diff --git a/CHANGELOG.md b/CHANGELOG.md
index b1d15e6..be117b3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,13 @@ is `0`, minor versions may carry breaking changes — they are called out here.
 
 ## [Unreleased]
 
+### Fixed
+- **A burst of connections during boot could end the version watch for good.** Its
+  first connection panicked when the OS had no thread to give: nothing showed it,
+  and a promoted version no longer reached that instance until a restart. It
+  needs a thread limit and a burst at that moment. The watch and the control loop
+  now carry a name and start again after a panic.
+
 ## [0.4.0] — 2026-11-06
 
 A security release: upgrade from any 0.3.x, and rotate any secret a shared or
diff --git a/core/src/control.rs b/core/src/control.rs
index 7bb4b66..65f724e 100644
--- a/core/src/control.rs
+++ b/core/src/control.rs
@@ -27,7 +27,9 @@ pub fn start(registry: Registry, root: PathBuf) {
     let Some(tenant) = env::var("VEJAS_TENANT").ok().filter(|t| !t.is_empty()) else {
         return;
     };
-    thread::spawn(move || supervise(registry, root, tenant));
+    crate::background("control", move || {
+        supervise(registry.clone(), root.clone(), tenant.clone())
+    });
 }
 
 fn supervise(registry: Registry, root: PathBuf, tenant: String) {
diff --git a/core/src/main.rs b/core/src/main.rs
index 1092efd..b056160 100644
--- a/core/src/main.rs
+++ b/core/src/main.rs
@@ -1929,6 +1929,32 @@ fn contained(attempt: impl FnOnce() -> Result<(), String>) -> Result<(), String>
     })
 }
 
+/// Start a background loop of the runtime itself, which no supervisor watches.
+/// `nats::connect` spawns threads and panics when the OS has none, and that used
+/// to end the loop for good with nothing to show for it: a burst of connections
+/// during boot took `version_watch`, and a promoted version no longer reached this
+/// instance. The thread carries a name, and a panic starts the loop again.
+fn background(name: &'static str, body: impl Fn() + Send + 'static) {
+    let spawned = thread::Builder::new().name(name.into()).spawn(move || {
+        while RUNNING.load(Ordering::SeqCst) {
+            let run = contained(|| {
+                body();
+                Ok(())
+            });
+            match run {
+                Ok(()) => break,
+                Err(e) => {
+                    eprintln!("[vejas] {name}: {e}; starting it again");
+                    thread::sleep(Duration::from_secs(2));
+                }
+            }
+        }
+    });
+    if let Err(e) = spawned {
+        eprintln!("[vejas] {name}: cannot start its thread: {e}");
+    }
+}
+
 /// Raises a flag when dropped, so that a helper thread stops on every way out of
 /// an attempt, a panic included.
 struct StopOnDrop(Arc<AtomicBool>);
@@ -4709,7 +4735,7 @@ fn main() {
     {
         let registry = registry.clone();
         let root = root.clone();
-        thread::spawn(move || version_watch(registry, root));
+        background("version-watch", move || version_watch(registry.clone(), root.clone()));
     }
 
     {
@@ -4741,6 +4767,22 @@ mod supervision_tests {
         assert_eq!(err, "panic: boom 7");
     }
 
+    #[test]
+    fn a_background_loop_starts_again_after_a_panic() {
+        let (tx, rx) = std::sync::mpsc::channel();
+        let runs = AtomicU64::new(0);
+        background("test-loop", move || {
+            let n = runs.fetch_add(1, Ordering::SeqCst);
+            tx.send(n).unwrap();
+            if n == 0 {
+                panic!("no thread to give");
+            }
+        });
+        let timeout = Duration::from_secs(10);
+        assert_eq!(rx.recv_timeout(timeout), Ok(0));
+        assert_eq!(rx.recv_timeout(timeout), Ok(1), "the loop came back");
+    }
+
     #[test]
     fn a_slot_comes_back_when_its_thread_panics() {
         let open = Arc::new(std::sync::atomic::AtomicUsize::new(0));
diff --git a/e2e/http-in-bound.sh b/e2e/http-in-bound.sh
index d7f2625..501712f 100755
--- a/e2e/http-in-bound.sh
+++ b/e2e/http-in-bound.sh
@@ -77,7 +77,8 @@ case "$C" in
 esac
 wait $HOLD 2>/dev/null
 wait_ingest $IN && ok "ingestion is back once the connections are gone" || ko "http-in did not recover"
-curl -s -m 5 "$H/metrics" | grep -q 'vejas_http_rejected_total{unit="connector:http_in"}' \
+METRICS=$(curl -s -m 5 "$H/metrics")
+[[ "$METRICS" == *'vejas_http_rejected_total{unit="connector:http_in"}'* ]] \
   && ok "the refusals are on /metrics" || ko "no vejas_http_rejected_total on /metrics"
 topology "$H" | grep -q 'running' && ok "the connector still reports running" || ko "the connector left running"
 
@@ -97,6 +98,9 @@ else
     debian:trixie-slim vejas-runtime > /dev/null
   for _ in $(seq 1 100); do curl -sf -o /dev/null "$H2/healthz" && break; sleep 0.1; done
   wait_ingest $IN2 || ko "http-in never came up in the container"
+  # Past the boot: a burst that lands on the first NATS connections is another
+  # case, covered by background(), and it made this check flaky.
+  sleep 5
   hold $IN2 300 3; HOLD=$!
   sleep 1
   wait $HOLD 2>/dev/null
@@ -104,6 +108,8 @@ else
     || ko "http-in died on the pids limit"
   P=$(docker logs "$CT" 2>&1 | grep -c panicked)
   [ "$P" = "0" ] && ok "no panic in the log" || ko "$P panic(s) in the log"
+  W=$(docker exec "$CT" sh -c 'cat /proc/1/task/*/comm' | grep -c version-watch)
+  [ "$W" -ge 1 ] && ok "the version watch is alive" || ko "the version watch died"
   UNIT=$(topology "$H2")
   echo "$UNIT" | grep -q '"restarts": 0' && ok "no restart was needed" || ko "the connector restarted: $UNIT"
 fi