diff --git a/benchmarks/aft-search/slice-descriptors/land-flake-batch-11.json b/benchmarks/aft-search/slice-descriptors/land-flake-batch-11.json
new file mode 100644
index 000000000..a39f7240c
--- /dev/null
+++ b/benchmarks/aft-search/slice-descriptors/land-flake-batch-11.json
@@ -0,0 +1,12 @@
+{
+  "slice_class": "non_ranking",
+  "targeted_mechanism": "none",
+  "kind": "harness",
+  "fixtures": [
+    "harness-goldens"
+  ],
+  "slices": [
+    "flake-batch-11"
+  ],
+  "note": "Test-only: the watcher-drain rewind test waits by time and yields between slices instead of a 32-slice spin. No changed path matches RANKING_FENCE_PREFIXES."
+}
diff --git a/crates/aft/src/runtime_drain.rs b/crates/aft/src/runtime_drain.rs
index bd1bc80cf..c283e60c2 100644
--- a/crates/aft/src/runtime_drain.rs
+++ b/crates/aft/src/runtime_drain.rs
@@ -6785,10 +6785,18 @@ mod tests {
         drain.join().unwrap();
 
         ctx.mark_subc_bound();
-        let mut guard = 0;
+        // The replay must finish: the manual dispatch channel holds no
+        // platform watcher, so `has_more` can only come from the retained
+        // Apply continuation. A bound replay advances every slice; the only
+        // zero-progress park is transient LSP contention from the
+        // applicability worker spawned by the initial batch. Yield between
+        // slices so that worker can release the manager, and bound the wait
+        // by time rather than a fixed slice count, which is brittle under
+        // CI load.
+        let deadline = Instant::now() + Duration::from_secs(30);
         while drain_watcher_events_bounded(&ctx, WATCHER_PATH_DRAIN_BATCH_CAP).has_more {
-            guard += 1;
-            assert!(guard < 32, "rebased replay must finish");
+            assert!(Instant::now() < deadline, "rebased replay must finish");
+            std::thread::yield_now();
         }
         let search = ctx
             .search_index()