diff --git a/src/proxy/client.rs b/src/proxy/client.rs
index 849e409..5d32e34 100644
--- a/src/proxy/client.rs
+++ b/src/proxy/client.rs
@@ -878,7 +878,7 @@ impl RunningClientHandler {
             {
                 Ok(reservation) => reservation,
                 Err(e) => {
-                    warn!(user = %user, error = %e, "User limit exceeded");
+                    warn!(user = %user, error = %e, "User admission check failed");
                     return Err(e);
                 }
             };
@@ -998,8 +998,8 @@ impl RunningClientHandler {
 
     #[cfg(test)]
     async fn check_user_limits_static(
-        user: &str, 
-        config: &ProxyConfig, 
+        user: &str,
+        config: &ProxyConfig,
         stats: &Stats,
         peer_addr: SocketAddr,
         ip_tracker: &UserIpTracker,
diff --git a/src/proxy/client_security_tests.rs b/src/proxy/client_security_tests.rs
index 8bdb234..6ca2d4b 100644
--- a/src/proxy/client_security_tests.rs
+++ b/src/proxy/client_security_tests.rs
@@ -1420,6 +1420,105 @@ async fn tcp_limit_rejection_does_not_reserve_ip_or_trigger_rollback() {
     );
 }
 
+#[tokio::test]
+async fn zero_tcp_limit_rejects_without_ip_or_counter_side_effects() {
+    let mut config = ProxyConfig::default();
+    config
+        .access
+        .user_max_tcp_conns
+        .insert("user".to_string(), 0);
+
+    let stats = Stats::new();
+    let ip_tracker = UserIpTracker::new();
+    let peer_addr: SocketAddr = "198.51.100.211:50001".parse().unwrap();
+
+    let result = RunningClientHandler::check_user_limits_static(
+        "user",
+        &config,
+        &stats,
+        peer_addr,
+        &ip_tracker,
+    )
+    .await;
+
+    assert!(matches!(
+        result,
+        Err(ProxyError::ConnectionLimitExceeded { user }) if user == "user"
+    ));
+    assert_eq!(stats.get_user_curr_connects("user"), 0);
+    assert_eq!(ip_tracker.get_active_ip_count("user").await, 0);
+}
+
+#[tokio::test]
+async fn concurrent_distinct_ip_rejections_rollback_user_counter_without_leak() {
+    let user = "rollback-storm-user";
+    let mut config = ProxyConfig::default();
+    config
+        .access
+        .user_max_tcp_conns
+        .insert(user.to_string(), 128);
+
+    let config = Arc::new(config);
+    let stats = Arc::new(Stats::new());
+    let ip_tracker = Arc::new(UserIpTracker::new());
+    ip_tracker.set_user_limit(user, 1).await;
+
+    let keeper_peer: SocketAddr = "198.51.100.212:50002".parse().unwrap();
+    let keeper = RunningClientHandler::acquire_user_connection_reservation_static(
+        user,
+        &config,
+        stats.clone(),
+        keeper_peer,
+        ip_tracker.clone(),
+    )
+    .await
+    .expect("keeper reservation must succeed");
+
+    let mut tasks = tokio::task::JoinSet::new();
+    for i in 0..64u8 {
+        let config = config.clone();
+        let stats = stats.clone();
+        let ip_tracker = ip_tracker.clone();
+        tasks.spawn(async move {
+            let peer = SocketAddr::new(
+                IpAddr::V4(std::net::Ipv4Addr::new(198, 51, 101, i.saturating_add(1))),
+                41000 + i as u16,
+            );
+            let result = RunningClientHandler::acquire_user_connection_reservation_static(
+                user,
+                &config,
+                stats,
+                peer,
+                ip_tracker,
+            )
+            .await;
+            assert!(matches!(
+                result,
+                Err(ProxyError::ConnectionLimitExceeded { user }) if user == "rollback-storm-user"
+            ));
+        });
+    }
+
+    while let Some(joined) = tasks.join_next().await {
+        joined.unwrap();
+    }
+
+    assert_eq!(
+        stats.get_user_curr_connects(user),
+        1,
+        "failed distinct-IP attempts must rollback acquired user slots"
+    );
+    assert_eq!(
+        ip_tracker.get_active_ip_count(user).await,
+        1,
+        "failed distinct-IP attempts must not leave extra active IPs"
+    );
+
+    keeper.release().await;
+    assert_eq!(stats.get_user_curr_connects(user), 0);
+    assert_eq!(ip_tracker.get_active_ip_count(user).await, 0);
+}
+
 #[tokio::test]
 async fn explicit_reservation_release_cleans_user_and_ip_immediately() {
     let user = "release-user";
@@ -2990,3 +3089,478 @@ async fn tls_record_len_16384_is_accepted_in_client_handler_pipeline() {
         "Valid max-length ClientHello must not increment bad counter"
     );
 }
+
+fn lcg_next(state: &mut u64) -> u64 {
+    *state = state.wrapping_mul(6364136223846793005).wrapping_add(1);
+    *state
+}
+
+async fn wait_for_user_and_ip_zero(
+    stats: &Arc<Stats>,
+    ip_tracker: &Arc<UserIpTracker>,
+    user: &str,
+) {
+    tokio::time::timeout(Duration::from_secs(2), async {
+        loop {
+            if stats.get_user_curr_connects(user) == 0
+                && ip_tracker.get_active_ip_count(user).await == 0
+            {
+                break;
+            }
+            tokio::time::sleep(Duration::from_millis(5)).await;
+        }
+    })
+    .await
+    .expect("cleanup must converge to zero user and IP footprint");
+}
+
+async fn burst_acquire_distinct_ips(
+    user: &'static str,
+    config: Arc<ProxyConfig>,
+    stats: Arc<Stats>,
+    ip_tracker: Arc<UserIpTracker>,
+    third_octet: u8,
+    attempts: u16,
+) -> (Vec<UserConnectionReservation>, usize) {
+    let mut tasks = tokio::task::JoinSet::new();
+    for i in 0..attempts {
+        let config = config.clone();
+        let stats = stats.clone();
+        let ip_tracker = ip_tracker.clone();
+        tasks.spawn(async move {
+            let host = (i as u8).saturating_add(1);
+            let peer = SocketAddr::new(
+                IpAddr::V4(std::net::Ipv4Addr::new(198, 51, third_octet, host)),
+                55000 + i,
+            );
+            RunningClientHandler::acquire_user_connection_reservation_static(
+                user,
+                &config,
+                stats,
+                peer,
+                ip_tracker,
+            )
+            .await
+        });
+    }
+
+    let mut successes = Vec::new();
+    let mut failures = 0usize;
+    while let Some(joined) = tasks.join_next().await {
+        match joined.expect("burst acquire task must not panic") {
+            Ok(reservation) => successes.push(reservation),
+            Err(err) => {
+                assert!(matches!(
+                    err,
+                    ProxyError::ConnectionLimitExceeded { user: ref denied_user }
+                        if denied_user == user
+                ));
+                failures = failures.saturating_add(1);
+            }
+        }
+    }
+
+    (successes, failures)
+}
+
+#[tokio::test]
+async fn deterministic_mixed_reservation_churn_preserves_counter_and_eventual_cleanup() {
+    let user = "deterministic-churn-user";
+    let mut config = ProxyConfig::default();
+    config
+        .access
+        .user_max_tcp_conns
+        .insert(user.to_string(), 12);
+
+    let config = Arc::new(config);
+    let stats = Arc::new(Stats::new());
+    let ip_tracker = Arc::new(UserIpTracker::new());
+    ip_tracker.set_user_limit(user, 4).await;
+
+    let mut seed = 0xD1F2_A4C8_991B_77E1u64;
+    let mut reservations: Vec<Option<UserConnectionReservation>> = Vec::new();
+
+    for step in 0..220u64 {
+        let op = (lcg_next(&mut seed) % 100) as u8;
+        let active = reservations.iter().filter(|entry| entry.is_some()).count();
+
+        if active == 0 || op < 55 {
+            let ip_octet = (lcg_next(&mut seed) % 16 + 1) as u8;
+            let peer = SocketAddr::new(
+                IpAddr::V4(std::net::Ipv4Addr::new(198, 51, 120, ip_octet)),
+                52000 + (step % 4000) as u16,
+            );
+            let result = RunningClientHandler::acquire_user_connection_reservation_static(
+                user,
+                &config,
+                stats.clone(),
+                peer,
+                ip_tracker.clone(),
+            )
+            .await;
+
+            if let Ok(reservation) = result {
+                reservations.push(Some(reservation));
+            } else {
+                assert!(matches!(
+                    result,
+                    Err(ProxyError::ConnectionLimitExceeded { user }) if user == "deterministic-churn-user"
+                ));
+            }
+        } else {
+            let selected = reservations
+                .iter()
+                .enumerate()
+                .filter(|(_, entry)| entry.is_some())
+                .map(|(idx, _)| idx)
+                .nth((lcg_next(&mut seed) as usize) % active)
+                .unwrap();
+
+            let reservation = reservations[selected].take().unwrap();
+            if op < 80 {
+                reservation.release().await;
+            } else {
+                std::thread::spawn(move || {
+                    drop(reservation);
+                })
+                .join()
+                .expect("cross-thread drop must not panic");
+            }
+        }
+
+        let live_slots = reservations.iter().filter(|entry| entry.is_some()).count() as u64;
+        assert_eq!(
+            stats.get_user_curr_connects(user),
+            live_slots,
+            "current-connects counter must match number of live reservations"
+        );
+        assert!(
+            stats.get_user_curr_connects(user) <= 12,
+            "current-connects must stay within configured TCP limit"
+        );
+        assert!(
+            ip_tracker.get_active_ip_count(user).await <= 4,
+            "active unique IPs must stay within configured per-user IP limit"
+        );
+    }
+
+    for reservation in reservations.into_iter().flatten() {
+        reservation.release().await;
+    }
+    wait_for_user_and_ip_zero(&stats, &ip_tracker, user).await;
+}
+
+#[tokio::test]
+async fn cross_thread_drop_storm_then_parallel_reacquire_wave_has_no_leak() {
+    let user = "drop-storm-reacquire-user";
+    let mut config = ProxyConfig::default();
+    config
+        .access
+        .user_max_tcp_conns
+        .insert(user.to_string(), 64);
+
+    let config = Arc::new(config);
+    let stats = Arc::new(Stats::new());
+    let ip_tracker = Arc::new(UserIpTracker::new());
+    ip_tracker.set_user_limit(user, 8).await;
+
+    let mut initial = Vec::new();
+    for i in 0..32u16 {
+        let ip_octet = (i % 8 + 1) as u8;
+        let peer = SocketAddr::new(
+            IpAddr::V4(std::net::Ipv4Addr::new(203, 0, 120, ip_octet)),
+            53000 + i,
+        );
+        let reservation = RunningClientHandler::acquire_user_connection_reservation_static(
+            user,
+            &config,
+            stats.clone(),
+            peer,
+            ip_tracker.clone(),
+        )
+        .await
+        .expect("initial reservation must succeed");
+        initial.push(reservation);
+    }
+
+    let mut second_half = initial.split_off(16);
+
+    let mut releases = Vec::new();
+    for reservation in initial {
+        releases.push(tokio::spawn(async move {
+            reservation.release().await;
+        }));
+    }
+    for release_task in releases {
+        release_task.await.expect("release task must not panic");
+    }
+
+    let mut drop_threads = Vec::new();
+    for reservation in second_half.drain(..) {
+        drop_threads.push(std::thread::spawn(move || {
+            drop(reservation);
+        }));
+    }
+    for drop_thread in drop_threads {
+        drop_thread
+            .join()
+            .expect("cross-thread drop worker must not panic");
+    }
+
+    wait_for_user_and_ip_zero(&stats, &ip_tracker, user).await;
+
+    let mut reacquire_tasks = tokio::task::JoinSet::new();
+    for i in 0..16u16 {
+        let config = config.clone();
+        let stats = stats.clone();
+        let ip_tracker = ip_tracker.clone();
+        reacquire_tasks.spawn(async move {
+            let peer = SocketAddr::new(
+                IpAddr::V4(std::net::Ipv4Addr::new(198, 51, 121, (i + 1) as u8)),
+                54000 + i,
+            );
+            RunningClientHandler::acquire_user_connection_reservation_static(
+                user,
+                &config,
+                stats,
+                peer,
+                ip_tracker,
+            )
+            .await
+        });
+    }
+
+    let mut acquired = Vec::new();
+    while let Some(joined) = reacquire_tasks.join_next().await {
+        match joined.expect("reacquire task must not panic") {
+            Ok(reservation) => acquired.push(reservation),
+            Err(err) => {
+                assert!(matches!(
+                    err,
+                    ProxyError::ConnectionLimitExceeded { user }
+                        if user == "drop-storm-reacquire-user"
+                ));
+            }
+        }
+    }
+
+    assert!(
+        acquired.len() <= 8,
+        "parallel distinct-IP reacquire wave must not exceed per-user unique IP limit"
+    );
+    for reservation in acquired {
+        reservation.release().await;
+    }
+    wait_for_user_and_ip_zero(&stats, &ip_tracker, user).await;
+}
+
+#[tokio::test]
+async fn scheduled_near_limit_and_burst_windows_preserve_admission_invariants() {
+    let user: &'static str = "scheduled-attack-user";
+    let mut config = ProxyConfig::default();
+    config
+        .access
+        .user_max_tcp_conns
+        .insert(user.to_string(), 6);
+
+    let config = Arc::new(config);
+    let stats = Arc::new(Stats::new());
+    let ip_tracker = Arc::new(UserIpTracker::new());
+    ip_tracker.set_user_limit(user, 2).await;
+
+    let mut base = Vec::new();
+    for i in 0..5u16 {
+        let peer = SocketAddr::new(IpAddr::V4(std::net::Ipv4Addr::new(198, 51, 130, 1)), 56000 + i);
+        let reservation = RunningClientHandler::acquire_user_connection_reservation_static(
+            user,
+            &config,
+            stats.clone(),
+            peer,
+            ip_tracker.clone(),
+        )
+        .await
+        .expect("near-limit warmup reservation must succeed");
+        base.push(reservation);
+    }
+    assert_eq!(stats.get_user_curr_connects(user), 5);
+    assert_eq!(ip_tracker.get_active_ip_count(user).await, 1);
+
+    let (wave1_success, wave1_fail) = burst_acquire_distinct_ips(
+        user,
+        config.clone(),
+        stats.clone(),
+        ip_tracker.clone(),
+        131,
+        32,
+    )
+    .await;
+    assert_eq!(wave1_success.len(), 1);
+    assert_eq!(wave1_fail, 31);
+    assert_eq!(stats.get_user_curr_connects(user), 6);
+    assert_eq!(ip_tracker.get_active_ip_count(user).await, 2);
+
+    let released = base.pop().expect("must have releasable reservation");
+    released.release().await;
+    for reservation in wave1_success {
+        reservation.release().await;
+    }
+
+    tokio::time::timeout(Duration::from_secs(1), async {
+        loop {
+            if stats.get_user_curr_connects(user) == 4
+                && ip_tracker.get_active_ip_count(user).await == 1
+            {
+                break;
+            }
+            tokio::time::sleep(Duration::from_millis(5)).await;
+        }
+    })
+    .await
+    .expect("window cleanup must settle to expected occupancy");
+
+    let (wave2_success, wave2_fail) = burst_acquire_distinct_ips(
+        user,
+        config,
+        stats.clone(),
+        ip_tracker.clone(),
+        132,
+        32,
+    )
+    .await;
+    assert_eq!(wave2_success.len(), 1);
+    assert_eq!(wave2_fail, 31);
+    assert_eq!(stats.get_user_curr_connects(user), 5);
+    assert_eq!(ip_tracker.get_active_ip_count(user).await, 2);
+
+    let tail = base.split_off(2);
+
+    let mut drop_threads = Vec::new();
+    for reservation in base {
+        drop_threads.push(std::thread::spawn(move || {
+            drop(reservation);
+        }));
+    }
+    for drop_thread in drop_threads {
+        drop_thread
+            .join()
+            .expect("cross-thread scheduled cleanup must not panic");
+    }
+
+    for reservation in tail {
+        reservation.release().await;
+    }
+    for reservation in wave2_success {
+        reservation.release().await;
+    }
+
+    wait_for_user_and_ip_zero(&stats, &ip_tracker, user).await;
+}
+
+#[tokio::test]
+async fn scheduled_mode_switch_burst_churn_preserves_limits_and_cleanup() {
+    let user: &'static str = "scheduled-mode-switch-user";
+    let mut config = ProxyConfig::default();
+    config
+        .access
+        .user_max_tcp_conns
+        .insert(user.to_string(), 10);
+
+    let config = Arc::new(config);
+    let stats = Arc::new(Stats::new());
+    let ip_tracker = Arc::new(UserIpTracker::new());
+    ip_tracker.set_user_limit(user, 3).await;
+
+    let base_peer = SocketAddr::new(IpAddr::V4(std::net::Ipv4Addr::new(198, 51, 140, 1)), 57000);
+    let mut base = Vec::new();
+    for i in 0..7u16 {
+        let peer = SocketAddr::new(base_peer.ip(), base_peer.port().saturating_add(i));
+        let reservation = RunningClientHandler::acquire_user_connection_reservation_static(
+            user,
+            &config,
+            stats.clone(),
+            peer,
+            ip_tracker.clone(),
+        )
+        .await
+        .expect("base occupancy reservation must succeed");
+        base.push(reservation);
+    }
+
+    assert_eq!(stats.get_user_curr_connects(user), 7);
+    assert_eq!(ip_tracker.get_active_ip_count(user).await, 1);
+
+    for round in 0..8u8 {
+        let (wave_success, wave_fail) = burst_acquire_distinct_ips(
+            user,
+            config.clone(),
+            stats.clone(),
+            ip_tracker.clone(),
+            141u8.saturating_add(round),
+            24,
+        )
+        .await;
+
+        assert!(
+            wave_success.len() <= 2,
+            "burst must not exceed available unique-IP headroom under limit=3"
+        );
+        assert_eq!(wave_success.len() + wave_fail, 24);
+        assert_eq!(
+            stats.get_user_curr_connects(user),
+            7 + wave_success.len() as u64,
+            "slot counter must reflect base occupancy plus successful burst leases"
+        );
+        assert!(ip_tracker.get_active_ip_count(user).await <= 3);
+
+        if round % 2 == 0 {
+            for reservation in wave_success {
+                reservation.release().await;
+            }
+            let rotated = base.pop().expect("base rotation reservation must exist");
+            rotated.release().await;
+        } else {
+            for reservation in wave_success {
+                std::thread::spawn(move || {
+                    drop(reservation);
+                })
+                .join()
+                .expect("drop-heavy burst cleanup thread must not panic");
+            }
+            let rotated = base.pop().expect("base rotation reservation must exist");
+            std::thread::spawn(move || {
+                drop(rotated);
+            })
+            .join()
+            .expect("drop-heavy base cleanup thread must not panic");
+        }
+
+        let replacement = RunningClientHandler::acquire_user_connection_reservation_static(
+            user,
+            &config,
+            stats.clone(),
+            base_peer,
+            ip_tracker.clone(),
+        )
+        .await
+        .expect("base replacement reservation must succeed after each round");
+        base.push(replacement);
+
+        tokio::time::timeout(Duration::from_secs(1), async {
+            loop {
+                if stats.get_user_curr_connects(user) == 7
+                    && ip_tracker.get_active_ip_count(user).await <= 1
+                {
+                    break;
+                }
+                tokio::time::sleep(Duration::from_millis(5)).await;
+            }
+        })
+        .await
+        .expect("round cleanup must converge to steady base occupancy");
+    }
+
+    for reservation in base {
+        reservation.release().await;
+    }
+    wait_for_user_and_ip_zero(&stats, &ip_tracker, user).await;
+}
diff --git a/src/transport/middle_proxy/health.rs b/src/transport/middle_proxy/health.rs
index edc9598..1c2c648 100644
--- a/src/transport/middle_proxy/health.rs
+++ b/src/transport/middle_proxy/health.rs
@@ -186,9 +186,7 @@ pub(super) async fn reap_draining_writers(
         }
     }
 
-    let mut active_draining_writer_ids = HashSet::with_capacity(draining_writers.len());
     for writer in draining_writers {
-        active_draining_writer_ids.insert(writer.id);
         if drain_ttl_secs > 0
             && writer.draining_started_at_epoch_secs != 0
             && now_epoch_secs.saturating_sub(writer.draining_started_at_epoch_secs) > drain_ttl_secs
@@ -214,12 +212,9 @@ pub(super) async fn reap_draining_writers(
         {
             warn!(writer_id = writer.id, "Drain timeout, force-closing");
             force_close_writer_ids.push(writer.id);
-            active_draining_writer_ids.remove(&writer.id);
         }
     }
 
-    warn_next_allowed.retain(|writer_id, _| active_draining_writer_ids.contains(writer_id));
-
     let close_budget = health_drain_close_budget();
     let requested_force_close = force_close_writer_ids.len();
     let requested_empty_close = empty_writer_ids.len();
@@ -257,6 +252,18 @@ pub(super) async fn reap_draining_writers(
             "ME draining close backlog deferred to next health cycle"
         );
     }
+
+    // Keep warn cooldown state for draining writers still present in the pool;
+    // drop state only once a writer is actually removed.
+    let active_draining_writer_ids = {
+        let writers = pool.writers.read().await;
+        writers
+            .iter()
+            .filter(|writer| writer.draining.load(std::sync::atomic::Ordering::Relaxed))
+            .map(|writer| writer.id)
+            .collect::<HashSet<u64>>()
+    };
+    warn_next_allowed.retain(|writer_id, _| active_draining_writer_ids.contains(writer_id));
 }
 
 pub(super) fn health_drain_close_budget() -> usize {
diff --git a/src/transport/middle_proxy/health_adversarial_tests.rs b/src/transport/middle_proxy/health_adversarial_tests.rs
index 675005a..cd06fdf 100644
--- a/src/transport/middle_proxy/health_adversarial_tests.rs
+++ b/src/transport/middle_proxy/health_adversarial_tests.rs
@@ -1,4 +1,5 @@
 use std::collections::HashMap;
+use std::collections::HashSet;
 use std::net::{IpAddr, Ipv4Addr, SocketAddr};
 use std::sync::Arc;
 use std::sync::atomic::{AtomicBool, AtomicU8, AtomicU32, AtomicU64, Ordering};
@@ -181,6 +182,40 @@ async fn sorted_writer_ids(pool: &Arc<MePool>) -> Vec<u64> {
     ids
 }
 
+fn lcg_next(state: &mut u64) -> u64 {
+    *state = state.wrapping_mul(6364136223846793005).wrapping_add(1);
+    *state
+}
+
+async fn draining_writer_ids(pool: &Arc<MePool>) -> HashSet<u64> {
+    pool.writers
+        .read()
+        .await
+        .iter()
+        .filter(|writer| writer.draining.load(Ordering::Relaxed))
+        .map(|writer| writer.id)
+        .collect::<HashSet<u64>>()
+}
+
+async fn set_writer_runtime_state(
+    pool: &Arc<MePool>,
+    writer_id: u64,
+    draining: bool,
+    drain_started_at_epoch_secs: u64,
+    drain_deadline_epoch_secs: u64,
+) {
+    let writers = pool.writers.read().await;
+    if let Some(writer) = writers.iter().find(|writer| writer.id == writer_id) {
+        writer.draining.store(draining, Ordering::Relaxed);
+        writer
+            .draining_started_at_epoch_secs
+            .store(drain_started_at_epoch_secs, Ordering::Relaxed);
+        writer
+            .drain_deadline_epoch_secs
+            .store(drain_deadline_epoch_secs, Ordering::Relaxed);
+    }
+}
+
 #[tokio::test]
 async fn reap_draining_writers_clears_warn_state_when_pool_empty() {
     let (pool, _rng) = make_pool(128, 1, 1).await;
@@ -430,6 +465,149 @@ async fn me_health_monitor_eliminates_mixed_empty_and_deadline_backlog() {
     assert!(writer_count(&pool).await <= threshold as usize);
 }
 
+#[tokio::test]
+async fn reap_draining_writers_deterministic_mixed_state_churn_preserves_invariants() {
+    let threshold = 9u64;
+    let (pool, _rng) = make_pool(threshold, 1, 1).await;
+    let mut warn_next_allowed = HashMap::new();
+    let mut seed = 0x9E37_79B9_7F4A_7C15u64;
+    let mut next_writer_id = 20_000u64;
+    let now_epoch_secs = MePool::now_epoch_secs();
+
+    for writer_id in 1..=72u64 {
+        let bound_clients = if writer_id % 4 == 0 { 0 } else { 1 };
+        let deadline = if writer_id % 5 == 0 {
+            now_epoch_secs.saturating_sub(1)
+        } else {
+            0
+        };
+        insert_draining_writer(
+            &pool,
+            writer_id,
+            now_epoch_secs.saturating_sub(500).saturating_add(writer_id),
+            bound_clients,
+            deadline,
+        )
+        .await;
+    }
+
+    for _round in 0..90 {
+        reap_draining_writers(&pool, &mut warn_next_allowed).await;
+
+        let draining_ids = draining_writer_ids(&pool).await;
+        assert!(
+            warn_next_allowed.keys().all(|id| draining_ids.contains(id)),
+            "warn-state keys must always be a subset of live draining writers"
+        );
+
+        let writer_ids = sorted_writer_ids(&pool).await;
+        if writer_ids.is_empty() {
+            continue;
+        }
+
+        let remove_n = (lcg_next(&mut seed) % 3) as usize;
+        for writer_id in writer_ids.iter().copied().take(remove_n) {
+            let _ = pool.remove_writer_and_close_clients(writer_id).await;
+        }
+
+        let survivors = sorted_writer_ids(&pool).await;
+        if !survivors.is_empty() {
+            let idx = (lcg_next(&mut seed) as usize) % survivors.len();
+            let target = survivors[idx];
+            set_writer_runtime_state(&pool, target, false, 0, 0).await;
+        }
+
+        let survivors = sorted_writer_ids(&pool).await;
+        if survivors.len() > 1 {
+            let idx = (lcg_next(&mut seed) as usize) % survivors.len();
+            let target = survivors[idx];
+            let expired_deadline = if lcg_next(&mut seed) & 1 == 0 {
+                now_epoch_secs.saturating_sub(1)
+            } else {
+                0
+            };
+            set_writer_runtime_state(
+                &pool,
+                target,
+                true,
+                now_epoch_secs.saturating_sub(120),
+                expired_deadline,
+            )
+            .await;
+        }
+
+        let inject_n = (lcg_next(&mut seed) % 4) as usize;
+        for _ in 0..inject_n {
+            let bound_clients = if lcg_next(&mut seed) & 1 == 0 { 0 } else { 1 };
+            let deadline = if lcg_next(&mut seed) & 1 == 0 {
+                now_epoch_secs.saturating_sub(1)
+            } else {
+                0
+            };
+            insert_draining_writer(
+                &pool,
+                next_writer_id,
+                now_epoch_secs.saturating_sub(240),
+                bound_clients,
+                deadline,
+            )
+            .await;
+            next_writer_id = next_writer_id.saturating_add(1);
+        }
+    }
+
+    for _ in 0..64 {
+        reap_draining_writers(&pool, &mut warn_next_allowed).await;
+        if writer_count(&pool).await <= threshold as usize {
+            break;
+        }
+    }
+
+    assert!(writer_count(&pool).await <= threshold as usize);
+    let draining_ids = draining_writer_ids(&pool).await;
+    assert!(warn_next_allowed.keys().all(|id| draining_ids.contains(id)));
+}
+
+#[tokio::test]
+async fn reap_draining_writers_repeated_draining_flips_never_leave_stale_warn_state() {
+    let (pool, _rng) = make_pool(64, 1, 1).await;
+    let now_epoch_secs = MePool::now_epoch_secs();
+
+    for writer_id in 1..=24u64 {
+        insert_draining_writer(
+            &pool,
+            writer_id,
+            now_epoch_secs.saturating_sub(240),
+            1,
+            0,
+        )
+        .await;
+    }
+
+    let mut warn_next_allowed = HashMap::new();
+    for _round in 0..48u64 {
+        for writer_id in 1..=24u64 {
+            let draining = (writer_id + _round) % 3 != 0;
+            set_writer_runtime_state(
+                &pool,
+                writer_id,
+                draining,
+                now_epoch_secs.saturating_sub(120),
+                0,
+            )
+            .await;
+        }
+
+        reap_draining_writers(&pool, &mut warn_next_allowed).await;
+
+        let draining_ids = draining_writer_ids(&pool).await;
+        assert!(
+            warn_next_allowed.keys().all(|id| draining_ids.contains(id)),
+            "warn-state map must not retain entries for writers outside draining set"
+        );
+    }
+}
+
 #[test]
 fn health_drain_close_budget_is_within_expected_bounds() {
     let budget = health_drain_close_budget();
diff --git a/src/transport/middle_proxy/health_regression_tests.rs b/src/transport/middle_proxy/health_regression_tests.rs
index 05a8e6a..fe73670 100644
--- a/src/transport/middle_proxy/health_regression_tests.rs
+++ b/src/transport/middle_proxy/health_regression_tests.rs
@@ -168,6 +168,21 @@ async fn current_writer_ids(pool: &Arc<MePool>) -> Vec<u64> {
     writer_ids
 }
 
+async fn writer_exists(pool: &Arc<MePool>, writer_id: u64) -> bool {
+    pool.writers
+        .read()
+        .await
+        .iter()
+        .any(|writer| writer.id == writer_id)
+}
+
+async fn set_writer_draining(pool: &Arc<MePool>, writer_id: u64, draining: bool) {
+    let writers = pool.writers.read().await;
+    if let Some(writer) = writers.iter().find(|writer| writer.id == writer_id) {
+        writer.draining.store(draining, Ordering::Relaxed);
+    }
+}
+
 #[tokio::test]
 async fn reap_draining_writers_drops_warn_state_for_removed_writer() {
     let pool = make_pool(128).await;
@@ -257,6 +272,123 @@ async fn reap_draining_writers_limits_closes_per_health_tick() {
     assert_eq!(pool.writers.read().await.len(), writer_total - close_budget);
 }
 
+#[tokio::test]
+async fn reap_draining_writers_keeps_warn_state_for_deadline_backlog_writers() {
+    let pool = make_pool(0).await;
+    let now_epoch_secs = MePool::now_epoch_secs();
+    let close_budget = health_drain_close_budget();
+    let writer_total = close_budget.saturating_add(5);
+    for writer_id in 1..=writer_total as u64 {
+        insert_draining_writer(
+            &pool,
+            writer_id,
+            now_epoch_secs.saturating_sub(60),
+            1,
+            now_epoch_secs.saturating_sub(1),
+        )
+        .await;
+    }
+    let target_writer_id = writer_total as u64;
+    let mut warn_next_allowed = HashMap::new();
+    warn_next_allowed.insert(
+        target_writer_id,
+        Instant::now() + Duration::from_secs(300),
+    );
+
+    reap_draining_writers(&pool, &mut warn_next_allowed).await;
+
+    assert!(writer_exists(&pool, target_writer_id).await);
+    assert!(warn_next_allowed.contains_key(&target_writer_id));
+}
+
+#[tokio::test]
+async fn reap_draining_writers_keeps_warn_state_for_overflow_backlog_writers() {
+    let pool = make_pool(1).await;
+    let now_epoch_secs = MePool::now_epoch_secs();
+    let close_budget = health_drain_close_budget();
+    let writer_total = close_budget.saturating_add(6);
+    for writer_id in 1..=writer_total as u64 {
+        insert_draining_writer(
+            &pool,
+            writer_id,
+            now_epoch_secs.saturating_sub(300).saturating_add(writer_id),
+            1,
+            0,
+        )
+        .await;
+    }
+    let target_writer_id = writer_total.saturating_sub(1) as u64;
+    let mut warn_next_allowed = HashMap::new();
+    warn_next_allowed.insert(
+        target_writer_id,
+        Instant::now() + Duration::from_secs(300),
+    );
+
+    reap_draining_writers(&pool, &mut warn_next_allowed).await;
+
+    assert!(writer_exists(&pool, target_writer_id).await);
+    assert!(warn_next_allowed.contains_key(&target_writer_id));
+}
+
+#[tokio::test]
+async fn reap_draining_writers_drops_warn_state_when_writer_exits_draining_state() {
+    let pool = make_pool(128).await;
+    let now_epoch_secs = MePool::now_epoch_secs();
+    insert_draining_writer(&pool, 71, now_epoch_secs.saturating_sub(60), 1, 0).await;
+
+    let mut warn_next_allowed = HashMap::new();
+    warn_next_allowed.insert(71, Instant::now() + Duration::from_secs(300));
+
+    set_writer_draining(&pool, 71, false).await;
+    reap_draining_writers(&pool, &mut warn_next_allowed).await;
+
+    assert!(writer_exists(&pool, 71).await);
+    assert!(
+        !warn_next_allowed.contains_key(&71),
+        "warn cooldown state must be dropped after writer leaves draining state"
+    );
+}
+
+#[tokio::test]
+async fn reap_draining_writers_preserves_warn_state_across_multiple_budget_deferrals() {
+    let pool = make_pool(0).await;
+    let now_epoch_secs = MePool::now_epoch_secs();
+    let close_budget = health_drain_close_budget();
+    let writer_total = close_budget.saturating_mul(2).saturating_add(1);
+    for writer_id in 1..=writer_total as u64 {
+        insert_draining_writer(
+            &pool,
+            writer_id,
+            now_epoch_secs.saturating_sub(120),
+            1,
+            now_epoch_secs.saturating_sub(1),
+        )
+        .await;
+    }
+
+    let tail_writer_id = writer_total as u64;
+    let mut warn_next_allowed = HashMap::new();
+    warn_next_allowed.insert(
+        tail_writer_id,
+        Instant::now() + Duration::from_secs(300),
+    );
+
+    reap_draining_writers(&pool, &mut warn_next_allowed).await;
+    assert!(writer_exists(&pool, tail_writer_id).await);
+    assert!(warn_next_allowed.contains_key(&tail_writer_id));
+
+    reap_draining_writers(&pool, &mut warn_next_allowed).await;
+    assert!(writer_exists(&pool, tail_writer_id).await);
+    assert!(warn_next_allowed.contains_key(&tail_writer_id));
+
+    reap_draining_writers(&pool, &mut warn_next_allowed).await;
+    assert!(!writer_exists(&pool, tail_writer_id).await);
+    assert!(
+        !warn_next_allowed.contains_key(&tail_writer_id),
+        "warn cooldown state must clear once writer is actually removed"
+    );
+}
+
 #[tokio::test]
 async fn reap_draining_writers_backlog_drains_across_ticks() {
     let pool = make_pool(128).await;