diff --git a/CHANGELOG.md b/CHANGELOG.md
index 715b13d..9372458 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -145,6 +145,17 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   `handshake.rs:1114` is unchanged. Introduces a shared
   `Node::outbound_admission_check()` helper so the invariant is
   grep-able and unit-testable.
+- Inbound `handle_msg1` now silent-drops at `node.limits.max_peers`
+  saturation *before* building/sending Msg2, instead of replying with
+  Msg2 and then rejecting at `promote_connection`. Adds an early cap
+  check positioned after identity verification (so the
+  reconnect / cross-connection bypass for known peers still fires) and
+  before index allocation + Msg2 wire send. The late cap check inside
+  `promote_connection` is intentionally retained as
+  defense-in-depth. Wire savings observed in a 45 s ops tcpdump at
+  saturation: ~3.6 cap-denials/s × Msg2 (~104 B + AEAD compute) each.
+  Bigger win is cleaner peer-side semantics — no fake-completed
+  handshake whose subsequent data frames fail decryption on this side.
 - Mesh-size estimator (`compute_mesh_size`) no longer double-counts the
   parent's bloom cardinality during the transient cache window after a
   local parent-switch. Symptom: `fipsctl show status` / fipstop displayed
diff --git a/src/node/handlers/handshake.rs b/src/node/handlers/handshake.rs
index 2702e25..0743bea 100644
--- a/src/node/handlers/handshake.rs
+++ b/src/node/handlers/handshake.rs
@@ -207,6 +207,35 @@ impl Node {
             possible_restart = true;
         }
 
+        // Early cap check: at max_peers and this is a net-new identity?
+        // Bypass for known peers (reconnect / cross-connection) — admitting
+        // them doesn't grow peers.len(). This silent-drops the Msg1 before
+        // the Msg2 build/send and index allocation, avoiding wasted wire
+        // bytes and giving the peer cleaner semantics (no fake-completed
+        // handshake whose data frames subsequently fail decryption here).
+        // The late cap check inside promote_connection() is intentionally
+        // left in place as defense-in-depth.
+        if self.max_peers > 0 && self.peers.len() >= self.max_peers {
+            let is_known_active = self.peers.contains_key(&peer_node_addr);
+            let is_pending_outbound = self.connections.iter().any(|(_, conn)| {
+                conn.expected_identity()
+                    .map(|id| *id.node_addr() == peer_node_addr)
+                    .unwrap_or(false)
+            });
+            if !is_known_active && !is_pending_outbound {
+                debug!(
+                    peer = %self.peer_display_name(&peer_node_addr),
+                    max = self.max_peers,
+                    "Silent-dropping Msg1 at max_peers cap (early gate; no Msg2 sent)"
+                );
+                // `link_id` was allocated above but `conn` is still a local
+                // (not yet inserted into self.connections / self.links /
+                // self.addr_to_link), so the local drop suffices.
+                self.msg1_rate_limiter.complete_handshake();
+                return;
+            }
+        }
+
         // Epoch-based restart detection and duplicate msg1 handling.
         //
         // If we fell through from the addr_to_link check above with
diff --git a/src/node/tests/unit.rs b/src/node/tests/unit.rs
index e04f7b5..ed654a4 100644
--- a/src/node/tests/unit.rs
+++ b/src/node/tests/unit.rs
@@ -1526,3 +1526,227 @@ fn nostr_discovery_outbound_admission_atomic_roundtrip() {
         "after recovery store: traversal initiator/responder must see true"
     );
 }
+
+/// Sender-side helper: build a wire-format Msg1 from a fresh peer
+/// identity targeting `node_b`, *and* send it on the wire over `socket_a`
+/// to `addr_b`. Returns the sender's NodeAddr so the test can assert on
+/// identity-keyed maps.
+///
+/// Uses the same outbound-PeerConnection->Noise IK pattern as the
+/// integration handshake tests, but inlined and unit-scoped.
+async fn craft_and_send_msg1(
+    node_b: &Node,
+    sender_identity: &Identity,
+    socket_a: &tokio::net::UdpSocket,
+    addr_b: std::net::SocketAddr,
+    timestamp_ms: u64,
+) -> NodeAddr {
+    use crate::node::wire::build_msg1;
+    use crate::utils::index::SessionIndex;
+
+    let peer_b_identity = PeerIdentity::from_pubkey_full(node_b.identity.pubkey_full());
+    let sender_pubkey_id = PeerIdentity::from_pubkey_full(sender_identity.pubkey_full());
+    let sender_node_addr = *sender_pubkey_id.node_addr();
+
+    let link_id = LinkId::new(0xDEAD_BEEF);
+    let mut conn = PeerConnection::outbound(link_id, peer_b_identity, timestamp_ms);
+
+    let sender_keypair = sender_identity.keypair();
+    let mut startup_epoch = [0u8; 8];
+    rand::Rng::fill_bytes(&mut rand::rng(), &mut startup_epoch);
+    let noise_msg1 = conn
+        .start_handshake(sender_keypair, startup_epoch, timestamp_ms)
+        .expect("start_handshake should produce noise msg1");
+
+    let sender_index = SessionIndex::new(0x5151);
+    let wire_msg1 = build_msg1(sender_index, &noise_msg1);
+
+    socket_a
+        .send_to(&wire_msg1, addr_b)
+        .await
+        .expect("sender_socket.send_to");
+    sender_node_addr
+}
+
+/// Helper: deliver a packet from `node`'s registered UDP transport to
+/// `node.handle_msg1`. Returns Ok(()) on success or Err if the packet
+/// was not received within `timeout`.
+async fn pump_one_msg1_into_node(
+    node: &mut Node,
+    packet_rx: &mut crate::transport::PacketRx,
+    timeout_ms: u64,
+) -> Result<(), &'static str> {
+    use tokio::time::{Duration, timeout};
+    let packet = timeout(Duration::from_millis(timeout_ms), packet_rx.recv())
+        .await
+        .map_err(|_| "timed out waiting for msg1 on packet_rx")?
+        .ok_or("packet channel closed")?;
+    node.handle_msg1(packet).await;
+    Ok(())
+}
+
+/// Verifies the early max_peers cap check in `handle_msg1` silent-drops
+/// a Msg1 from a brand-new identity at saturation: no peer is admitted,
+/// no Msg2 response goes back on the wire, and the msg1 rate-limiter
+/// pending_count returns to baseline.
+///
+/// Wire-observable Msg2 absence is the load-bearing discriminator. With
+/// the early cap gate removed (stash-verify), the late gate inside
+/// `promote_connection` still rejects the new identity — but only
+/// *after* `handle_msg1` has already built the Msg2 frame and
+/// `transport.send(...wire_msg2)` has put it on the wire. The
+/// post-call wire-side poll catches that Msg2 (FAIL pre-fix; the
+/// silent timeout is the PASS post-fix).
+#[tokio::test]
+async fn handle_msg1_silent_drops_at_cap_for_new_peer() {
+    use crate::config::UdpConfig;
+    use tokio::time::{Duration, timeout};
+
+    let mut node = make_node();
+    node.set_max_peers(2);
+    inject_dummy_peers(&mut node, 2);
+    assert_eq!(node.peer_count(), 2, "precondition: at cap");
+
+    // === UDP transport setup for node_b (the unit under test) ===
+    let transport_id_b = TransportId::new(1);
+    let udp_config = UdpConfig {
+        bind_addr: Some("127.0.0.1:0".to_string()),
+        mtu: Some(1280),
+        ..Default::default()
+    };
+    let (packet_tx_b, mut packet_rx_b) = packet_channel(64);
+    let mut transport_b = UdpTransport::new(transport_id_b, None, udp_config, packet_tx_b);
+    transport_b.start_async().await.unwrap();
+    let addr_b = transport_b.local_addr().unwrap();
+    node.transports
+        .insert(transport_id_b, TransportHandle::Udp(transport_b));
+
+    // === Sender-side socket ===
+    let socket_a = tokio::net::UdpSocket::bind("127.0.0.1:0")
+        .await
+        .expect("bind sender socket");
+
+    let before_peers = node.peer_count();
+    let before_pending = node.msg1_rate_limiter.pending_count();
+
+    // Fresh sender identity — never seen by `node`.
+    let sender = Identity::generate();
+    let sender_node_addr = craft_and_send_msg1(&node, &sender, &socket_a, addr_b, 1000).await;
+
+    // Sanity: new identity is not currently a peer.
+    assert!(
+        !node.peers.contains_key(&sender_node_addr),
+        "precondition: new sender not yet a peer"
+    );
+
+    // Pump the wire-arrived Msg1 into the node's handler.
+    pump_one_msg1_into_node(&mut node, &mut packet_rx_b, 1000)
+        .await
+        .expect("msg1 must reach packet_rx_b");
+
+    // Post-call state checks.
+    assert_eq!(
+        node.peer_count(),
+        before_peers,
+        "early cap gate must not adopt a new peer at saturation"
+    );
+    assert!(
+        !node.peers.contains_key(&sender_node_addr),
+        "new sender must not appear in peers map"
+    );
+    assert_eq!(
+        node.msg1_rate_limiter.pending_count(),
+        before_pending,
+        "rate limiter must rebalance: start_handshake() then \
+         complete_handshake() before silent-drop return"
+    );
+
+    // Wire-observable discriminator: with the early gate in place, no
+    // Msg2 should come back. With the gate removed, Msg2 IS sent
+    // before promote_connection rejects.
+    let mut buf = [0u8; 2048];
+    let recv = timeout(Duration::from_millis(300), socket_a.recv_from(&mut buf)).await;
+    let received_bytes = recv.ok().and_then(|inner| inner.ok()).map(|(n, _)| n);
+    assert!(
+        received_bytes.is_none(),
+        "Msg2 must NOT be sent in response when at max_peers cap; \
+         observed {received_bytes:?} wire bytes — the fingerprint of \
+         the late-gate path replying with Msg2 before rejecting"
+    );
+}
+
+/// Verifies the bypass: at saturation, an inbound Msg1 from an
+/// *existing* peer's identity is not silent-dropped by the early cap
+/// check (the gate would otherwise wedge legitimate
+/// reconnect/restart/rekey traffic against an at-cap node).
+///
+/// The cap-gate's `is_known_active = self.peers.contains_key(&peer_node_addr)`
+/// branch admits this case; the downstream handling (restart-detect or
+/// duplicate-msg1 resend) then runs per existing semantics. The
+/// observable assertion here is the existing peer's continued
+/// presence — the rate-limiter rebalance is the same in
+/// bypass-admit and silent-drop, so this test isn't a discriminator
+/// against the no-gate (stash) build; it's a regression check that the
+/// gate doesn't accidentally evict known peers.
+#[tokio::test]
+async fn handle_msg1_admits_existing_peer_at_cap() {
+    use crate::config::UdpConfig;
+
+    let mut node = make_node();
+    node.set_max_peers(2);
+
+    inject_dummy_peers(&mut node, 1);
+
+    let existing_sender = Identity::generate();
+    let existing_pid = PeerIdentity::from_pubkey_full(existing_sender.pubkey_full());
+    let existing_node_addr = *existing_pid.node_addr();
+    let existing_link_id = LinkId::new(7777);
+    {
+        use crate::peer::ActivePeer;
+        let peer = ActivePeer::new(existing_pid, existing_link_id, 0);
+        node.peers.insert(existing_node_addr, peer);
+    }
+    assert_eq!(node.peer_count(), 2, "precondition: at cap");
+
+    let transport_id_b = TransportId::new(1);
+    let udp_config = UdpConfig {
+        bind_addr: Some("127.0.0.1:0".to_string()),
+        mtu: Some(1280),
+        ..Default::default()
+    };
+    let (packet_tx_b, mut packet_rx_b) = packet_channel(64);
+    let mut transport_b = UdpTransport::new(transport_id_b, None, udp_config, packet_tx_b);
+    transport_b.start_async().await.unwrap();
+    let addr_b = transport_b.local_addr().unwrap();
+    node.transports
+        .insert(transport_id_b, TransportHandle::Udp(transport_b));
+
+    let socket_a = tokio::net::UdpSocket::bind("127.0.0.1:0")
+        .await
+        .expect("bind sender socket");
+
+    let before_pending = node.msg1_rate_limiter.pending_count();
+
+    let sender_node_addr =
+        craft_and_send_msg1(&node, &existing_sender, &socket_a, addr_b, 2000).await;
+    assert_eq!(
+        sender_node_addr, existing_node_addr,
+        "sanity: crafted msg1 carries the existing peer's NodeAddr"
+    );
+
+    pump_one_msg1_into_node(&mut node, &mut packet_rx_b, 1000)
+        .await
+        .expect("msg1 must reach packet_rx_b");
+
+    // Bypass must not evict the existing peer or grow peer count.
+    assert_eq!(node.peer_count(), 2, "peer count unchanged");
+    assert!(
+        node.peers.contains_key(&existing_node_addr),
+        "existing peer must still be present after bypass-admitted msg1"
+    );
+    assert_eq!(
+        node.msg1_rate_limiter.pending_count(),
+        before_pending,
+        "rate limiter must rebalance after the (bypass-admitted) handler returns"
+    );
+}
diff --git a/testing/ci-local.sh b/testing/ci-local.sh
index 34661e5..c74701c 100755
--- a/testing/ci-local.sh
+++ b/testing/ci-local.sh
@@ -57,6 +57,7 @@ ONLY_SUITE=""
 # All integration suites matching ci.yml
 STATIC_SUITES=(static-mesh static-chain)
 REKEY_SUITES=(rekey rekey-accept-off rekey-outbound-only)
+ADMISSION_SUITES=(admission-cap)
 # Each entry: "display-name scenario [--flag value ...]"
 CHAOS_SUITES=(
     "smoke-10 smoke-10"
@@ -111,6 +112,9 @@ list_suites() {
     echo "  Rekey:"
     for s in "${REKEY_SUITES[@]}"; do echo "    $s"; done
     echo ""
+    echo "  Admission cap:"
+    for s in "${ADMISSION_SUITES[@]}"; do echo "    $s"; done
+    echo ""
     echo "  Gateway:"
     for s in "${GATEWAY_SUITES[@]}"; do echo "    $s"; done
     echo ""
@@ -320,6 +324,34 @@ run_rekey() {
     record "rekey" $rc
 }
 
+# Run the admission-cap integration test
+# Verifies the inbound max_peers early-gate silent-drops at scale by
+# lowering node.max_peers on one mesh node and asserting via tcpdump
+# that no Msg2 responses go to the sustained-retrying denied peers.
+run_admission_cap() {
+    local compose="testing/static/docker-compose.yml"
+    local rc=0
+
+    info "[admission-cap] Generating configs"
+    bash testing/static/scripts/generate-configs.sh mesh || { record "admission-cap" 1; return; }
+    bash testing/static/scripts/admission-cap-test.sh inject-config || { record "admission-cap" 1; return; }
+
+    info "[admission-cap] Starting containers (mesh profile)"
+    docker compose -f "$compose" --profile mesh up -d || { record "admission-cap" 1; return; }
+
+    info "[admission-cap] Running admission-cap test"
+    if bash testing/static/scripts/admission-cap-test.sh; then
+        rc=0
+    else
+        rc=1
+        info "[admission-cap] Collecting failure logs"
+        docker compose -f "$compose" --profile mesh logs --no-color 2>&1 | tail -100
+    fi
+
+    docker compose -f "$compose" --profile mesh down --volumes --remove-orphans 2>/dev/null
+    record "admission-cap" $rc
+}
+
 # Run a chaos scenario
 run_chaos() {
     local name="$1"
@@ -568,6 +600,11 @@ run_integration() {
     run_rekey_accept_off
     run_rekey_outbound_only
 
+    # Admission cap (mesh profile, max_peers=1 on one node)
+    for _suite in "${ADMISSION_SUITES[@]}"; do
+        run_admission_cap
+    done
+
     # Gateway
     run_gateway
 
@@ -670,6 +707,8 @@ run_suite() {
             run_rekey_accept_off ;;
         rekey-outbound-only)
             run_rekey_outbound_only ;;
+        admission-cap)
+            run_admission_cap ;;
         gateway)
             run_gateway ;;
         acl-allowlist)
diff --git a/testing/static/scripts/admission-cap-test.sh b/testing/static/scripts/admission-cap-test.sh
new file mode 100755
index 0000000..c88448f
--- /dev/null
+++ b/testing/static/scripts/admission-cap-test.sh
@@ -0,0 +1,205 @@
+#!/bin/bash
+# Integration test for the inbound max_peers admission gate.
+#
+# Verifies the silent-drop behavior of the early-gate in handle_msg1 at
+# scale, using the mesh topology with one node's node.max_peers lowered
+# to 1. This forces 2 of node-c's 3 configured peers (b, d, e) into a
+# sustained denied state, and asserts via tcpdump that no Msg2 responses
+# go back to those denied peers across a 60s capture window.
+#
+# Tested behavior:
+#   - Denied peers DO arrive at the cap'd node (inbound FMP-IK Msg1, 84 B)
+#   - Cap'd node sends NO Msg2 responses (104 B) to denied peers
+#   - Cap'd node maintains exactly max_peers active sessions
+#   - Admitted peer's session stays healthy throughout the window
+#
+# Usage:
+#   ./admission-cap-test.sh                Run the test (containers must be up)
+#   ./admission-cap-test.sh inject-config  Inject node.max_peers into generated configs
+
+set -e
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+CAP_NODE="${ADMISSION_CAP_NODE:-c}"
+MAX_PEERS="${ADMISSION_MAX_PEERS:-1}"
+CAPTURE_SECS="${ADMISSION_CAPTURE_SECS:-60}"
+TOPOLOGY="mesh"
+TOPO_FILE="$SCRIPT_DIR/../configs/topologies/$TOPOLOGY.yaml"
+
+# ── inject-config subcommand ─────────────────────────────────────────
+# Inject node.max_peers into generated configs. Called separately by CI
+# before building Docker images.
+if [ "${1:-}" = "inject-config" ]; then
+    echo "Injecting node.limits.max_peers: $MAX_PEERS into node-$CAP_NODE ($TOPOLOGY topology)..."
+    cfg="$SCRIPT_DIR/../generated-configs/$TOPOLOGY/node-$CAP_NODE.yaml"
+    if [ ! -f "$cfg" ]; then
+        echo "  Error: $cfg not found (run generate-configs.sh $TOPOLOGY first)" >&2
+        exit 1
+    fi
+    # Insert under node.limits (the actual config path per src/config/node.rs).
+    # Three cases: limits.max_peers already present (update), limits: present
+    # without max_peers (append), or no limits block (insert full subtree).
+    if grep -qE "^    max_peers:" "$cfg"; then
+        sed -i -E "s/^    max_peers: *[0-9]+/    max_peers: $MAX_PEERS/" "$cfg"
+    elif grep -qE "^  limits:" "$cfg"; then
+        sed -i "/^  limits:/a\\
+    max_peers: $MAX_PEERS" "$cfg"
+    else
+        sed -i "/^node:/a\\
+  limits:\\
+    max_peers: $MAX_PEERS" "$cfg"
+    fi
+    echo "  node-$CAP_NODE limits block:"
+    sed -n '/^  limits:/,/^  [a-z]/p' "$cfg" | head -5
+    exit 0
+fi
+
+stamp() { date '+%H:%M:%S'; }
+info() { echo "[$(stamp)] $*"; }
+fail() { echo "[$(stamp)] FAIL: $*"; exit 1; }
+pass() { echo "[$(stamp)] PASS: $*"; }
+
+# Extract docker_ip for a node from the topology file
+node_ip() {
+    grep -A 5 "^  $1:" "$TOPO_FILE" \
+        | grep -m1 'docker_ip:' \
+        | sed 's/.*: *"*\([^"]*\)".*/\1/'
+}
+
+# Extract npub for a node from the topology file
+node_npub() {
+    grep -A 5 "^  $1:" "$TOPO_FILE" \
+        | grep -m1 'npub:' \
+        | sed 's/.*: *"*\([^"]*\)".*/\1/'
+}
+
+# Extract configured peers list for a node from the topology file
+node_peers() {
+    grep -A 5 "^  $1:" "$TOPO_FILE" \
+        | grep -m1 'peers:' \
+        | sed 's/.*\[\(.*\)\].*/\1/' \
+        | tr -d ' ' \
+        | tr ',' ' '
+}
+
+CAP_IP=$(node_ip "$CAP_NODE")
+[ -n "$CAP_IP" ] || fail "could not resolve docker_ip for node-$CAP_NODE in $TOPO_FILE"
+info "cap'd node: node-$CAP_NODE (ip $CAP_IP, max_peers=$MAX_PEERS)"
+
+# ── Phase 1: wait for convergence ────────────────────────────────────
+info "phase 1: wait for node-$CAP_NODE peer_count to reach $MAX_PEERS (90s timeout)"
+deadline=$(($(date +%s) + 90))
+pc=0
+while [ "$(date +%s)" -lt "$deadline" ]; do
+    pc=$(docker exec fips-node-$CAP_NODE fipsctl show status 2>/dev/null \
+        | grep -m1 peer_count | sed 's/.*: *//' | tr -d ',' || echo 0)
+    [ "$pc" = "$MAX_PEERS" ] && break
+    sleep 2
+done
+[ "$pc" = "$MAX_PEERS" ] \
+    || fail "node-$CAP_NODE peer_count=$pc after 90s, expected $MAX_PEERS"
+info "node-$CAP_NODE converged: peer_count=$pc"
+
+# Identify admitted vs denied peers among configured peers
+ADMITTED_NPUBS=$(docker exec fips-node-$CAP_NODE fipsctl show peers 2>/dev/null \
+    | grep -oE 'npub1[a-z0-9]+' | sort -u || true)
+DENIED=""
+ADMITTED=""
+for p in $(node_peers "$CAP_NODE"); do
+    npub=$(node_npub "$p")
+    if echo "$ADMITTED_NPUBS" | grep -q "$npub"; then
+        ADMITTED="$ADMITTED $p"
+    else
+        DENIED="$DENIED $p"
+    fi
+done
+ADMITTED=$(echo $ADMITTED | xargs)
+DENIED=$(echo $DENIED | xargs)
+info "admitted: ${ADMITTED:-<none>}"
+info "denied (sustained-retry): ${DENIED:-<none>}"
+[ -n "$DENIED" ] \
+    || fail "no denied peers — test setup wrong (cap=$MAX_PEERS too high vs configured peers)"
+
+# ── Phase 2: capture wire traffic for CAPTURE_SECS seconds ───────────
+# Drives sustained load by restarting denied peer containers on a cadence
+# during the capture window. Each restart resets the auto-reconnect
+# exponential backoff (5s base / 300s cap), producing a fresh burst of
+# Msg1s that exercises the silent-drop gate at meaningful rate. Without
+# this loop the gate fires ~3-4 times per denied peer in a 60s window;
+# with restarts every 15s we get ~30-50 firings across both denied peers.
+info "phase 2: capture UDP/2121 on node-$CAP_NODE for ${CAPTURE_SECS}s, with denied-peer restart loop"
+CAP_FILE=$(mktemp /tmp/admission-cap-pcap.XXXXXX.txt)
+HELPER_IMAGE=$(docker inspect -f '{{.Config.Image}}' fips-node-$CAP_NODE 2>/dev/null)
+[ -n "$HELPER_IMAGE" ] || fail "could not resolve helper image from fips-node-$CAP_NODE"
+
+# Background: cycle denied peers to reset their backoff and drive load.
+(
+    elapsed=0
+    while [ $elapsed -lt $((CAPTURE_SECS - 5)) ]; do
+        sleep 15
+        elapsed=$((elapsed + 15))
+        for n in $DENIED; do
+            docker restart "fips-node-$n" >/dev/null 2>&1 &
+        done
+        wait
+        info "  [load-driver] restarted denied peers ($DENIED) at t+${elapsed}s"
+    done
+) &
+LOAD_PID=$!
+
+# Foreground: tcpdump capture for CAPTURE_SECS
+docker run --rm --net=container:fips-node-$CAP_NODE \
+    --cap-add NET_ADMIN --cap-add NET_RAW \
+    --entrypoint sh "$HELPER_IMAGE" \
+    -c "timeout $CAPTURE_SECS tcpdump -nn -i any 'udp port 2121' -l 2>&1 || true" \
+    > "$CAP_FILE" 2>&1
+
+# Reap load-driver if it's still running (should be ~done)
+wait $LOAD_PID 2>/dev/null || true
+
+captured=$(wc -l < "$CAP_FILE")
+info "captured $captured tcpdump lines → $CAP_FILE"
+
+# ── Phase 3: per-denied-peer wire-level assertion ────────────────────
+info "phase 3: per-denied-peer assertion (inbound Msg1 > 0, outbound Msg2 == 0)"
+OVERALL=0
+TOTAL_MSG1_IN=0
+TOTAL_MSG2_OUT=0
+for n in $DENIED; do
+    n_ip=$(node_ip "$n")
+    # Inbound: src=n_ip:* → dst=cap_ip:2121; FMP-IK Msg1 wire size = 84 B
+    msg1_in=$(grep -cE "IP $n_ip\.[0-9]+ > $CAP_IP\.2121: UDP, length 84" "$CAP_FILE" || true)
+    # Outbound: src=cap_ip:2121 → dst=n_ip:*; FMP-IK Msg2 wire size = 104 B
+    msg2_out=$(grep -cE "IP $CAP_IP\.2121 > $n_ip\.[0-9]+: UDP, length 104" "$CAP_FILE" || true)
+    info "  node-$n ($n_ip): inbound Msg1 (len 84) = $msg1_in, outbound Msg2 (len 104) = $msg2_out"
+    TOTAL_MSG1_IN=$((TOTAL_MSG1_IN + msg1_in))
+    TOTAL_MSG2_OUT=$((TOTAL_MSG2_OUT + msg2_out))
+    if [ "$msg1_in" -eq 0 ]; then
+        info "    FAIL: expected inbound Msg1 retries from denied peer (peer not sustained-retrying?)"
+        OVERALL=1
+    fi
+    if [ "$msg2_out" -gt 0 ]; then
+        info "    FAIL: silent-drop gate leaked — expected 0 outbound Msg2, got $msg2_out"
+        OVERALL=1
+    fi
+done
+
+# ── Phase 4: cap'd node still at exactly max_peers ───────────────────
+pc_final=$(docker exec fips-node-$CAP_NODE fipsctl show status 2>/dev/null \
+    | grep -m1 peer_count | sed 's/.*: *//' | tr -d ',' || echo 0)
+info "node-$CAP_NODE final peer_count=$pc_final (expected $MAX_PEERS)"
+[ "$pc_final" = "$MAX_PEERS" ] || OVERALL=1
+
+if [ "$OVERALL" -eq 0 ]; then
+    pass "admission-cap: silent-drop gate verified at scale"
+    pass "  denied peers: $(echo $DENIED | wc -w), capture: ${CAPTURE_SECS}s"
+    pass "  total inbound Msg1 from denied: $TOTAL_MSG1_IN (sustained retries observed)"
+    pass "  total outbound Msg2 to denied:  $TOTAL_MSG2_OUT (silent-drop holds)"
+    rm -f "$CAP_FILE"
+    exit 0
+else
+    info "--- tcpdump capture tail (last 50 lines) ---"
+    tail -50 "$CAP_FILE"
+    fail "admission-cap: see failures above (capture preserved at $CAP_FILE)"
+fi