diff --git a/Dockerfile b/Dockerfile
index 4fb159f..f038006 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -13,4 +13,13 @@ COPY . .
 
 EXPOSE 8004
 
+# PR-B0: no compose-level healthcheck existed for this service prior to
+# this change (unlike mysql/redis, which gate downstream depends_on
+# conditions) -- api-gateway's depends_on only waited on
+# condition: service_started. This HEALTHCHECK makes container health
+# introspectable via `docker ps`/`docker inspect`; it does not by itself
+# change any depends_on condition in docker-compose.yml.
+HEALTHCHECK --interval=10s --timeout=3s --start-period=5s --retries=3 \
+  CMD curl -f http://localhost:8004/health || exit 1
+
 CMD ["uvicorn", "api.main:app", "--host", "0.0.0.0", "--port", "8004"]
diff --git a/tests/test_worker_integration_real_backends.py b/tests/test_worker_integration_real_backends.py
new file mode 100644
index 0000000..59e3829
--- /dev/null
+++ b/tests/test_worker_integration_real_backends.py
@@ -0,0 +1,236 @@
+"""PR-B0: end-to-end regression against REAL Redis and REAL MySQL, not the
+mocks/SQLite used by the rest of the suite (tests/test_worker.py,
+tests/test_sink.py, tests/test_stream.py, tests/test_migrations.py).
+
+This exists specifically to prove the runtime path B0 makes deployable --
+Redis Streams consumer group -> worker -> MySQL persistence -- actually
+works against real backends, since nothing in the repo previously did.
+It is intentionally isolated from any other data on those backends:
+
+* Uses its own Redis stream name (never the production `audit:events`
+  stream other services are actively writing to), destroyed at teardown.
+* Uses its own throwaway MySQL database (never the `omnibioai_audit`
+  database the real worker/API use), dropped at teardown.
+
+Skips automatically (does not fail) when a real Redis/MySQL isn't
+reachable -- e.g. in CI, which has no service containers configured for
+this repo today (see .github/workflows/ci.yml). This is opportunistic
+regression coverage for environments (like local dev-stack validation)
+where real backends exist, not a hard CI requirement introduced by this
+PR -- see the B0 report for why that's a deliberate, separately-flagged
+follow-up rather than bundled into this change.
+"""
+import os
+import uuid
+
+import pytest
+import redis as redis_lib
+from sqlalchemy import create_engine, text
+
+TEST_REDIS_URL = os.getenv("B0_TEST_REDIS_URL", "redis://localhost:6380")
+TEST_MYSQL_ROOT_URL = os.getenv(
+    "B0_TEST_MYSQL_ROOT_URL", "mysql+pymysql://root:root@localhost:3306/mysql"
+)
+TEST_DB_NAME = "omnibioai_audit_b0_test"
+TEST_STREAM = f"audit:events:b0-test-{uuid.uuid4().hex[:8]}"
+
+
+def _real_backends_available():
+    try:
+        r = redis_lib.from_url(TEST_REDIS_URL, socket_connect_timeout=2)
+        r.ping()
+    except Exception:
+        return False
+    try:
+        engine = create_engine(TEST_MYSQL_ROOT_URL, connect_args={"connect_timeout": 2})
+        with engine.connect():
+            pass
+    except Exception:
+        return False
+    return True
+
+
+pytestmark = pytest.mark.skipif(
+    not _real_backends_available(),
+    reason="real Redis/MySQL not reachable (set B0_TEST_REDIS_URL / "
+    "B0_TEST_MYSQL_ROOT_URL, or run against the dev docker-compose stack) "
+    "-- skipped, not failed, since CI has no service containers for this "
+    "repo today",
+)
+
+
+@pytest.fixture
+def real_mysql_url():
+    """Creates a throwaway database, runs the real alembic migration
+    against it (proving the actual migration -- not a hand-built schema --
+    is what's being validated), yields its URL, then drops it."""
+    root_engine = create_engine(TEST_MYSQL_ROOT_URL)
+    with root_engine.connect() as conn:
+        conn.execute(text(f"DROP DATABASE IF EXISTS {TEST_DB_NAME}"))
+        conn.execute(text(f"CREATE DATABASE {TEST_DB_NAME}"))
+        conn.commit()
+
+    db_url = TEST_MYSQL_ROOT_URL.rsplit("/", 1)[0] + f"/{TEST_DB_NAME}"
+
+    from pathlib import Path
+
+    from alembic.config import Config
+
+    from alembic import command
+
+    repo_root = Path(__file__).resolve().parent.parent
+    cfg = Config(str(repo_root / "alembic.ini"))
+    cfg.set_main_option("script_location", str(repo_root / "alembic"))
+    cfg.set_main_option("sqlalchemy.url", db_url)
+    command.upgrade(cfg, "head")
+
+    yield db_url
+
+    with root_engine.connect() as conn:
+        conn.execute(text(f"DROP DATABASE IF EXISTS {TEST_DB_NAME}"))
+        conn.commit()
+
+
+@pytest.fixture
+def real_redis_stream():
+    """Yields a real StreamReader pointed at an isolated test stream name,
+    then destroys the consumer group and deletes the stream -- never
+    touches the production `audit:events` stream."""
+    from audit.config import AuditConfig
+    from consumers.stream_reader import StreamReader
+
+    original_stream = AuditConfig.STREAM_NAME
+    AuditConfig.STREAM_NAME = TEST_STREAM
+    try:
+        reader = StreamReader()
+        reader.ensure_group()
+        yield reader
+    finally:
+        # Best-effort teardown of the isolated test stream/group -- if the
+        # test itself already failed, the group/stream may be partially
+        # created or already gone; swallow here so teardown never masks
+        # the real assertion failure with a cleanup error.
+        try:
+            reader.redis.xgroup_destroy(TEST_STREAM, AuditConfig.CONSUMER_GROUP)
+        except Exception as cleanup_err:
+            print(f"[TEST TEARDOWN] xgroup_destroy failed (non-fatal): {cleanup_err}")
+        try:
+            reader.redis.delete(TEST_STREAM)
+        except Exception as cleanup_err:
+            print(f"[TEST TEARDOWN] stream delete failed (non-fatal): {cleanup_err}")
+        AuditConfig.STREAM_NAME = original_stream
+
+
+def test_real_stream_group_bootstrap_is_idempotent(real_redis_stream):
+    """ensure_group() must be safely callable more than once against a
+    real Redis instance (BUSYGROUP swallowed) -- the mocked version in
+    tests/test_stream.py can't prove the real ResponseError string match
+    holds against a real server's actual error text."""
+    real_redis_stream.ensure_group()  # second call, must not raise
+
+
+def test_real_produce_consume_persist_ack_round_trip(real_redis_stream, real_mysql_url, monkeypatch):
+    """The exact chain B0 makes deployable: producer XADD -> real Redis
+    Streams consumer group -> worker.handle_message -> real MySQL insert
+    -> XACK. Uses the worker's actual handle_message(), not a
+    reimplementation of it."""
+    import json
+    from datetime import datetime, timezone
+
+    from sqlalchemy import create_engine
+    from sqlalchemy.orm import sessionmaker
+
+    import worker.main as worker_module
+
+    engine = create_engine(real_mysql_url)
+    TestSessionLocal = sessionmaker(bind=engine)
+    monkeypatch.setattr(worker_module, "SessionLocal", TestSessionLocal)
+
+    event_id = f"b0-test-{uuid.uuid4()}"
+    payload = {
+        "event_id": event_id,
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "service": "b0-integration-test",
+        "event_type": "test",
+        "user_id": "test-user",
+        "action": "b0_smoke",
+        "resource": None,
+        "decision": "success",
+        "reason": None,
+        "trace_id": "b0-trace-1",
+        "context": {},
+    }
+    real_redis_stream.redis.xadd(TEST_STREAM, {"data": json.dumps(payload)})
+
+    response = real_redis_stream.read_group(consumer_name="b0-test-consumer", block=3000)
+    assert response, "expected the real event just XADDed to be delivered"
+
+    acked = False
+    for _stream_name, messages in response:
+        for message_id, fields in messages:
+            result = worker_module.handle_message(real_redis_stream, message_id, fields)
+            assert result is True
+            acked = True
+    assert acked
+
+    with TestSessionLocal() as session:
+        from db.models import AuditEventRecord
+
+        row = session.get(AuditEventRecord, event_id)
+        assert row is not None
+        assert row.service == "b0-integration-test"
+        assert row.action == "b0_smoke"
+
+    # Pending entries list must be empty -- the message was acked, not
+    # left for retry.
+    pending = real_redis_stream.redis.xpending(TEST_STREAM, "audit-workers")
+    assert pending["pending"] == 0
+
+
+def test_real_duplicate_delivery_does_not_duplicate_row(real_mysql_url):
+    """Sink.write()'s IntegrityError-swallow behavior (consumers/sink.py)
+    against a real MySQL unique-constraint violation, not SQLite's -- the
+    exact exception type/rollback path differs enough between backends
+    that this is worth proving directly, not just inferring from the
+    SQLite-backed tests/test_sink.py coverage."""
+    from datetime import datetime, timezone
+
+    from sqlalchemy import create_engine
+    from sqlalchemy.orm import sessionmaker
+
+    from consumers.sink import Sink
+
+    engine = create_engine(real_mysql_url)
+    TestSessionLocal = sessionmaker(bind=engine)
+
+    event = {
+        "event_id": f"b0-dup-test-{uuid.uuid4()}",
+        "timestamp": datetime.now(timezone.utc),
+        "service": "b0-integration-test",
+        "event_type": "test",
+        "user_id": None,
+        "action": "dup_check",
+        "resource": None,
+        "decision": "success",
+        "reason": None,
+        "trace_id": None,
+        "context": {},
+    }
+
+    with TestSessionLocal() as session1:
+        first_result = Sink(session1).write(event)
+    with TestSessionLocal() as session2:
+        second_result = Sink(session2).write(event)  # redelivery, same event_id
+
+    assert first_result is True
+    assert second_result is True  # duplicate is a no-op success, not a crash
+
+    with TestSessionLocal() as session:
+        from sqlalchemy import func
+
+        from db.models import AuditEventRecord
+
+        count = session.query(func.count(AuditEventRecord.event_id)).filter(
+            AuditEventRecord.event_id == event["event_id"]
+        ).scalar()
+        assert count == 1  # exactly one row, not two