Fix segment avg_speed: derive from distance/time; tighten speed bounds to reject false matches

bincio/segments/detect.py

@@ -25,10 +25,17 @@ CONFORMANCE_MAX_FRAC = 0.30  # max fraction of interior points allowed to deviat
 # Rejects false matches from long circuit rides where the track passes the
 # segment start early and the segment end hours later.
 _MIN_SPEED_MS: dict[str, float] = {
-    'cycling': 1.0,   # ~3.6 km/h — slowest realistic climb
+    'cycling': 2.0,   # ~7.2 km/h — below any realistic cyclist even on brutal climbs
-    'running': 0.5,   # ~1.8 km/h
+    'running': 0.8,   # ~2.9 km/h
 }
-_MIN_SPEED_DEFAULT = 0.2  # hiking / walking / unknown
+_MIN_SPEED_DEFAULT = 0.3  # hiking / walking / unknown
+
+# Maximum geometric speed per sport in m/s — rejects GPS glitch matches.
+_MAX_SPEED_MS: dict[str, float] = {
+    'cycling': 30.0,  # ~108 km/h
+    'running': 12.0,  # ~43 km/h
+}
+_MAX_SPEED_DEFAULT = 20.0
 
 # ── fast distance approximation ───────────────────────────────────────────────
 
@@ -207,7 +214,10 @@ def _extract_effort(
 ) -> SegmentEffort:
     elapsed_s = track.times[j] - track.times[i]
     started_at = (track.started_at + timedelta(seconds=track.times[i])).replace(microsecond=0)
-    avg_speed = _avg_nonnull(track.speeds, i, j)
+    # Always derive avg speed from segment distance / elapsed time.  Device-recorded
+    # speed is unreliable across formats (m/s vs km/h in older FIT files) and
+    # averaging instantaneous GPS speed over a slice gives different results anyway.
+    avg_speed = (seg.distance_m / elapsed_s * 3.6) if elapsed_s > 0 else None
     avg_hr_raw = _avg_nonnull(track.hrs, i, j)
     avg_hr = int(round(avg_hr_raw)) if avg_hr_raw is not None else None
     avg_pwr_raw = _avg_nonnull(track.powers, i, j)
@@ -259,12 +269,13 @@ def detect_one(track: ActivityTrack, seg: Segment) -> list[SegmentEffort]:
             # No end found — no more efforts possible starting at or after start_idx.
             break
 
-        # Reject implausibly slow matches (e.g. circuit rides matching start
+        # Reject implausibly slow or fast matches.
-        # early and end hours later on a second pass through the area).
         elapsed = track.times[end_idx] - track.times[start_idx]
         if elapsed > 0:
+            geo_speed = seg.distance_m / elapsed
             min_speed = _MIN_SPEED_MS.get(track.sport, _MIN_SPEED_DEFAULT)
-            if seg.distance_m / elapsed < min_speed:
+            max_speed = _MAX_SPEED_MS.get(track.sport, _MAX_SPEED_DEFAULT)
+            if geo_speed < min_speed or geo_speed > max_speed:
                 search_from = start_idx + 1
                 continue