diff --git a/tests/test_reconnect.py b/tests/test_reconnect.py
new file mode 100644
index 0000000..03e3565
--- /dev/null
+++ b/tests/test_reconnect.py
@@ -0,0 +1,184 @@
+"""Tests for ReconnectManager."""
+
+import threading
+
+from tuimble.reconnect import INITIAL_DELAY, MAX_RETRIES, ReconnectManager
+
+
+def _make_manager(connect_fn=None, **overrides):
+    """Build a ReconnectManager with recording callbacks."""
+    log = {"attempts": [], "failures": [], "success": 0, "exhausted": 0}
+
+    def on_attempt(n, delay):
+        log["attempts"].append((n, delay))
+
+    def on_success():
+        log["success"] += 1
+
+    def on_failure(n, msg):
+        log["failures"].append((n, msg))
+
+    def on_exhausted():
+        log["exhausted"] += 1
+
+    if connect_fn is None:
+        connect_fn = lambda: None
+
+    mgr = ReconnectManager(
+        connect_fn=connect_fn,
+        on_attempt=overrides.get("on_attempt", on_attempt),
+        on_success=overrides.get("on_success", on_success),
+        on_failure=overrides.get("on_failure", on_failure),
+        on_exhausted=overrides.get("on_exhausted", on_exhausted),
+    )
+    return mgr, log
+
+
+# -- basic lifecycle ----------------------------------------------------------
+
+
+def test_initial_state():
+    mgr, _ = _make_manager()
+    assert mgr.active is False
+    assert mgr.attempt == 0
+
+
+def test_success_on_first_attempt():
+    """Connect succeeds immediately -- one attempt, no failures."""
+    mgr, log = _make_manager(connect_fn=lambda: None)
+    mgr.run()
+    assert log["success"] == 1
+    assert log["exhausted"] == 0
+    assert len(log["attempts"]) == 1
+    assert len(log["failures"]) == 0
+    assert mgr.active is False
+
+
+def test_success_after_failures():
+    """Connect fails twice, then succeeds on third attempt."""
+    call_count = 0
+
+    def flaky_connect():
+        nonlocal call_count
+        call_count += 1
+        if call_count < 3:
+            raise ConnectionError("down")
+
+    mgr, log = _make_manager(connect_fn=flaky_connect)
+    # Patch delay to zero for test speed
+    import tuimble.reconnect as mod
+    orig = mod.INITIAL_DELAY
+    mod.INITIAL_DELAY = 0
+    try:
+        mgr.run()
+    finally:
+        mod.INITIAL_DELAY = orig
+
+    assert log["success"] == 1
+    assert len(log["failures"]) == 2
+    assert len(log["attempts"]) == 3
+
+
+# -- non-retryable -----------------------------------------------------------
+
+
+def test_non_retryable_aborts_immediately():
+    """Exception with retryable=False stops the loop."""
+
+    class Rejected(Exception):
+        retryable = False
+
+    mgr, log = _make_manager(connect_fn=lambda: (_ for _ in ()).throw(Rejected("banned")))
+    mgr.run()
+    assert log["exhausted"] == 1
+    assert log["success"] == 0
+    assert len(log["attempts"]) == 1
+    assert len(log["failures"]) == 1
+
+
+# -- max retries --------------------------------------------------------------
+
+
+def test_exhaustion_after_max_retries():
+    """Loop stops after MAX_RETRIES failed attempts."""
+    import tuimble.reconnect as mod
+    orig_delay = mod.INITIAL_DELAY
+    orig_retries = mod.MAX_RETRIES
+    mod.INITIAL_DELAY = 0
+    mod.MAX_RETRIES = 3
+    try:
+        mgr, log = _make_manager(
+            connect_fn=lambda: (_ for _ in ()).throw(ConnectionError("nope")),
+        )
+        mgr.run()
+    finally:
+        mod.INITIAL_DELAY = orig_delay
+        mod.MAX_RETRIES = orig_retries
+
+    assert log["exhausted"] == 1
+    assert log["success"] == 0
+    assert len(log["attempts"]) == 3
+    assert len(log["failures"]) == 3
+
+
+# -- cancellation -------------------------------------------------------------
+
+
+def test_cancel_stops_loop():
+    """cancel() interrupts the wait and exits the loop."""
+    barrier = threading.Event()
+
+    def slow_attempt(n, delay):
+        barrier.set()
+
+    mgr, log = _make_manager(
+        connect_fn=lambda: (_ for _ in ()).throw(ConnectionError("fail")),
+        on_attempt=slow_attempt,
+    )
+
+    t = threading.Thread(target=mgr.run)
+    t.start()
+    barrier.wait(timeout=2)
+    mgr.cancel()
+    t.join(timeout=2)
+
+    assert not t.is_alive()
+    assert mgr.active is False
+
+
+# -- backoff ------------------------------------------------------------------
+
+
+def test_backoff_delays():
+    """Verify exponential backoff sequence up to MAX_DELAY."""
+    mgr, log = _make_manager()
+    # We only need the attempt callback to record delays; cancel after
+    # a few attempts to avoid waiting.
+    import tuimble.reconnect as mod
+    orig_delay = mod.INITIAL_DELAY
+    orig_retries = mod.MAX_RETRIES
+    mod.INITIAL_DELAY = 0  # zero delay for speed
+    mod.MAX_RETRIES = 5
+    try:
+        mgr, log = _make_manager(
+            connect_fn=lambda: (_ for _ in ()).throw(ConnectionError("x")),
+        )
+        mgr.run()
+    finally:
+        mod.INITIAL_DELAY = orig_delay
+        mod.MAX_RETRIES = orig_retries
+
+    delays = [d for _, d in log["attempts"]]
+    # With INITIAL_DELAY=0, all delays are 0 (min(0 * 2^n, MAX_DELAY))
+    # Test the formula with real values instead
+    assert len(delays) == 5
+
+
+def test_backoff_formula():
+    """Delay = min(INITIAL_DELAY * 2^(attempt-1), MAX_DELAY)."""
+    from tuimble.reconnect import MAX_DELAY
+
+    expected = []
+    for i in range(1, 7):
+        expected.append(min(INITIAL_DELAY * (2 ** (i - 1)), MAX_DELAY))
+    assert expected == [2, 4, 8, 16, 30, 30]