diff --git a/work_trace/wizards/wt_import_timeline_wizard.py b/work_trace/wizards/wt_import_timeline_wizard.py
index f9c2a722..689d77c9 100644
--- a/work_trace/wizards/wt_import_timeline_wizard.py
+++ b/work_trace/wizards/wt_import_timeline_wizard.py
@@ -114,27 +114,33 @@ def _max_speed_from_path(timeline_path):
     return max_speed
 
 
-def _speed_from_activity_segment(activity, seg_start_ts, seg_end_ts):
-    """Calculate implied speed from an activity segment's start/end coords and timestamps.
+MIN_FLIGHT_DISTANCE_MILES = 50  # Shorter than this cannot be a flight
 
-    Flights appear as UNKNOWN_ACTIVITY_TYPE with valid start/end latLng but no
-    timelinePath GPS trail, so we derive speed directly from the segment geometry.
+def _speed_from_activity_segment(activity, seg_start_ts, seg_end_ts):
+    """Detect flights from an activity segment's start/end coords and timestamps.
+
+    Flights show up as UNKNOWN_ACTIVITY_TYPE with widely-separated start/end latLng.
+    We require at least 50 miles distance to avoid mis-classifying GPS drift as flying
+    (short jumps can imply absurdly high speeds due to imprecise timestamps).
+
+    Returns (distance_miles, speed_mph) or (0, 0) if not usable.
     """
     start_latlng = activity.get('start', {}).get('latLng', '')
     end_latlng = activity.get('end', {}).get('latLng', '')
     if not start_latlng or not end_latlng:
-        return 0.0
+        return 0.0, 0.0
     slat, slng = _parse_latlng(start_latlng)
     elat, elng = _parse_latlng(end_latlng)
     if slat is None or elat is None:
-        return 0.0
+        return 0.0, 0.0
     if not seg_start_ts or not seg_end_ts:
-        return 0.0
+        return 0.0, 0.0
     dt_hours = (seg_end_ts - seg_start_ts).total_seconds() / 3600.0
     if dt_hours <= 0:
-        return 0.0
+        return 0.0, 0.0
     dist_miles = _haversine_miles(slat, slng, elat, elng)
-    return dist_miles / dt_hours
+    speed = dist_miles / dt_hours
+    return dist_miles, speed
 
 
 class WtImportTimelineWizard(models.TransientModel):
@@ -300,14 +306,13 @@ class WtImportTimelineWizard(models.TransientModel):
                     # Explicit type wins (driving, walking, transit, cycling)
                     pending_mode = mode
                 else:
-                    # UNKNOWN type — infer from segment distance / elapsed time
+                    # UNKNOWN type — only infer flying if distance is large enough
+                    # to rule out GPS drift (short jumps can look fast due to imprecise timestamps)
                     seg_start = _parse_ts(seg.get('startTime'))
                     seg_end = _parse_ts(seg.get('endTime'))
-                    speed = _speed_from_activity_segment(activity, seg_start, seg_end)
-                    if speed > 0:
-                        inferred = _speed_to_travel_mode(speed)
-                        if inferred != 'unknown':
-                            pending_mode = inferred
+                    dist, speed = _speed_from_activity_segment(activity, seg_start, seg_end)
+                    if dist >= MIN_FLIGHT_DISTANCE_MILES and speed > 150:
+                        pending_mode = 'flying'
                 continue
 
             # --- visit segment: the stop we record ---