diff --git a/zebra-network/src/peer_set/candidate_set.rs b/zebra-network/src/peer_set/candidate_set.rs
index eb02c464..9458bbea 100644
--- a/zebra-network/src/peer_set/candidate_set.rs
+++ b/zebra-network/src/peer_set/candidate_set.rs
@@ -368,7 +368,8 @@ where
     /// Zebra resists distributed denial of service attacks by making sure that
     /// new peer connections are initiated at least
     /// [`MIN_OUTBOUND_PEER_CONNECTION_INTERVAL`][constants::MIN_OUTBOUND_PEER_CONNECTION_INTERVAL]
-    /// apart.
+    /// apart. If `next()` has recently provided a peer, then its future will sleep
+    /// until the rate-limit has passed.
     ///
     /// [`Responded`]: crate::PeerAddrState::Responded
     pub async fn next(&mut self) -> Option<MetaAddr> {
diff --git a/zebra-network/src/peer_set/initialize.rs b/zebra-network/src/peer_set/initialize.rs
index 30404ac6..6b2ae33e 100644
--- a/zebra-network/src/peer_set/initialize.rs
+++ b/zebra-network/src/peer_set/initialize.rs
@@ -870,6 +870,12 @@ where
 
                 // Spawn each handshake or crawl into an independent task, so handshakes can make
                 // progress while crawls are running.
+                //
+                // # Concurrency
+                //
+                // The peer crawler must be able to make progress even if some handshakes are
+                // rate-limited. So the async mutex and next peer timeout are awaited inside the
+                // spawned task.
                 let handshake_or_crawl_handle = tokio::spawn(
                     async move {
                         // Try to get the next available peer for a handshake.