fix(transform): prefer latest overlapping subwatcher event

Author: TimeToBuildBob

aw_transform/merge_subwatcher_fields.py

@@ -61,7 +61,9 @@ def merge_subwatcher_fields(
     Note on overlap:
         Base events are split at the clipped subwatcher boundaries so each
         output segment only carries the subwatcher fields that were actually
-        present during that slice of time.
+        present during that slice of time. If multiple subwatcher events cover
+        the same slice, the most recent one wins so a later transition does not
+        get smeared backward by an older, longer pulse.
     """
     if conflict not in ("base_wins", "sub_wins"):
         raise ValueError(
@@ -102,17 +104,25 @@ def merge_subwatcher_fields(
         for start, end in zip(boundary_points, boundary_points[1:]):
             segment_period = Timeslot(start, end)
             best_sub: Optional[Event] = None
-            best_overlap_secs: float = 0.0
+            best_sub_period: Optional[Timeslot] = None
 
             for sub, sub_period in overlapping:
-                ip = segment_period.intersection(sub_period)
-                if not ip:
+                if not segment_period.intersection(sub_period):
                     continue
 
-                overlap_secs = ip.duration.total_seconds()
-                if overlap_secs > best_overlap_secs:
-                    best_overlap_secs = overlap_secs
+                # Later subwatcher events should supersede older overlapping
+                # pulses on the shared slice instead of letting stale data linger.
+                if (
+                    best_sub is None
+                    or sub.timestamp > best_sub.timestamp
+                    or (
+                        sub.timestamp == best_sub.timestamp
+                        and best_sub_period is not None
+                        and sub_period.end > best_sub_period.end
+                    )
+                ):
                     best_sub = sub
+                    best_sub_period = sub_period
 
             enriched = deepcopy(base)
             enriched.timestamp = start

tests/test_transforms.py

@@ -537,6 +537,35 @@ def test_merge_subwatcher_fields_partial_overlap_splits_base():
     assert sum_durations(result) == td1h
 
 
+def test_merge_subwatcher_fields_overlapping_subwatchers_prefer_latest_start():
+    """Newer overlapping subwatcher events take over from their own start time."""
+    now = datetime(2024, 1, 1, 12, 0, tzinfo=timezone.utc)
+    td20m = timedelta(minutes=20)
+    td40m = timedelta(minutes=40)
+    td1h = timedelta(hours=1)
+
+    base = [Event(timestamp=now, duration=td1h, data={"app": "vim"})]
+    sub = [
+        Event(timestamp=now, duration=td40m, data={"project": "alpha"}),
+        Event(timestamp=now + td20m, duration=td40m, data={"project": "beta"}),
+    ]
+
+    result = merge_subwatcher_fields(base, sub, ["project"])
+
+    assert len(result) == 2
+    assert [event.timestamp for event in result] == [now, now + td20m]
+    assert [event.duration for event in result] == [td20m, td40m]
+    assert [event.data.get("project") for event in result] == ["alpha", "beta"]
+
+    by_project = {
+        event.data.get("project"): event.duration
+        for event in merge_events_by_keys(result, ["project"])
+    }
+    assert by_project["alpha"] == td20m
+    assert by_project["beta"] == td40m
+    assert sum_durations(result) == td1h
+
+
 def test_merge_subwatcher_fields_no_overlap():
     """Base events with no overlapping subwatcher event are returned unchanged."""
     now = datetime(2024, 1, 1, 12, 0, tzinfo=timezone.utc)