await world.nextMembership()).succeed();
+  const inFlight = await world.nextFetch("agent-a");
+  world.show("workspace", ["agent-b"]);
+  (await world.nextMembership()).fail("subscription unavailable");
+  await world.nextError();
+  inFlight.respond({ hasNewer: false });
+  await Promise.resolve();
+  await Promise.resolve();
+  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error");
+});
+
+test("a failed membership retry returns to error", async () => {
+  const world = new TimelineWorld();
+  world.sync.setConnected(true);
+  world.show("workspace", ["agent-a"]);
+  (await world.nextMembership()).fail("unavailable");
+  await world.nextError();
+  world.sync.retryVisibleAgentTimeline("agent-a");
+  expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("retrying");
+  (await world.nextMembership()).fail("still unavailable");
+  await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-a")).toBe("error"));
+});
+
+test("redeclaring unchanged visibility does not bypass membership error", async () => {
   const world = new TimelineWorld();
   world.sync.setConnected(true);
   world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
 
   const failed = await world.nextMembership();
   failed.fail("subscription unavailable");
-  await world.nextRetry();
+  await world.nextError();
 
   world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
+  world.sync.setActive(false);
+  world.sync.setActive(true);
 
   world.expectNoPendingMembership();
   world.expectNoPendingFetch();
@@ -1050,7 +1062,6 @@ test("backgrounding preserves subscriptions and returning verifies the visible c
 
   world.sync.setActive(false);
   world.expectNoPendingMembership();
-  world.elapse(60_000);
   world.sync.setActive(true);
 
   world.expectNoPendingMembership();
@@ -1067,11 +1078,10 @@ test("stale membership retry cannot overwrite a newer effective set", async () =
   world.show("workspace", ["agent-a"]);
   const failed = await world.nextMembership();
   failed.fail("subscription unavailable");
-  const staleRetry = await world.nextRetry();
+  await world.nextError();
 
   world.show("workspace", ["agent-b"]);
   const current = await world.nextMembership();
-  staleRetry();
   current.succeed();
   const agentB = await world.nextFetch("agent-b");
   agentB.respond({ hasNewer: false });
@@ -1088,10 +1098,9 @@ test("membership retry cannot run while disconnected", async () => {
   world.sync.replaceVisibleAgentIds("workspace", ["agent-a"]);
   const failed = await world.nextMembership();
   failed.fail("subscription unavailable");
-  const disconnectedRetry = await world.nextRetry();
+  await world.nextError();
 
   world.sync.setConnected(false);
-  disconnectedRetry();
   world.expectNoPendingMembership();
   world.sync.setConnected(true);
   const restored = await world.nextMembership();
@@ -1118,7 +1127,6 @@ test("returning to an open hidden agent stays live after inactivity", async () =
   agentBCatchUp.respond({ hasNewer: false });
   await vi.waitFor(() => expect(world.sync.getAgentTimelineStatus("agent-b")).toBe("ready"));
 
-  world.elapse(60_000);
   world.show("workspace", ["agent-a"]);
 
   world.expectNoPendingMembership();

packages/app/src/timeline/viewed-timeline-sync.ts

@@ -329,7 +329,6 @@ export interface ViewedTimelineSyncPorts {
    */
   onCatchUpEnded(agentId: string): void;
   reportError(error: unknown): void;
-  schedule(task: () => void, delayMs: number): () => void;
 }
 
 export type ViewedTimelineStatus = "ready" | "pending" | "error" | "retrying";
@@ -423,26 +422,14 @@ export function createViewedTimelineOwner(input: {
   };
 }
 
-const RETRY_DELAY_MS = 1_000;
-const MAX_RETRY_DELAY_MS = 30_000;
-
 type CatchUpStatus = "running" | "complete" | "error";
 
 interface CatchUpState {
   generation: number;
   status: CatchUpStatus;
   request?: ProjectedTimelineForwardFetchPlan;
-  cancelRetry?: () => void;
-  retryDelayMs?: number;
 }
 
-const getNextRetryDelayMs = (previousDelayMs: number | undefined): number => {
-  if (previousDelayMs == null) {
-    return RETRY_DELAY_MS;
-  }
-  return Math.min(previousDelayMs * 2, MAX_RETRY_DELAY_MS);
-};
-
 function isSameCatchUpRequest(
   left: ProjectedTimelineForwardFetchPlan | undefined,
   right: ProjectedTimelineForwardFetchPlan | undefined,
@@ -472,9 +459,7 @@ function decideCatchUp(input: {
     }
     return "replace";
   }
-  return input.current.status === "running" || input.current.status === "complete"
-    ? "keep"
-    : "replace";
+  return "keep";
 }
 
 function normalizeAgentIds(agentIds: string[]): string[] {
@@ -494,8 +479,7 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
   const pendingCatchUps = new Map<string, ProjectedTimelineForwardFetchPlan>();
   const visibilityCatchUpPending = new Set<string>();
   const visibilityCatchUpErrors = new Map<string, string>();
-  // User-initiated retries only. Background retries stay silent; a retry the user asked for
-  // owes them a pending state until it settles.
+  // A user-initiated retry owes a pending state until it settles.
   const manualRetries = new Set<string>();
   const loadedCache = new Set<string>();
   const cacheLoads = new Map<string, Promise<void>>();
@@ -511,8 +495,6 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
   let acknowledged: string[] = [];
   let membershipGeneration = 0;
   let membershipNeedsRetry = false;
-  let membershipRetryDelayMs: number | undefined;
-  let cancelMembershipRetry: (() => void) | null = null;
   // Chats this session has opened. Visibility is the only way in, so an agent is never
   // subscribed — and never resumed on the daemon — until the user actually opens it.
   // Membership then survives hiding, workspace switches, backgrounding and reconnect;
@@ -563,7 +545,6 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
 
   const cancelCatchUp = (agentId: string) => {
     catchUpGenerations.set(agentId, (catchUpGenerations.get(agentId) ?? 0) + 1);
-    catchUps.get(agentId)?.cancelRetry?.();
     catchUps.delete(agentId);
     pendingCatchUps.delete(agentId);
   };
@@ -607,19 +588,7 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
       setVisibilityCatchUpReady(agentId);
     } catch (error) {
       if (catchUps.get(agentId)?.generation === generation) {
-        const nextRetryDelayMs = getNextRetryDelayMs(catchUps.get(agentId)?.retryDelayMs);
-        const cancelRetry = ports.schedule(() => {
-          const current = catchUps.get(agentId);
-          if (current?.generation !== generation || current.status !== "error") return;
-          startCatchUp(agentId);
-        }, nextRetryDelayMs);
-        catchUps.set(agentId, {
-          generation,
-          status: "error",
-          request,
-          cancelRetry,
-          retryDelayMs: nextRetryDelayMs,
-        });
+        catchUps.set(agentId, { generation, status: "error", request });
         setVisibilityCatchUpError([agentId], error);
         ports.reportError(error);
       }
@@ -686,17 +655,9 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
     if (decision === "keep") {
       return;
     }
-    current?.cancelRetry?.();
     const generation = (catchUpGenerations.get(agentId) ?? 0) + 1;
     catchUpGenerations.set(agentId, generation);
-    const retryDelayMs =
-      supersede || current?.status !== "error" ? undefined : current.retryDelayMs;
-    catchUps.set(agentId, {
-      generation,
-      status: "running",
-      request: nextRequest,
-      retryDelayMs,
-    });
+    catchUps.set(agentId, { generation, status: "running", request: nextRequest });
     pendingCatchUps.delete(agentId);
     void fetchUntilCurrent(
       agentId,
@@ -715,7 +676,7 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
     for (const agentId of ordered) {
       const pendingCatchUp = pendingCatchUps.get(agentId);
       const current = catchUps.get(agentId);
-      // Completion of a different chat must not bypass a failed chat's backoff or
+      // Completion of a different chat must not bypass a failed chat's error or
       // supersede an in-flight tail that owns its parked gap recovery.
       if (current && !(current.status === "complete" && pendingCatchUp)) continue;
       startCatchUp(agentId, {
@@ -741,42 +702,31 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
     } catch (error) {
       if (disposed || !connected || generation !== membershipGeneration) return;
       membershipNeedsRetry = true;
+      acknowledged = [];
+      for (const agentId of requested) cancelCatchUp(agentId);
       setVisibilityCatchUpError(requested, error);
-      cancelMembershipRetry?.();
-      const nextRetryDelayMs = getNextRetryDelayMs(membershipRetryDelayMs);
-      cancelMembershipRetry = ports.schedule(() => {
-        cancelMembershipRetry = null;
-        if (disposed || !connected || membershipGeneration !== generation) return;
-        void reconcileMembership();
-      }, nextRetryDelayMs);
-      membershipRetryDelayMs = nextRetryDelayMs;
       ports.reportError(error);
       return;
     }
     if (disposed || !connected || generation !== membershipGeneration) return;
-    cancelMembershipRetry?.();
-    cancelMembershipRetry = null;
-    membershipRetryDelayMs = undefined;
     acknowledged = requested;
     startAcknowledgedCatchUps();
   };
 
   const retryVisibleAgentTimeline = (agentId: string) => {
-    if (!isDesired(agentId) || manualRetries.has(agentId)) return;
+    if (!connected || !isDesired(agentId) || manualRetries.has(agentId)) return;
     const catchUp = catchUps.get(agentId);
     const membershipRetryable = membershipNeedsRetry && connected;
     if (catchUp?.status !== "error" && !membershipRetryable) return;
     manualRetries.add(agentId);
     notifyListeners();
-    if (catchUp?.status === "error") {
-      catchUp.cancelRetry?.();
-      startCatchUp(agentId, { request: catchUp.request, supersede: true });
+    if (membershipRetryable) {
+      void reconcileMembership();
       return;
     }
-    cancelMembershipRetry?.();
-    cancelMembershipRetry = null;
-    membershipRetryDelayMs = undefined;
-    void reconcileMembership();
+    if (catchUp?.status === "error") {
+      startCatchUp(agentId, { request: catchUp.request, supersede: true });
+    }
   };
 
   const commitDesiredMembership = (
@@ -817,8 +767,6 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
         ensureCacheLoaded(agentId);
       }
     }
-    cancelMembershipRetry?.();
-    cancelMembershipRetry = null;
     desired = nextDesired;
     ports.replaceDemandedAgentIds(desired);
     membershipGeneration += 1;
@@ -869,6 +817,7 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
         return;
       }
       for (const agentId of visibleAgentIds()) {
+        if (visibilityCatchUpErrors.has(agentId)) continue;
         visibilityCatchUpPending.add(agentId);
         visibilityCatchUpErrors.delete(agentId);
         startCatchUp(agentId, { supersede: true });
@@ -880,9 +829,6 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
       connected = nextConnected;
       if (!connected) {
         commitDesiredMembership(desired, { resetCatchUpStatus: true });
-        cancelMembershipRetry?.();
-        cancelMembershipRetry = null;
-        membershipRetryDelayMs = undefined;
         void observation?.handle.release().catch(ports.reportError);
         observation = null;
         acknowledged = [];
@@ -894,7 +840,7 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
       void reconcileMembership();
     },
     recoverGap(agentId, cursor) {
-      if (!isDesired(agentId)) return;
+      if (!isDesired(agentId) || catchUps.get(agentId)?.status === "error") return;
       startCatchUp(agentId, {
         request: planTimelineCatchUpAfter({ epoch: cursor.epoch, seq: cursor.endSeq }),
         supersede: true,
@@ -904,10 +850,7 @@ export function createViewedTimelineSync(ports: ViewedTimelineSyncPorts): Viewed
       disposed = true;
       void observation?.handle.release().catch(ports.reportError);
       observation = null;
-      cancelMembershipRetry?.();
-      cancelMembershipRetry = null;
       membershipNeedsRetry = false;
-      membershipRetryDelayMs = undefined;
       sources.clear();
       membershipGeneration += 1;
       for (const agentId of desired) cancelCatchUp(agentId);