diff --git a/src/paperscout/__main__.py b/src/paperscout/__main__.py
index c0a42a6..fc6e7bd 100644
--- a/src/paperscout/__main__.py
+++ b/src/paperscout/__main__.py
@@ -14,7 +14,7 @@
 from typing import Any, cast
 
 from .config import settings
-from .db import init_db, init_pool
+from .db import init_db, init_pool, pool_status
 from .health import start_health_server
 from .monitor import PollResult, Scheduler
 from .protocols import DataSource, OpsAlertFn
@@ -171,6 +171,7 @@ async def _async_main() -> None:
     bolt_thread = None
     mq = None
     app = None
+    pool = None
 
     _register_shutdown_signals(asyncio.get_running_loop(), shutdown_event, shutdown_reason)
 
@@ -228,62 +229,51 @@ async def _async_main() -> None:
 
     launch_time = datetime.now(timezone.utc)
 
-    pool = init_pool(settings.database_url)
-    init_db(pool)
-
-    state = ProbeState(pool)
-    user_watchlist = UserWatchlist(pool)
-    index = WG21Index(pool, cfg=settings)
-    prober = ISOProber(index, state, user_watchlist)
-    sources: list[DataSource] = [index, prober]
-    if settings.enable_open_std:
-        sources.append(OpenStdSource())
-    app = create_app()
-    mq = MessageQueue(app)
-    mq.start()
-
-    def paper_count_fn() -> int:
-        return len(index.papers)
-
-    def _on_poll_result(result: PollResult) -> None:
-        notify_channel(app, result, mq)
-        notify_users(app, result, mq)
-
-    def _ops_alert(msg: str) -> None:
-        if settings.ops_alert_channel:
-            mq.enqueue(
-                settings.ops_alert_channel,
-                f":rotating_light: PaperScout alert: {msg}",
-            )
-
-    def _pool_status(p: Any) -> dict[str, Any]:
-        """Best-effort pool stats (psycopg2 ThreadedConnectionPool uses private attrs)."""
-        status: dict[str, Any] = {"max": getattr(p, "maxconn", None)}
-        try:
-            status["in_use"] = len(p._used)
-            status["available"] = len(p._pool)
-        except AttributeError:
-            status["in_use"] = None
-            status["available"] = None
-        return status
-
-    scheduler = Scheduler(
-        sources=sources,
-        user_watchlist=user_watchlist,
-        state=state,
-        cfg=settings,
-        notify_callback=_on_poll_result,
-        ops_alert_fn=cast(OpsAlertFn, _ops_alert),
-    )
+    try:
+        pool = init_pool(settings.database_url)
+        init_db(pool)
+
+        state = ProbeState(pool)
+        user_watchlist = UserWatchlist(pool)
+        index = WG21Index(pool, cfg=settings)
+        prober = ISOProber(index, state, user_watchlist)
+        sources: list[DataSource] = [index, prober]
+        if settings.enable_open_std:
+            sources.append(OpenStdSource())
+        app = create_app()
+        mq = MessageQueue(app)
+        mq.start()
+
+        def paper_count_fn() -> int:
+            return len(index.papers)
+
+        def _on_poll_result(result: PollResult) -> None:
+            notify_channel(app, result, mq)
+            notify_users(app, result, mq)
+
+        def _ops_alert(msg: str) -> None:
+            if settings.ops_alert_channel:
+                mq.enqueue(
+                    settings.ops_alert_channel,
+                    f":rotating_light: PaperScout alert: {msg}",
+                )
 
-    def _extra_health_fields() -> dict[str, Any]:
-        return _merge_extra_health_fields(
-            scheduler.health_snapshot(),
-            _mq_health_fields(mq),
-            _pool_status(pool),
+        scheduler = Scheduler(
+            sources=sources,
+            user_watchlist=user_watchlist,
+            state=state,
+            cfg=settings,
+            notify_callback=_on_poll_result,
+            ops_alert_fn=cast(OpsAlertFn, _ops_alert),
         )
 
-    try:
+        def _extra_health_fields() -> dict[str, Any]:
+            return _merge_extra_health_fields(
+                scheduler.health_snapshot(),
+                _mq_health_fields(mq),
+                pool_status(pool),
+            )
+
         register_handlers(app, user_watchlist, state, paper_count_fn, launch_time)
 
         health_server = start_health_server(
@@ -318,6 +308,7 @@ def _extra_health_fields() -> dict[str, Any]:
             bolt_thread=bolt_thread,
             mq_drain_timeout=settings.shutdown_mq_drain_timeout_seconds,
             thread_join_timeout=settings.shutdown_thread_join_timeout_seconds,
+            pool=pool,
         )
 
 
diff --git a/src/paperscout/db.py b/src/paperscout/db.py
index 02a627b..91b9f91 100644
--- a/src/paperscout/db.py
+++ b/src/paperscout/db.py
@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import logging
+from typing import Any
 
 from psycopg2 import pool as pg_pool
 
@@ -48,6 +49,29 @@ def init_pool(dsn: str, minconn: int = 1, maxconn: int = 10) -> pg_pool.Threaded
     return p
 
 
+def pool_status(p: pg_pool.ThreadedConnectionPool) -> dict[str, Any]:
+    """Report pool reachability via documented ``getconn``/``putconn`` only.
+
+    Borrows and immediately returns one connection as a liveness probe.
+    Does not read undocumented pool attributes or private internals.
+    """
+    try:
+        conn = p.getconn()
+    except AttributeError:
+        raise
+    except Exception:
+        return {"reachable": False}
+    try:
+        p.putconn(conn)
+    except Exception:
+        try:
+            conn.close()
+        except Exception:
+            pass
+        return {"reachable": False}
+    return {"reachable": True}
+
+
 def init_db(p: pg_pool.ThreadedConnectionPool) -> None:
     """Create all tables (idempotent)."""
     conn = p.getconn()
diff --git a/src/paperscout/shutdown.py b/src/paperscout/shutdown.py
index b3646b3..7a7560f 100644
--- a/src/paperscout/shutdown.py
+++ b/src/paperscout/shutdown.py
@@ -1,4 +1,4 @@
-"""Graceful process shutdown: drain MQ, stop HTTP servers, join worker threads."""
+"""Graceful process shutdown: drain MQ, stop HTTP servers, join worker threads, close DB pool."""
 
 from __future__ import annotations
 
@@ -6,6 +6,7 @@
 import threading
 from http.server import HTTPServer
 
+from psycopg2 import pool as pg_pool
 from slack_bolt import App
 
 from .scout import MessageQueue
@@ -49,6 +50,7 @@ def shutdown_services(
     bolt_thread: threading.Thread | None,
     mq_drain_timeout: float,
     thread_join_timeout: float,
+    pool: pg_pool.ThreadedConnectionPool | None = None,
 ) -> int:
     """Ordered teardown. Returns the number of messages drained from the queue."""
     drained = 0
@@ -83,4 +85,12 @@ def shutdown_services(
         reason,
         drained,
     )
+
+    if pool is not None:
+        try:
+            pool.closeall()
+            log.info("shutdown: DB pool closed")
+        except Exception:
+            log.exception("shutdown: DB pool close failed")
+
     return drained
diff --git a/tests/conftest.py b/tests/conftest.py
index 13249fb..39fe9e2 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -194,6 +194,7 @@ def __init__(self):
         self._store = _FakeStore()
         self.fail_on_commit = False
         self.rollback_count = 0
+        self.closeall_called = False
         self.call_log: list[tuple[str, Sequence]] = []
 
     def calls_matching(self, sql_fragment: str) -> list[tuple[str, Sequence]]:
@@ -221,6 +222,9 @@ def getconn(self):
     def putconn(self, conn):
         pass
 
+    def closeall(self) -> None:
+        self.closeall_called = True
+
     def seed_watchlist_raw(self, rows: list[tuple[str, str, str]]) -> None:
         """Directly populate ``user_watchlist`` rows for edge-case tests."""
         for uid, entry, etype in rows:
diff --git a/tests/test_db.py b/tests/test_db.py
index 3df8e63..1b19675 100644
--- a/tests/test_db.py
+++ b/tests/test_db.py
@@ -6,7 +6,7 @@
 
 import pytest
 
-from paperscout.db import init_db, init_pool
+from paperscout.db import init_db, init_pool, pool_status
 
 
 @patch("paperscout.db.pg_pool.ThreadedConnectionPool")
@@ -63,3 +63,32 @@ def test_init_db_putconn_even_when_execute_fails():
         init_db(pool)
 
     pool.putconn.assert_called_once_with(conn)
+
+
+def test_pool_status_uses_public_api_only():
+    pool = MagicMock()
+    conn = MagicMock()
+    pool.getconn.return_value = conn
+    assert pool_status(pool) == {"reachable": True}
+    pool.getconn.assert_called_once()
+    pool.putconn.assert_called_once_with(conn)
+
+
+def test_pool_status_returns_false_when_getconn_fails():
+    pool = MagicMock()
+    pool.getconn.side_effect = RuntimeError("pool exhausted")
+    assert pool_status(pool) == {"reachable": False}
+
+
+def test_pool_status_closes_conn_when_putconn_fails():
+    pool = MagicMock()
+    conn = MagicMock()
+    pool.getconn.return_value = conn
+    pool.putconn.side_effect = RuntimeError("putconn failed")
+    assert pool_status(pool) == {"reachable": False}
+    conn.close.assert_called_once()
+
+
+def test_pool_status_raises_for_incompatible_pool():
+    with pytest.raises(AttributeError):
+        pool_status(object())  # type: ignore[arg-type]
diff --git a/tests/test_health.py b/tests/test_health.py
index c258d97..8d981f4 100644
--- a/tests/test_health.py
+++ b/tests/test_health.py
@@ -51,7 +51,7 @@ def health_url_with_extras():
             "probe_stats": {},
             "probe_success_rate": 0.5,
             "mq_depth": 3,
-            "db_pool": {"max": 10, "in_use": 1, "available": 9},
+            "db_pool": {"reachable": True},
         },
     )
     yield f"http://127.0.0.1:{port}"
@@ -121,7 +121,7 @@ def test_health_extra_fields_merged(self, health_url_with_extras):
         assert data["last_successful_poll"] == "2026-03-16T12:00:00+00:00"
         assert data["probe_success_rate"] == 0.5
         assert data["mq_depth"] == 3
-        assert data["db_pool"] == {"max": 10, "in_use": 1, "available": 9}
+        assert data["db_pool"] == {"reachable": True}
 
 
 @dataclass(frozen=True, slots=True)
diff --git a/tests/test_main_health_merge.py b/tests/test_main_health_merge.py
index ab376bc..4a3d69d 100644
--- a/tests/test_main_health_merge.py
+++ b/tests/test_main_health_merge.py
@@ -21,11 +21,11 @@ def test_merge_scheduler_wins_on_key_conflict(caplog):
         "poll_count": 99,
     }
     with caplog.at_level(logging.DEBUG, logger="paperscout"):
-        out = _merge_extra_health_fields(scheduler, mq_extra, {"max": 10})
+        out = _merge_extra_health_fields(scheduler, mq_extra, {"reachable": True})
     assert out["last_updated"] == "2026-01-01T00:00:00+00:00"
     assert out["poll_count"] == 1
     assert out["mq_depth"] == 5
-    assert out["db_pool"] == {"max": 10}
+    assert out["db_pool"] == {"reachable": True}
     assert any("not allow-listed" in r.message for r in caplog.records)
 
 
diff --git a/tests/test_shutdown.py b/tests/test_shutdown.py
index 1acb092..838184a 100644
--- a/tests/test_shutdown.py
+++ b/tests/test_shutdown.py
@@ -79,3 +79,53 @@ def test_shutdown_services_continues_after_mq_drain_failure(self, caplog):
         assert drained == 0
         health_server.shutdown.assert_called_once()
         assert any("drained 0" in r.message for r in caplog.records)
+
+    def test_shutdown_services_closes_pool(self, caplog):
+        pool = MagicMock()
+        with caplog.at_level(logging.INFO, logger="paperscout"):
+            shutdown_services(
+                reason="SIGTERM",
+                mq=None,
+                health_server=None,
+                health_thread=None,
+                app=None,
+                bolt_thread=None,
+                mq_drain_timeout=30.0,
+                thread_join_timeout=5.0,
+                pool=pool,
+            )
+        pool.closeall.assert_called_once()
+        assert any("DB pool closed" in r.message for r in caplog.records)
+
+    def test_shutdown_services_pool_close_raises_continues(self, caplog):
+        mq = MagicMock()
+        mq.drain.return_value = 1
+        pool = MagicMock()
+        pool.closeall.side_effect = RuntimeError("pool close boom")
+        with caplog.at_level(logging.ERROR, logger="paperscout"):
+            drained = shutdown_services(
+                reason="SIGTERM",
+                mq=mq,
+                health_server=None,
+                health_thread=None,
+                app=None,
+                bolt_thread=None,
+                mq_drain_timeout=30.0,
+                thread_join_timeout=5.0,
+                pool=pool,
+            )
+        assert drained == 1
+        pool.closeall.assert_called_once()
+
+    def test_shutdown_services_skips_pool_when_none(self):
+        shutdown_services(
+            reason="unknown",
+            mq=None,
+            health_server=None,
+            health_thread=None,
+            app=None,
+            bolt_thread=None,
+            mq_drain_timeout=30.0,
+            thread_join_timeout=5.0,
+            pool=None,
+        )