diff --git a/bincio/extract/metrics.py b/bincio/extract/metrics.py
index 67ef765..1823bd0 100644
--- a/bincio/extract/metrics.py
+++ b/bincio/extract/metrics.py
@@ -181,16 +181,32 @@ def compute_best_efforts(
     # Build dense 1 Hz speed (km/h) and elevation (m) arrays with gap zero-filling.
     # Zero-filling speed gaps (0 km/h) prevents best-effort windows from spanning
     # recording pauses and producing artificially fast times.
+    # When the device didn't record speed (common in older FIT files), fall back to
+    # GPS-derived speed: spread the haversine segment speed evenly across the interval
+    # so the sliding window accumulates the correct distance.
     sparse_speed: dict[int, float] = {}
     sparse_ele: dict[int, Optional[float]] = {}
     last_t = -1
+    _prev: Optional[DataPoint] = None
     for p in pts:
         t = int((p.timestamp - started_at).total_seconds())
         if t < 0 or t == last_t:
             continue
-        last_t = t
-        sparse_speed[t] = p.speed_kmh if p.speed_kmh is not None else 0.0
         sparse_ele[t] = p.elevation_m
+        if p.speed_kmh is not None:
+            sparse_speed[t] = p.speed_kmh
+        elif (_prev is not None
+              and _prev.lat is not None and _prev.lon is not None
+              and p.lat is not None and p.lon is not None):
+            dt_s = t - last_t
+            seg_m = _haversine_m(_prev.lat, _prev.lon, p.lat, p.lon)
+            seg_kmh = (seg_m / dt_s) * 3.6
+            for slot in range(last_t, t):
+                sparse_speed[slot] = seg_kmh
+        else:
+            sparse_speed[t] = 0.0
+        last_t = t
+        _prev = p
 
     if not sparse_speed:
         return None, None