diff --git a/fintick/storage.py b/fintick/storage.py
index 5ef6136..0997876 100644
--- a/fintick/storage.py
+++ b/fintick/storage.py
@@ -264,14 +264,38 @@ def _bootstrap_post_aggregation_decisions(connection: sqlite3.Connection) -> Non
     )
 
 
+# How long a connection waits for a lock before raising. Python's sqlite3 default is 5s,
+# which is not enough when several workers open the database at the same instant and each
+# runs the migration block in a write transaction. WAL lets readers and one writer coexist;
+# this covers the writer-vs-writer moment at startup.
+BUSY_TIMEOUT_SECONDS = 30.0
+
+
 @contextmanager
 def open_database(path: str | Path) -> Iterator[sqlite3.Connection]:
     """Open and initialize a FinTick database, committing on success."""
     database = Path(path)
     database.parent.mkdir(parents=True, exist_ok=True)
-    connection = sqlite3.connect(database)
+    connection = sqlite3.connect(database, timeout=BUSY_TIMEOUT_SECONDS)
     try:
-        connection.execute("PRAGMA journal_mode=WAL")
+        # The timeout on connect() above is what actually makes a blocked open wait; the
+        # PRAGMA restates it for any connection opened elsewhere and makes the value visible
+        # to `PRAGMA busy_timeout`. Set it before the WAL switch, which can itself need a lock.
+        #
+        # Why this is needed: every open runs the schema/migration block inside a write
+        # transaction, so four workers starting in the same second all contend for one write
+        # lock before doing any real work. That surfaced as "OperationalError: database is
+        # locked" one second after every boot and deploy, costing a full 900s cycle. The
+        # contention lasts milliseconds — waiting it out is the correct response.
+        connection.execute(f"PRAGMA busy_timeout={BUSY_TIMEOUT_SECONDS * 1000:.0f}")
+        # journal_mode is a PERSISTENT property of the database file, so re-setting it on
+        # every open is pointless — and not free. Switching journal mode needs an exclusive
+        # lock, and SQLite answers SQLITE_BUSY for it IMMEDIATELY rather than honouring the
+        # busy handler, so no timeout can save it. Four workers opening in the same second
+        # therefore raced on a switch that was already a no-op. Read first, write only if
+        # it actually needs changing.
+        if connection.execute("PRAGMA journal_mode").fetchone()[0].lower() != "wal":
+            connection.execute("PRAGMA journal_mode=WAL")
         connection.execute("PRAGMA foreign_keys=ON")
         connection.execute("BEGIN")
         # Legacy databases need columns before indexes can refer to them.
diff --git a/tests/test_storage_locking.py b/tests/test_storage_locking.py
new file mode 100644
index 0000000..b369d77
--- /dev/null
+++ b/tests/test_storage_locking.py
@@ -0,0 +1,100 @@
+"""Concurrent opens must wait for the lock, not raise."""
+
+from __future__ import annotations
+
+import sqlite3
+import tempfile
+import threading
+import time
+import unittest
+from unittest import mock
+from pathlib import Path
+
+from fintick.storage import BUSY_TIMEOUT_SECONDS, open_database
+
+
+class BusyTimeoutTests(unittest.TestCase):
+    def setUp(self) -> None:
+        self.tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(self.tmp.cleanup)
+        self.database = Path(self.tmp.name) / "fintick.db"
+
+    def test_busy_timeout_is_raised_above_the_python_default(self) -> None:
+        with open_database(self.database) as connection:
+            timeout = connection.execute("PRAGMA busy_timeout").fetchone()[0]
+        # Python's default is 5000ms, which is not enough for several workers opening at once.
+        self.assertGreaterEqual(timeout, 30000)
+        self.assertEqual(BUSY_TIMEOUT_SECONDS, 30.0)
+
+    def test_wal_is_still_enabled(self) -> None:
+        with open_database(self.database) as connection:
+            self.assertEqual(
+                connection.execute("PRAGMA journal_mode").fetchone()[0].lower(), "wal"
+            )
+
+    def test_the_wal_switch_is_skipped_when_already_enabled(self) -> None:
+        """The actual cause of the startup failure.
+
+        Switching journal_mode needs an exclusive lock, and SQLite answers SQLITE_BUSY for it
+        IMMEDIATELY rather than honouring the busy handler — so no timeout can rescue it. The
+        switch was being re-issued on every open despite journal_mode being persistent, which
+        is what four simultaneous workers actually collided on.
+        """
+        with open_database(self.database):
+            pass  # first open sets WAL
+
+        # sqlite3.Connection is an immutable C type, so the statements are captured with the
+        # connection's own trace callback, attached as it is created.
+        statements: list[str] = []
+        real_connect = sqlite3.connect
+
+        def traced(*args, **kwargs):  # type: ignore[no-untyped-def]
+            connection = real_connect(*args, **kwargs)
+            connection.set_trace_callback(statements.append)
+            return connection
+
+        with mock.patch("fintick.storage.sqlite3.connect", traced):
+            with open_database(self.database):
+                pass
+
+        writes = [s for s in statements if "journal_mode=" in s.replace(" ", "")]
+        self.assertEqual(writes, [], f"re-issued the WAL switch unnecessarily: {writes}")
+
+    def test_a_concurrent_open_waits_instead_of_raising(self) -> None:
+        """Behavioural check: a held write lock delays an open rather than failing it.
+
+        Note this does not by itself prove the fix — Python's 5s connect() default would
+        also pass a 0.4s hold. The guard for the raised timeout is the pragma test above;
+        the guard for the real cause is the WAL test above that.
+        """
+        with open_database(self.database):
+            pass  # create the schema first
+
+        holder = sqlite3.connect(self.database)
+        holder.execute("PRAGMA busy_timeout=30000")
+        holder.execute("BEGIN IMMEDIATE")  # take the write lock
+        self.addCleanup(holder.close)
+
+        failure: list[BaseException] = []
+
+        def opener() -> None:
+            try:
+                with open_database(self.database):
+                    pass
+            except BaseException as error:  # noqa: BLE001 - recorded for the assertion
+                failure.append(error)
+
+        thread = threading.Thread(target=opener)
+        thread.start()
+        # Hold the lock briefly, then release it from THIS thread — a sqlite3 connection
+        # may only be used by the thread that created it.
+        time.sleep(0.4)
+        holder.rollback()
+        thread.join(timeout=20)
+
+        self.assertFalse(thread.is_alive(), "the concurrent open never completed")
+        self.assertEqual(failure, [], f"concurrent open raised instead of waiting: {failure}")
+
+
+if __name__ == "__main__":
+    unittest.main()