diff --git a/crates/chatd/src/adapters/tiny/e2ee.rs b/crates/chatd/src/adapters/tiny/e2ee.rs
index fb81eeca..efdfe269 100644
--- a/crates/chatd/src/adapters/tiny/e2ee.rs
+++ b/crates/chatd/src/adapters/tiny/e2ee.rs
@@ -989,6 +989,8 @@ pub(crate) struct Device {
     /// Conversations where someone the server is adding has no device in the group yet, as the
     /// last sync saw them.
     joining: BTreeSet<String>,
+    /// Coalesce reconnect and device-link membership catch-up workers.
+    membership_nudge: bool,
     /// Conversations with a row that did not open because this device was not in the group, or
     /// not at its epoch, yet. Read again when the group's handshake next comes.
     unopened: BTreeSet<String>,
@@ -1141,6 +1143,14 @@ pub(crate) fn begin(saved: &Saved, commands: std::sync::mpsc::Sender<crate::adap
     device.enrol();
 }
 
+/// Live hints can be missed while offline (or before enrollment finished). Once the
+/// conversation list is known, revisit its encrypted groups without waiting for a send.
+pub(crate) fn resume(saved: &Saved) {
+    if let Some(shared) = started(saved) {
+        linking::nudge(shared);
+    }
+}
+
 /// Signing out: the device is revoked on the server and its state forgotten here. What it opened
 /// stays, sealed under the cache key [`keep_history`] keeps in the keychain, unless the person
 /// chose to erase it ([`erase_history`]); a later sign-in here is a new device that reads it.
@@ -1417,6 +1427,7 @@ impl Device {
             files: HashMap::new(),
             checked: BTreeSet::new(),
             joining: BTreeSet::new(),
+            membership_nudge: false,
             unopened: BTreeSet::new(),
             linking: false,
             commands: None,
diff --git a/crates/chatd/src/adapters/tiny/e2ee/fake.rs b/crates/chatd/src/adapters/tiny/e2ee/fake.rs
index 88697c81..fd1c4b46 100644
--- a/crates/chatd/src/adapters/tiny/e2ee/fake.rs
+++ b/crates/chatd/src/adapters/tiny/e2ee/fake.rs
@@ -126,6 +126,11 @@ impl Server {
         let body: Value = serde_json::from_str(body).unwrap_or(Value::Null);
         let conversation = format!("/v1/e2ee/conversations/{CONVERSATION}");
         match method {
+            "GET" if matches!(path, "/v1/channels" | "/v1/rooms") => (200, json!([])),
+            "GET" if path == "/v1/directs" => (200, json!([{
+                "id": CONVERSATION, "e2ee": true,
+                "peer": {"user_id": if caller == ADA { BEA } else { ADA }, "display_name": "Peer"}
+            }])),
             "GET" if path == "/v1/capabilities" => (
                 200,
                 json!({ "features": ["e2ee_text"], "e2ee_text": "optional" }),
@@ -422,6 +427,55 @@ fn the_first_message_of_a_new_conversation_opens_for_the_other_person() {
     prove_first_encrypted_message().expect("first message opens for the other person");
 }
 
+#[test]
+fn publishing_conversations_settles_an_add_missed_while_offline() {
+    let fake = Fake::start(0);
+    let mut creator = device(&fake, ADA, "offline-creator");
+    creator.sync(CONVERSATION).unwrap();
+    assert_eq!(lock(&fake.server).epoch, Some(0));
+    // The creator is now offline. The new recipient can trigger a signed Add,
+    // but cannot commit it or join until the creator processes it.
+    let mut recipient = device(&fake, BEA, "waiting-recipient");
+    recipient.sync(CONVERSATION).unwrap();
+    assert!(!recipient.engine().unwrap().has_group(ConversationId::parse(CONVERSATION).unwrap()));
+    assert!(lock(&fake.server).pending.iter().any(|add| add.issued));
+
+    let creator_saved = creator.saved.clone();
+    let recipient_saved = recipient.saved.clone();
+    let creator = Arc::new(Mutex::new(creator));
+    let recipient = Arc::new(Mutex::new(recipient));
+    let registry_key = |saved: &Saved| (saved.base_url.clone(), saved.user_id.clone());
+    lock(devices()).insert(registry_key(&creator_saved), creator.clone());
+    lock(devices()).insert(registry_key(&recipient_saved), recipient.clone());
+    let context = |saved: &Saved| {
+        let hub = Arc::new(crate::hub::Hub::new(crate::store::Store::open(Path::new(":memory:")).unwrap()));
+        crate::adapters::local(protocol::AccountKey { network: protocol::Network::Tiny, account: saved.user_id.clone() }, hub)
+    };
+    let (_adapter, creator_ctx, _inbox) = context(&creator_saved);
+    // The normal list refresh after connection must drive membership even with
+    // no live handshake event and no messages sent or opened by the creator.
+    super::super::messages::publish(&creator_ctx, &creator_saved, &HashMap::new());
+    let deadline = Instant::now() + Duration::from_secs(5);
+    while lock(&fake.server).epoch == Some(0) && Instant::now() < deadline {
+        std::thread::sleep(Duration::from_millis(10));
+    }
+    let committed = lock(&fake.server).epoch == Some(1);
+    if committed {
+        let (_adapter, ctx, _inbox) = context(&recipient_saved);
+        super::super::messages::publish(&ctx, &recipient_saved, &HashMap::new());
+        while !lock(&recipient).engine().unwrap().has_group(ConversationId::parse(CONVERSATION).unwrap()) && Instant::now() < deadline {
+            std::thread::sleep(Duration::from_millis(10));
+        }
+    }
+    let joined = lock(&recipient).engine().unwrap().has_group(ConversationId::parse(CONVERSATION).unwrap());
+    lock(devices()).remove(&registry_key(&creator_saved));
+    lock(devices()).remove(&registry_key(&recipient_saved));
+    assert!(committed, "returning creator must commit the missed Add without opening or sending a message");
+    assert!(joined, "returning recipient must take its Welcome without a live hint");
+    let answer = lock(&recipient).post(CONVERSATION, "after-resume", Kind::Message, None, &Envelope::message("joined after reconnect"), &[]).unwrap();
+    assert_eq!(text(&fake.read(&mut lock(&creator)), &id_of(&answer)), "joined after reconnect");
+}
+
 /// A row that did not open because this device was not in the group yet (here, the sync of the
 /// read failed) is read again when the group arrives, not only when another message does.
 #[test]
diff --git a/crates/chatd/src/adapters/tiny/e2ee/linking.rs b/crates/chatd/src/adapters/tiny/e2ee/linking.rs
index 9f021a4e..4e7a3368 100644
--- a/crates/chatd/src/adapters/tiny/e2ee/linking.rs
+++ b/crates/chatd/src/adapters/tiny/e2ee/linking.rs
@@ -286,9 +286,24 @@ const NUDGES: [Duration; 3] = [
 /// Settle a device addition or removal immediately, so a linked device gets its Welcome before
 /// its first send. Older servers batch membership changes, so retain the later retries too.
 pub(super) fn nudge(device: Shared) {
+    {
+        let mut state = lock(&device);
+        if !state.ready() || state.membership_nudge {
+            return;
+        }
+        state.membership_nudge = true;
+    }
+    struct Running(Shared);
+    impl Drop for Running {
+        fn drop(&mut self) {
+            lock(&self.0).membership_nudge = false;
+        }
+    }
+    let running = Running(device.clone());
     let _ = std::thread::Builder::new()
         .name("tiny-e2ee-nudge".into())
         .spawn(move || {
+            let _running = running;
             for wait in NUDGES {
                 std::thread::sleep(wait);
                 let conversations = lock(&device).known_encrypted();
@@ -298,8 +313,9 @@ pub(super) fn nudge(device: Shared) {
                     if !device.ready() {
                         return;
                     }
-                    if let Err(why) = device.sync(&conversation) {
-                        tracing::debug!("chatd: Tiny e2ee: {conversation}: {why}");
+                    match device.sync(&conversation) {
+                        Ok(()) => device.reload(&BTreeSet::from([conversation])),
+                        Err(why) => tracing::warn!("chatd: Tiny e2ee: {conversation}: {why}"),
                     }
                 }
             }
diff --git a/crates/chatd/src/adapters/tiny/live.rs b/crates/chatd/src/adapters/tiny/live.rs
index 0a2f60a2..71248b76 100644
--- a/crates/chatd/src/adapters/tiny/live.rs
+++ b/crates/chatd/src/adapters/tiny/live.rs
@@ -1809,6 +1809,8 @@ fn ready_heard(ctx: &Context, saved: &Saved, live: &Live, seat: &mut Seat, heard
     // without another members read: a resume is not fresh and reads nothing here.
     if fresh {
         super::messages::publish(ctx, saved, &seat.names);
+    } else {
+        super::e2ee::resume(saved);
     }
 }
 
diff --git a/crates/chatd/src/adapters/tiny/messages.rs b/crates/chatd/src/adapters/tiny/messages.rs
index 83ae665f..eac16c0b 100644
--- a/crates/chatd/src/adapters/tiny/messages.rs
+++ b/crates/chatd/src/adapters/tiny/messages.rs
@@ -212,6 +212,9 @@ pub(crate) fn publish(ctx: &Context, saved: &Saved, names: &HashMap<String, Stri
             record_call(ctx, saved, names, &direct.id, call);
         }
     }
+    // Includes groups with no messages yet: opening only cached history does not
+    // drive pending Adds, and their live hint may have arrived while we were offline.
+    e2ee::resume(saved);
 }
 
 /// The server lists direct calls. A server not yet checked is asked; a 404 is an older one.
diff --git a/docs/e2ee-text-client.md b/docs/e2ee-text-client.md
index 606f2daf..a60a5a5e 100644
--- a/docs/e2ee-text-client.md
+++ b/docs/e2ee-text-client.md
@@ -37,7 +37,12 @@ that account here (a new device) reads them. "Erase" deletes the directory and t
   starts the group if nobody has (`POST .../group`), applies the handshake feed in order (an Add
   for a device not yet verified makes it fetch that person's list and try once more), and commits
   what is pending, with the Welcome for every device it adds. It runs on `e2ee.handshake`, before
-  every send, and when a message is from an epoch ahead (once per read).
+  every send, and when a message is from an epoch ahead (once per read). After publishing the
+  conversation list at startup, and after socket resume, a background worker also revisits known
+  encrypted groups. This catches pending additions whose live hint was missed while offline,
+  including conversations with no messages. It retries after 35 and then 40 seconds for older
+  servers that batch membership changes, and requests a message backfill after successful sync.
+  Concurrent catch-up requests share one worker per device.
 - **Sending.** Never plaintext to an encrypted conversation: if this device is not ready, the send
   is an error. While the server lists an Add for someone with no device in the group yet, the send
   waits, syncing every 2 s, up to 45 s: the server signs Adds only after its commit batch
diff --git a/e2e/scenarios/encrypted-membership-reconnect/README.md b/e2e/scenarios/encrypted-membership-reconnect/README.md
new file mode 100644
index 00000000..70af6b50
--- /dev/null
+++ b/e2e/scenarios/encrypted-membership-reconnect/README.md
@@ -0,0 +1,36 @@
+# Encrypted membership after reconnect
+
+Required recorded real-app scenario, using isolated profiles and invented accounts:
+
+1. Sign in Ada and create an encrypted direct conversation with Bea before Bea
+   has registered an encryption device. Ada's device is the sole group member.
+2. Close Ada's app. Sign in Bea and open the direct conversation. Verify that
+   Bea cannot send yet. The server should have a signed, uncommitted Add for Bea.
+3. Start Ada's release-profile app again. Leave it at the conversation list;
+   do not open the conversation, send a message, or press Update encryption.
+4. Restart Bea's app, then send a unique message from Bea. It must send and
+   Ada must be able to read it. Reply from Ada and verify Bea reads it too.
+5. Repeat with Ada's process left running but its network disconnected while
+   Bea's Add is created. Restore the network and repeat step 4. This checks
+   socket resume as well as cold startup. Allow 90 seconds for bounded retries
+   on older servers that batch membership changes.
+
+Run the same scenario on the base and branch on every platform. The base must
+fail at step 4 because connection/list refresh does not process the pending Add.
+The lab must record and publish these runs; no full GUI proof is claimed here.
+Platform UI automation and lab recordings are still pending, so this scenario
+does not currently certify the change and is not in the release suite.
+
+The focused source regression uses real MLS devices and the actual conversation
+list publication path with an in-process server. It checks that the creator
+commits the missed Add, that the recipient takes its Welcome without a live
+hint, and that the recipient's first message opens on the creator:
+
+```sh
+cargo test --locked -p chatd --lib publishing_conversations_settles_an_add_missed_while_offline -- --nocapture
+```
+
+For a behavioral base comparison, transplant the test and its fake server list
+routes, leaving production code unchanged. A missing test is not a base failure.
+This source regression supplements the required real-app scenario; it is not
+`e2e-proof` or a substitute for recordings.
diff --git a/e2e/scenarios/encrypted-membership-reconnect/scenario.json b/e2e/scenarios/encrypted-membership-reconnect/scenario.json
new file mode 100644
index 00000000..c8de2cae
--- /dev/null
+++ b/e2e/scenarios/encrypted-membership-reconnect/scenario.json
@@ -0,0 +1,7 @@
+{
+  "id": "encrypted-membership-reconnect",
+  "title": "Returning clients finish a pending encrypted-chat admission after missing its live event",
+  "platforms": ["windows", "macos", "linux", "android", "ios"],
+  "needs": ["server", "two-people"],
+  "release": false
+}