diff --git a/tests/common/sync_helpers.rs b/tests/common/sync_helpers.rs
index 668b5fc..4084a05 100644
--- a/tests/common/sync_helpers.rs
+++ b/tests/common/sync_helpers.rs
@@ -135,52 +135,58 @@ impl TestClient {
         Err("Connection loop exited unexpectedly".to_string())
     }
 
-    /// Send an event with retry logic.
+    /// Send an event with bounded retry for transient transport failures.
     ///
-    /// Attempts to send up to 3 times with exponential backoff:
-    /// - Attempt 1: immediate
-    /// - Attempt 2: after 200ms
-    /// - Attempt 3: after 400ms
+    /// Attempts to send up to 3 times with short backoff (200ms, then 400ms),
+    /// reconnecting between attempts. Only errors that leave delivery in doubt
+    /// (disconnection, transport failure, OK timeout) are retried; a relay
+    /// `OK false` rejection is deterministic and fails immediately so tests
+    /// cannot mask write-policy regressions.
     ///
     /// # Arguments
     /// * `event` - The signed event to send
     ///
     /// # Returns
     /// * `Ok(EventId)` on successful send
-    /// * `Err(String)` if all attempts fail
+    /// * `Err(String)` on relay rejection or after all attempts fail
     pub async fn send_event(&self, event: &Event) -> Result<EventId, String> {
-        let delays = [0, 200, 400]; // Exponential backoff in ms
+        let delays = [0, 200, 400]; // Backoff in ms
+        let mut last_error = String::new();
 
         for (attempt, delay_ms) in delays.iter().enumerate() {
             if *delay_ms > 0 {
                 tokio::time::sleep(Duration::from_millis(*delay_ms)).await;
             }
 
-            match self.client.send_event(event).await {
-                Ok(output) => {
-                    if !output.success.is_empty() {
-                        return Ok(output.value);
-                    }
-                    // Log failures for debugging
-                    if !output.failed.is_empty() {
-                        eprintln!(
-                            "  Send attempt {} - failures: {:?}",
-                            attempt + 1,
-                            output.failed
-                        );
-                        // Try reconnecting if relay disconnected
-                        self.client.connect().await;
-                    }
+            // Send through the single tracked relay rather than the pool so
+            // the SDK error kind survives and OK-false can be told apart from
+            // transport failures.
+            let relay = self
+                .client
+                .relay(self.relay_url.as_str())
+                .await
+                .map_err(|e| format!("Failed to look up relay {}: {}", self.relay_url, e))?
+                .ok_or_else(|| format!("Relay {} missing from client pool", self.relay_url))?;
+
+            match relay.send_event(event).await {
+                Ok(output) => return Ok(*output.id()),
+                Err(e) if e.kind() == ErrorKind::Rejected => {
+                    return Err(format!("Relay rejected event {}: {}", event.id, e));
                 }
                 Err(e) => {
-                    eprintln!("  Send attempt {} - error: {}", attempt + 1, e);
+                    eprintln!("  Send attempt {} - transient error: {}", attempt + 1, e);
+                    last_error = e.to_string();
+                    // Re-establish the connection before the next attempt.
+                    let _ = self.connect().await;
                 }
             }
         }
 
         Err(format!(
-            "Failed to send event {} after 3 attempts",
-            event.id
+            "Failed to send event {} after {} attempts: {}",
+            event.id,
+            delays.len(),
+            last_error
         ))
     }
 
diff --git a/tests/sync/discovery.rs b/tests/sync/discovery.rs
index bfcc2cc..84f549a 100644
--- a/tests/sync/discovery.rs
+++ b/tests/sync/discovery.rs
@@ -106,14 +106,13 @@ async fn test_discovers_layer3_via_layer2() {
         setup_announcement_on_relay(&relay_b, &keys, &domain_refs, repo_id).await;
     println!("Announcement set up on relay_b (should trigger discovery of relay_a)");
 
-    // 9. Wait for relay_b to discover relay_a and sync the patch
-    println!("Waiting 3s for relay_b to discover relay_a and sync patch...");
-    tokio::time::sleep(Duration::from_secs(3)).await;
-
-    // 10. Verify patch was synced to relay_b
+    // 9/10. Verify the patch syncs to relay_b. The bounded poll on the synced
+    // event replaces a fixed discovery sleep, which is not a reliable proxy
+    // under CI load.
     let filter = Filter::new().kind(Kind::GitPatch).author(keys.public_key());
 
-    let patch_synced = wait_for_event_on_relay(relay_b.url(), filter, Duration::from_secs(5)).await;
+    let patch_synced =
+        wait_for_event_on_relay(relay_b.url(), filter, Duration::from_secs(30)).await;
 
     if patch_synced {
         println!(
@@ -212,14 +211,12 @@ async fn test_relay_discovery_via_announcements_with_historic_sync() {
         setup_announcement_on_relay(&relay_b, &keys, &domain_refs, repo_id).await;
     println!("Announcement set up on relay_b (should trigger discovery of relay_a)");
 
-    // 7. Wait for sync
-    println!("Waiting 3s for Layer 2 sync...");
-    tokio::time::sleep(Duration::from_secs(3)).await;
-
-    // 8. Verify Layer 2 event synced to relay_b
+    // 7/8. Verify the Layer 2 event syncs to relay_b. The bounded poll on the
+    // synced event replaces a fixed discovery sleep, which is not a reliable
+    // proxy under CI load.
     let issue_filter = Filter::new().kind(Kind::GitIssue).author(keys.public_key());
     let issue_synced =
-        wait_for_event_on_relay(relay_b.url(), issue_filter, Duration::from_secs(5)).await;
+        wait_for_event_on_relay(relay_b.url(), issue_filter, Duration::from_secs(30)).await;
 
     println!("Sync result:");
     println!("  Issue {} synced: {}", issue_id, issue_synced);
diff --git a/tests/sync/historic_sync.rs b/tests/sync/historic_sync.rs
index a4ca7df..8db7a43 100644
--- a/tests/sync/historic_sync.rs
+++ b/tests/sync/historic_sync.rs
@@ -33,7 +33,7 @@ async fn test_bootstrap_syncs_existing_layer2_events() {
         .author(result.maintainer_keys.public_key());
 
     let synced =
-        wait_for_event_on_relay(result.syncing_relay.url(), filter, Duration::from_secs(5)).await;
+        wait_for_event_on_relay(result.syncing_relay.url(), filter, Duration::from_secs(30)).await;
 
     // Cleanup
     result.syncing_relay.stop().await;
@@ -70,15 +70,15 @@ async fn test_relay_replays_events_after_restart() {
     let synced_first = wait_for_event_on_relay(
         result.syncing_relay.url(),
         filter.clone(),
-        Duration::from_secs(5),
+        Duration::from_secs(30),
     )
     .await;
 
     println!("First sync check: {}", synced_first);
 
-    // Stop syncing relay (simulates restart)
+    // Stop syncing relay (simulates restart). `stop` waits on the process, so
+    // the replacement instance can start immediately.
     result.syncing_relay.stop().await;
-    tokio::time::sleep(Duration::from_millis(500)).await;
 
     // Restart syncing relay (new instance with same bootstrap config)
     // Note: The new syncing relay will have a different domain, so it may not
@@ -90,10 +90,10 @@ async fn test_relay_replays_events_after_restart() {
         syncing_new.domain()
     );
 
-    // Wait for re-sync
-    tokio::time::sleep(Duration::from_secs(2)).await;
-
-    // Verify announcement is available on restarted syncing relay
+    // Check whether the announcement re-syncs to the restarted relay. The
+    // bounded poll is kept short because this outcome is informational only:
+    // the new instance has a different domain, so the announcement may
+    // legitimately never be accepted (see the note below).
     let synced_after_restart =
         wait_for_event_on_relay(syncing_new.url(), filter, Duration::from_secs(5)).await;
 
@@ -135,7 +135,7 @@ async fn test_announcement_not_listing_relay_is_not_synced() {
     let keys = Keys::generate();
 
     // Wait for sync connection to establish
-    match wait_for_sync_connection(syncing.url(), 1, Duration::from_secs(5)).await {
+    match wait_for_sync_connection(syncing.url(), 1, Duration::from_secs(30)).await {
         Ok(()) => println!("Sync connection established (verified via metrics)"),
         Err(e) => println!("Sync connection check: {} (continuing with test)", e),
     }
@@ -168,15 +168,14 @@ async fn test_announcement_not_listing_relay_is_not_synced() {
 
     client.disconnect().await;
 
-    // Wait for potential sync attempt
-    tokio::time::sleep(Duration::from_secs(3)).await;
-
-    // Verify announcement did NOT sync to syncing relay
+    // Verify announcement did NOT sync to syncing relay. The bounded wait is
+    // the observation window for a wrongful sync; the sync connection was
+    // already confirmed above, so any policy failure has this long to appear.
     let filter = Filter::new()
         .kind(Kind::GitRepoAnnouncement)
         .author(keys.public_key());
 
-    let synced = wait_for_event_on_relay(syncing.url(), filter, Duration::from_secs(2)).await;
+    let synced = wait_for_event_on_relay(syncing.url(), filter, Duration::from_secs(5)).await;
 
     // Cleanup
     syncing.stop().await;
@@ -246,8 +245,17 @@ async fn test_history_sync_without_negentropy() {
         announcement_id
     );
 
-    // Wait to ensure event is stored
-    tokio::time::sleep(Duration::from_millis(500)).await;
+    // Confirm the announcement is stored and served by the source before the
+    // syncing relay ever connects, so this genuinely exercises history sync.
+    assert!(
+        wait_for_event_on_relay(
+            source.url(),
+            Filter::new().id(announcement.id),
+            Duration::from_secs(30),
+        )
+        .await,
+        "announcement should be served by the source before the syncing relay starts"
+    );
 
     // NOW start syncing relay on the reserved port, with negentropy DISABLED
     // This syncing relay has never connected before - it needs to do HISTORY sync
@@ -263,15 +271,13 @@ async fn test_history_sync_without_negentropy() {
         syncing.domain()
     );
 
-    // Wait for history sync to complete (using REQ+EOSE, not negentropy)
-    tokio::time::sleep(Duration::from_secs(3)).await;
-
-    // Verify announcement synced to syncing relay via HISTORY sync
+    // Verify announcement syncs to the syncing relay via HISTORY sync. The
+    // bounded poll replaces a fixed sleep plus short deadline.
     let filter = Filter::new()
         .kind(Kind::GitRepoAnnouncement)
         .author(keys.public_key());
 
-    let synced = wait_for_event_on_relay(syncing.url(), filter, Duration::from_secs(5)).await;
+    let synced = wait_for_event_on_relay(syncing.url(), filter, Duration::from_secs(30)).await;
 
     // Cleanup
     syncing.stop().await;
@@ -368,8 +374,17 @@ async fn test_pagination_for_large_historic_sync() {
         .expect("Failed to send announcement to source");
     println!("Announcement sent to source");
 
-    // Wait for announcement to be stored
-    tokio::time::sleep(Duration::from_millis(200)).await;
+    // Confirm the announcement is stored before sending the issues that
+    // reference it.
+    assert!(
+        wait_for_event_on_relay(
+            source.url(),
+            Filter::new().id(announcement.id),
+            Duration::from_secs(30),
+        )
+        .await,
+        "announcement should be stored on the source relay"
+    );
 
     // Send all 40 issue events to source (before syncing relay starts)
     println!("Sending {} issues to source relay...", issue_events.len());
@@ -391,8 +406,18 @@ async fn test_pagination_for_large_historic_sync() {
 
     client.disconnect().await;
 
-    // Wait to ensure all events are stored
-    tokio::time::sleep(Duration::from_millis(500)).await;
+    // Confirm the last-sent issue is stored so every event exists before the
+    // syncing relay connects.
+    let last_issue = issue_events.last().expect("at least one issue");
+    assert!(
+        wait_for_event_on_relay(
+            source.url(),
+            Filter::new().id(last_issue.id),
+            Duration::from_secs(30),
+        )
+        .await,
+        "all issues should be stored on the source relay"
+    );
 
     // NOW start syncing relay on the reserved port, with negentropy DISABLED
     // This forces it to use REQ+EOSE historic sync with pagination
@@ -408,17 +433,14 @@ async fn test_pagination_for_large_historic_sync() {
         syncing.domain()
     );
 
-    // Wait for historic sync with pagination to complete
-    println!("Waiting for historic sync with pagination to complete...");
-    tokio::time::sleep(Duration::from_secs(8)).await;
-
-    // Verify announcement synced
+    // Verify announcement synced. The bounded poll replaces a fixed
+    // "wait for pagination" sleep plus short deadline.
     let announcement_filter = Filter::new()
         .kind(Kind::GitRepoAnnouncement)
         .author(keys.public_key());
 
     let announcement_synced =
-        wait_for_event_on_relay(syncing.url(), announcement_filter, Duration::from_secs(3)).await;
+        wait_for_event_on_relay(syncing.url(), announcement_filter, Duration::from_secs(30)).await;
 
     // Verify ALL 40 issues synced
     let issues_filter = Filter::new().kind(Kind::GitIssue).author(keys.public_key());
@@ -432,18 +454,21 @@ async fn test_pagination_for_large_historic_sync() {
         .add_relay(syncing.url())
         .await
         .expect("Failed to add syncing relay to client");
-    client.connect().await;
+    client.connect().and_wait(Duration::from_secs(30)).await;
 
-    // Wait for connection
-    tokio::time::sleep(Duration::from_millis(500)).await;
-
-    let synced_issues = client
-        .fetch_events(issues_filter)
-        .timeout(Duration::from_secs(5))
-        .await
-        .expect("Failed to fetch issues from syncing relay");
-
-    let synced_count = synced_issues.len();
+    // Poll until every issue has paginated across, up to a bounded deadline.
+    let deadline = tokio::time::Instant::now() + Duration::from_secs(30);
+    let synced_count = loop {
+        let synced_issues = client
+            .fetch_events(issues_filter.clone())
+            .timeout(Duration::from_secs(5))
+            .await
+            .expect("Failed to fetch issues from syncing relay");
+        if synced_issues.len() >= 40 || tokio::time::Instant::now() >= deadline {
+            break synced_issues.len();
+        }
+        tokio::time::sleep(Duration::from_millis(200)).await;
+    };
     println!("Synced {} out of 40 expected issues", synced_count);
 
     client.disconnect().await;
diff --git a/tests/sync/live_sync.rs b/tests/sync/live_sync.rs
index dc1582d..36dd2a3 100644
--- a/tests/sync/live_sync.rs
+++ b/tests/sync/live_sync.rs
@@ -486,8 +486,7 @@ async fn test_live_sync_layer3_events() {
             .authenticator(SignerAuthenticator::new(temp_keys))
             .build();
         if client.add_relay(relay_b.url()).await.is_ok() {
-            client.connect().await;
-            tokio::time::sleep(Duration::from_millis(500)).await;
+            client.connect().and_wait(Duration::from_secs(30)).await;
 
             let fetch_filter = Filter::new().kind(Kind::Comment).id(comment_id);
 
@@ -644,8 +643,7 @@ async fn test_live_sync_event_ordering() {
 
     let events_found: Vec<Event>;
     if client.add_relay(relay_b.url()).await.is_ok() {
-        client.connect().await;
-        tokio::time::sleep(Duration::from_millis(500)).await;
+        client.connect().and_wait(Duration::from_secs(30)).await;
 
         let filter = Filter::new().kind(Kind::GitIssue).author(keys.public_key());
 
diff --git a/tests/sync/maintainer_reprocessing.rs b/tests/sync/maintainer_reprocessing.rs
index 75438b3..4abbc05 100644
--- a/tests/sync/maintainer_reprocessing.rs
+++ b/tests/sync/maintainer_reprocessing.rs
@@ -508,6 +508,8 @@ async fn test_invitee_only_acceptance_recovers_cold_owner_invitation() {
         .await,
         "Invitee-only server should reject and index the owner invitation"
     );
+    // Passage of the configured one-second hot-cache TTL is required so the
+    // acceptance below exercises cold exact-ID recovery, not the hot copy.
     tokio::time::sleep(Duration::from_secs(2)).await;
 
     let pushes_before_acceptance = [
@@ -767,6 +769,8 @@ async fn test_invitation_applies_newer_invitee_state_to_owner_before_acceptance(
         .await;
     }
 
+    // Deliberate timestamp spacing: the invitee events below are dated one
+    // second after the owner state, so let wall-clock time pass that mark.
     tokio::time::sleep(Duration::from_secs(1)).await;
 
     let invitee_git = tempfile::tempdir().expect("Failed to create invitee repository directory");
@@ -982,6 +986,8 @@ async fn test_acceptance_replaces_existing_invitee_repository_with_newer_owner_s
         .await;
     }
 
+    // Deliberate timestamp spacing: the owner events below are dated one
+    // second after the invitee state, so let wall-clock time pass that mark.
     tokio::time::sleep(Duration::from_secs(1)).await;
 
     let owner_git = tempfile::tempdir().expect("Failed to create owner repository directory");
@@ -1062,6 +1068,8 @@ async fn test_acceptance_replaces_existing_invitee_repository_with_newer_owner_s
         .await,
         "Invitee-only server should process the unilateral invitation before acceptance"
     );
+    // Deliberate observation window: give a wrongful state replacement time
+    // to manifest before asserting the invitee state is still intact.
     tokio::time::sleep(Duration::from_millis(500)).await;
     for relay in invitee_servers {
         assert_exact_event_served(relay, &invitee_state, "Invitee state before acceptance").await;
@@ -1395,9 +1403,22 @@ async fn test_maintainer_announcement_reprocessed_immediately() {
     let relay_b = TestRelay::start_with_sync(Some(relay_a.url().to_string())).await;
     println!("relay_b started at {}", relay_b.url());
 
-    // Give relay_b's SyncManager time to complete the initial negentropy sync with relay_a.
-    tokio::time::sleep(Duration::from_secs(3)).await;
-    println!("✓ relay_b synced from relay_a (maintainer announcement should be in hot cache)");
+    // Wait for relay_b's initial negentropy sync on its observable outcome:
+    // the rejected maintainer announcement is logged by its note ID.
+    let maintainer_note = maintainer_announcement
+        .id
+        .to_bech32()
+        .expect("Failed to encode maintainer announcement ID");
+    assert!(
+        wait_for_log(
+            &relay_b.log_path(),
+            &maintainer_note,
+            Duration::from_secs(30),
+        )
+        .await,
+        "relay_b should sync, reject and index the maintainer announcement"
+    );
+    println!("✓ relay_b synced from relay_a (maintainer announcement in hot cache)");
 
     let start = std::time::Instant::now();
 
@@ -1441,19 +1462,15 @@ async fn test_maintainer_announcement_reprocessed_immediately() {
         "✓ Owner git data pushed to relay_b (owner announcement promoted, hot cache re-processed)"
     );
 
-    // Step 5: Wait briefly for async processing to complete.
-    tokio::time::sleep(Duration::from_secs(1)).await;
-
-    let elapsed = start.elapsed();
-
-    // Step 6: Verify both announcements are in relay_b's database.
+    // Step 5/6: Verify both announcements are in relay_b's database. Bounded
+    // polls on the served events replace a fixed post-push sleep.
     let owner_filter = Filter::new()
         .kind(Kind::GitRepoAnnouncement)
         .author(owner_keys.public_key())
         .identifier(identifier);
 
     let owner_found =
-        wait_for_event_on_relay(relay_b.url(), owner_filter, Duration::from_secs(2)).await;
+        wait_for_event_on_relay(relay_b.url(), owner_filter, Duration::from_secs(30)).await;
     assert!(owner_found, "Owner announcement should be in relay_b");
 
     let maintainer_filter = Filter::new()
@@ -1462,12 +1479,14 @@ async fn test_maintainer_announcement_reprocessed_immediately() {
         .identifier(identifier);
 
     let maintainer_found =
-        wait_for_event_on_relay(relay_b.url(), maintainer_filter, Duration::from_secs(2)).await;
+        wait_for_event_on_relay(relay_b.url(), maintainer_filter, Duration::from_secs(30)).await;
     assert!(
         maintainer_found,
         "Maintainer announcement should be re-processed and accepted in relay_b"
     );
 
+    // Immediate hot-cache re-processing must beat the scheduled retry path.
+    let elapsed = start.elapsed();
     assert!(
         elapsed.as_secs() < 15,
         "Re-processing should happen in <15 seconds, took {:?}",
@@ -1514,6 +1533,7 @@ async fn test_multiple_maintainers_all_reprocessed() {
     // land in relay_a's DB.  Each announcement lists relay_a only, so relay_b will reject
     // them when syncing (no owner announcement in relay_b's DB yet).
     let mut git_dirs = Vec::new();
+    let mut maintainer_notes = Vec::new();
     for (idx, maintainer_keys) in [&maintainer1_keys, &maintainer2_keys, &maintainer3_keys]
         .iter()
         .enumerate()
@@ -1541,6 +1561,12 @@ async fn test_multiple_maintainers_all_reprocessed() {
         ])
         .finalize(*maintainer_keys)
         .unwrap();
+        maintainer_notes.push(
+            announcement
+                .id
+                .to_bech32()
+                .expect("Failed to encode maintainer announcement ID"),
+        );
         send_to_relay(&relay_a, &announcement).await.unwrap();
         // Use push_unique_git_data_to_relay so each maintainer gets a distinct commit
         // hash.  Identical hashes cause git to skip pack transfer when the object
@@ -1584,11 +1610,15 @@ async fn test_multiple_maintainers_all_reprocessed() {
     let relay_b = TestRelay::start_with_sync(Some(relay_a.url().to_string())).await;
     println!("relay_b started at {}", relay_b.url());
 
-    // Give relay_b's SyncManager time to complete the initial negentropy sync with relay_a.
-    // The negentropy sync completes within ~200ms (NGIT_TEST=1 sets batch window to 200ms), but we
-    // allow extra time for slow CI environments.
-    tokio::time::sleep(Duration::from_secs(3)).await;
-    println!("✓ relay_b synced from relay_a (maintainer announcements should be in hot cache)");
+    // Wait for relay_b's initial negentropy sync on its observable outcome:
+    // every rejected maintainer announcement is logged by its note ID.
+    for note in &maintainer_notes {
+        assert!(
+            wait_for_log(&relay_b.log_path(), note, Duration::from_secs(30)).await,
+            "relay_b should sync, reject and index maintainer announcement {note}"
+        );
+    }
+    println!("✓ relay_b synced from relay_a (maintainer announcements in hot cache)");
 
     // Step 3: Send owner announcement to relay_b → goes to purgatory.
     let owner_npub = owner_keys
@@ -1634,10 +1664,8 @@ async fn test_multiple_maintainers_all_reprocessed() {
         push_git_data_to_relay(&relay_b, &owner_keys, identifier, &[&relay_b.domain()]).await;
     println!("✓ Owner git data pushed to relay_b (hot-cache re-processing should fire)");
 
-    // Step 5: Wait briefly for async processing to complete.
-    tokio::time::sleep(Duration::from_secs(1)).await;
-
-    // Step 6: Verify all four announcements are in relay_b's database.
+    // Step 5/6: Verify all four announcements are in relay_b's database.
+    // Bounded polls on the served events replace a fixed post-push sleep.
     for (name, keys) in [
         ("owner", &owner_keys),
         ("maintainer1", &maintainer1_keys),
@@ -1649,7 +1677,7 @@ async fn test_multiple_maintainers_all_reprocessed() {
             .author(keys.public_key())
             .identifier(identifier);
 
-        let found = wait_for_event_on_relay(relay_b.url(), filter, Duration::from_secs(2)).await;
+        let found = wait_for_event_on_relay(relay_b.url(), filter, Duration::from_secs(30)).await;
         assert!(found, "{} announcement should be in relay_b", name);
     }
 
@@ -1695,9 +1723,10 @@ async fn test_invalid_maintainer_pubkey_handled_gracefully() {
             .finalize(&maintainer_keys)
             .unwrap();
 
-    // Send maintainer announcement - expect it to be rejected
+    // Send maintainer announcement - expect it to be rejected. `send_event`
+    // waits for the relay's OK response, so the rejection has already been
+    // processed when it returns.
     let _ = client.send_event(&maintainer_announcement).await;
-    tokio::time::sleep(Duration::from_millis(200)).await;
 
     // Step 2: Send owner announcement with INVALID maintainer hex, then push git data.
     // The announcement goes to purgatory first; the git push promotes it.
@@ -1728,16 +1757,16 @@ async fn test_invalid_maintainer_pubkey_handled_gracefully() {
     send_to_relay(&relay, &owner_announcement).await.unwrap();
     let _git_dir =
         push_git_data_to_relay(&relay, &owner_keys, identifier, &[&relay.domain()]).await;
-    tokio::time::sleep(Duration::from_millis(500)).await;
 
-    // Step 3: Verify owner announcement accepted, maintainer not re-processed
+    // Step 3: Verify owner announcement accepted, maintainer not re-processed.
+    // The bounded poll on the served announcement replaces a fixed sleep.
     let owner_filter = Filter::new()
         .kind(Kind::GitRepoAnnouncement)
         .author(owner_keys.public_key())
         .identifier(identifier);
 
     let owner_found =
-        wait_for_event_on_relay(relay.url(), owner_filter, Duration::from_secs(2)).await;
+        wait_for_event_on_relay(relay.url(), owner_filter, Duration::from_secs(30)).await;
     assert!(
         owner_found,
         "Owner announcement should be accepted despite invalid maintainer"
diff --git a/tests/sync/metrics.rs b/tests/sync/metrics.rs
index 9c60bff..cbfb9be 100644
--- a/tests/sync/metrics.rs
+++ b/tests/sync/metrics.rs
@@ -28,15 +28,31 @@ use crate::common::{
 // Format and Availability Tests (Keepers)
 // ============================================================================
 
+/// Poll the metrics endpoint until it responds, returning the first scrape.
+///
+/// The HTTP endpoint can come up slightly after the relay's WebSocket
+/// listener, so a bounded poll replaces the fixed startup sleeps these tests
+/// used to rely on.
+async fn wait_for_metrics_ready(relay_url: &str, timeout: Duration) -> String {
+    let deadline = tokio::time::Instant::now() + timeout;
+    loop {
+        if let Ok(metrics) = fetch_metrics(relay_url).await {
+            return metrics;
+        }
+        assert!(
+            tokio::time::Instant::now() < deadline,
+            "metrics endpoint at {relay_url} did not become ready before the deadline"
+        );
+        tokio::time::sleep(Duration::from_millis(100)).await;
+    }
+}
+
 /// Test that Prometheus text format is valid
 #[tokio::test]
 async fn test_prometheus_format_valid() {
     let relay = TestRelay::start().await;
-    tokio::time::sleep(Duration::from_millis(500)).await;
 
-    let metrics = fetch_metrics(relay.url())
-        .await
-        .expect("Failed to fetch metrics");
+    let metrics = wait_for_metrics_ready(relay.url(), Duration::from_secs(30)).await;
 
     relay.stop().await;
 
@@ -65,9 +81,10 @@ async fn test_metrics_availability_during_sync() {
     let source_relay = TestRelay::start().await;
     let sync_relay = TestRelay::start_with_sync(Some(source_relay.url().into())).await;
 
-    tokio::time::sleep(Duration::from_millis(500)).await;
+    wait_for_metrics_ready(sync_relay.url(), Duration::from_secs(30)).await;
 
-    // Make multiple metrics requests while sync is active
+    // Make multiple metrics requests while sync is active. The 200ms gap is
+    // deliberate scrape spacing, not a wait for a condition.
     for i in 0..3 {
         let metrics = fetch_metrics(sync_relay.url()).await;
         assert!(
@@ -91,7 +108,7 @@ async fn test_concurrent_metrics_requests() {
     let source_relay = TestRelay::start().await;
     let sync_relay = TestRelay::start_with_sync(Some(source_relay.url().into())).await;
 
-    tokio::time::sleep(Duration::from_secs(1)).await;
+    wait_for_metrics_ready(sync_relay.url(), Duration::from_secs(30)).await;
 
     // Clone the URL string so we have an owned value for spawned tasks
     let sync_url: String = sync_relay.url().to_string();
@@ -135,11 +152,8 @@ async fn test_concurrent_metrics_requests() {
 #[tokio::test]
 async fn test_metric_values_are_numeric() {
     let relay = TestRelay::start().await;
-    tokio::time::sleep(Duration::from_millis(500)).await;
 
-    let metrics = fetch_metrics(relay.url())
-        .await
-        .expect("Should fetch metrics");
+    let metrics = wait_for_metrics_ready(relay.url(), Duration::from_secs(30)).await;
 
     relay.stop().await;
 
@@ -276,9 +290,18 @@ async fn test_startup_sync_event_count() {
         setup_announcement_on_relay(&syncing_relay, &keys, &domain_refs, repo_id).await;
     println!("Announcement set up on syncing relay (triggers discovery of source)");
 
-    // 9. Wait for discovery + sync to complete
-    println!("Waiting 5s for discovery and sync...");
-    tokio::time::sleep(Duration::from_secs(5)).await;
+    // 9. Wait for discovery + sync on the observable condition the test
+    // cares about: the patches arriving on the syncing relay.
+    let patches_filter = Filter::new()
+        .kind(Kind::Custom(Kind::GitPatch.as_u16()))
+        .author(keys.public_key());
+    let patches_synced = crate::common::sync_helpers::wait_for_event_on_relay(
+        syncing_relay.url(),
+        patches_filter,
+        Duration::from_secs(30),
+    )
+    .await;
+    println!("Patches synced to syncing relay: {}", patches_synced);
 
     // 10. Fetch and parse metrics
     let raw_metrics = fetch_metrics(syncing_relay.url())
@@ -305,19 +328,6 @@ async fn test_startup_sync_event_count() {
     println!("Relays connected: {:?}", connected);
     println!("Events synced total: {:?}", events_synced);
 
-    // 12. Verify patches actually synced (functional check)
-    let filter = Filter::new()
-        .kind(Kind::Custom(Kind::GitPatch.as_u16()))
-        .author(keys.public_key());
-
-    let patches_synced = crate::common::sync_helpers::wait_for_event_on_relay(
-        syncing_relay.url(),
-        filter,
-        Duration::from_secs(2),
-    )
-    .await;
-    println!("Patches synced to syncing relay: {}", patches_synced);
-
     // Cleanup
     syncing_relay.stop().await;
     source_relay.stop().await;
@@ -364,27 +374,30 @@ async fn test_connection_failure_increments_counter() {
     let mut harness = MetricsTestHarness::with_sources(0).await; // No sources
     harness.start_syncing_relay_to_nowhere().await;
 
-    // Wait for initial connection attempt to the unreachable bootstrap relay
-    tokio::time::sleep(Duration::from_secs(2)).await;
-
-    let metrics = harness.get_metrics().await.unwrap();
-
-    // Failure counter should be recorded when connecting to unreachable relay
-    let failures = metrics
-        .counter(
-            "ngit_sync_connection_attempts_total",
-            &[("result", "failure")],
-        )
-        .unwrap_or(0);
+    // Poll for the failure counter recorded by the connection attempt to the
+    // unreachable bootstrap relay, bounded rather than sleeping a fixed time.
+    let deadline = tokio::time::Instant::now() + Duration::from_secs(30);
+    let failures = loop {
+        let metrics = harness.get_metrics().await.unwrap();
+        let failures = metrics
+            .counter(
+                "ngit_sync_connection_attempts_total",
+                &[("result", "failure")],
+            )
+            .unwrap_or(0);
+        if failures >= 1 {
+            break failures;
+        }
+        assert!(
+            tokio::time::Instant::now() < deadline,
+            "Expected at least 1 connection failure to be recorded, got {}",
+            failures
+        );
+        tokio::time::sleep(Duration::from_millis(200)).await;
+    };
 
     println!("Connection failures recorded: {}", failures);
 
-    assert!(
-        failures >= 1,
-        "Expected at least 1 connection failure to be recorded, got {}",
-        failures
-    );
-
     harness.stop_all().await;
 }
 
@@ -473,16 +486,19 @@ async fn test_live_sync_event_count() {
     client.disconnect().await;
     println!("Two patches sent to source relay (live mode)");
 
-    // Wait for live events to be processed and metrics updated
-    tokio::time::sleep(Duration::from_secs(4)).await;
-
-    // Fetch metrics from syncing relay
-    let raw_metrics = fetch_metrics(&sync_url)
-        .await
-        .expect("Failed to fetch metrics");
-    let metrics = ParsedMetrics::parse(&raw_metrics);
-
-    let synced_count = metrics.events_synced_total();
+    // Poll for the live events being processed and counted in metrics,
+    // bounded rather than sleeping a fixed time.
+    let deadline = tokio::time::Instant::now() + Duration::from_secs(30);
+    let synced_count = loop {
+        let raw_metrics = fetch_metrics(&sync_url)
+            .await
+            .expect("Failed to fetch metrics");
+        let synced_count = ParsedMetrics::parse(&raw_metrics).events_synced_total();
+        if synced_count.is_some_and(|count| count >= 2) || tokio::time::Instant::now() >= deadline {
+            break synced_count;
+        }
+        tokio::time::sleep(Duration::from_millis(200)).await;
+    };
     println!("Events synced total: {:?}", synced_count);
 
     // Cleanup
@@ -521,7 +537,7 @@ async fn test_relay_connected_status() {
 
     // Stop the source
     harness.stop_source(0).await;
-    let deadline = tokio::time::Instant::now() + Duration::from_secs(10);
+    let deadline = tokio::time::Instant::now() + Duration::from_secs(30);
     let metrics = loop {
         let metrics = harness.get_metrics().await.unwrap();
         if metrics.relay_connected(&source_url).is_none() {
@@ -559,28 +575,24 @@ async fn test_health_state_degrades_on_failure() {
     let mut harness = MetricsTestHarness::with_sources(0).await;
     harness.start_syncing_relay_to_nowhere().await;
 
-    // Initially might be trying to connect
-    tokio::time::sleep(Duration::from_secs(1)).await;
-    let initial = harness.get_metrics().await.unwrap();
-
-    // After several failures, should degrade (status = 2 or 3)
-    tokio::time::sleep(Duration::from_secs(5)).await;
-    let later = harness.get_metrics().await.unwrap();
-
-    // Get the relay status (1=healthy, 2=degraded, 3=dead)
-    let status = later.gauge("ngit_sync_relay_status", &[]).unwrap_or(0);
-
-    println!(
-        "Initial metrics: {:?}",
-        initial.gauge("ngit_sync_relay_status", &[])
-    );
-    println!("Later status: {}", status);
-
-    assert!(
-        status >= 2,
-        "Health should degrade to 2 (degraded) or 3 (dead), got {}",
-        status
-    );
+    // Poll the relay status gauge (1=healthy, 2=degraded, 3=dead) until it
+    // reflects the repeated connection failures, bounded rather than sleeping
+    // a fixed time.
+    let deadline = tokio::time::Instant::now() + Duration::from_secs(30);
+    let status = loop {
+        let metrics = harness.get_metrics().await.unwrap();
+        let status = metrics.gauge("ngit_sync_relay_status", &[]).unwrap_or(0);
+        if status >= 2 {
+            break status;
+        }
+        assert!(
+            tokio::time::Instant::now() < deadline,
+            "Health should degrade to 2 (degraded) or 3 (dead), got {}",
+            status
+        );
+        tokio::time::sleep(Duration::from_millis(200)).await;
+    };
+    println!("Degraded status: {}", status);
 
     harness.stop_all().await;
 }
@@ -617,13 +629,14 @@ async fn test_multi_source_aggregate_counts() {
     );
     harness.submit_events(0, &[announcement]).await.unwrap();
 
-    // Now start syncing relay - it should sync the existing announcement
+    // Now start syncing relay - it should sync the existing announcement.
+    // The loop below already polls for the connected state, so no fixed
+    // settling sleep is needed first.
     harness
         .start_syncing_relay_on_reservation(0, sync_reservation)
         .await;
-    tokio::time::sleep(Duration::from_secs(2)).await;
 
-    let deadline = tokio::time::Instant::now() + Duration::from_secs(10);
+    let deadline = tokio::time::Instant::now() + Duration::from_secs(30);
     let metrics = loop {
         let metrics = harness.get_metrics().await.unwrap();
         if metrics.relays_tracked_total() == Some(1) && metrics.relays_connected_total() == Some(1)
diff --git a/tests/sync/purgatory_fetch.rs b/tests/sync/purgatory_fetch.rs
index 0f09fcb..76d5788 100644
--- a/tests/sync/purgatory_fetch.rs
+++ b/tests/sync/purgatory_fetch.rs
@@ -158,8 +158,10 @@ async fn purgatory_fetch_batches_available_tips_and_isolates_missing_oids() {
         .add_relay(source.url())
         .await
         .expect("add source relay");
-    source_client.connect().await;
-    tokio::time::sleep(Duration::from_millis(500)).await;
+    source_client
+        .connect()
+        .and_wait(Duration::from_secs(30))
+        .await;
     source_client
         .send_event(&source_announcement)
         .await
@@ -236,8 +238,10 @@ async fn purgatory_fetch_batches_available_tips_and_isolates_missing_oids() {
         .add_relay(mock.url())
         .await
         .expect("add mock relay");
-    mock_client.connect().await;
-    tokio::time::sleep(Duration::from_millis(500)).await;
+    mock_client
+        .connect()
+        .and_wait(Duration::from_secs(30))
+        .await;
     mock_client
         .send_event(&syncing_announcement)
         .await